Spaces:
Runtime error
Runtime error
karolmajek
commited on
Commit
•
1a1ee1f
1
Parent(s):
799a750
app
Browse filesThis view is limited to 50 files because it contains too many changes.
See raw diff
- LICENSE +674 -0
- README.md +224 -1
- app.py +104 -0
- cfg/yolor_csp.cfg +1376 -0
- cfg/yolor_csp_x.cfg +1576 -0
- cfg/yolor_p6.cfg +1760 -0
- cfg/yolor_w6.cfg +1760 -0
- cfg/yolov4_csp.cfg +1334 -0
- cfg/yolov4_csp_x.cfg +1534 -0
- cfg/yolov4_p6.cfg +2260 -0
- cfg/yolov4_p7.cfg +2714 -0
- darknet/README.md +63 -0
- darknet/cfg/yolov4-csp-x.cfg +1555 -0
- darknet/cfg/yolov4-csp.cfg +1354 -0
- darknet/new_layers.md +329 -0
- data/coco.names +80 -0
- data/coco.yaml +18 -0
- data/hyp.finetune.1280.yaml +28 -0
- data/hyp.scratch.1280.yaml +28 -0
- data/hyp.scratch.640.yaml +28 -0
- figure/implicit_modeling.png +0 -0
- figure/performance.png +0 -0
- figure/schedule.png +0 -0
- figure/unifued_network.png +0 -0
- inference/images/horses.jpg +0 -0
- inference/output/horses.jpg +0 -0
- models/__init__.py +1 -0
- models/__pycache__/__init__.cpython-37.pyc +0 -0
- models/__pycache__/models.cpython-37.pyc +0 -0
- models/export.py +68 -0
- models/models.py +761 -0
- requirements.txt +33 -0
- scripts/get_coco.sh +27 -0
- scripts/get_pretrain.sh +7 -0
- test.py +344 -0
- train.py +619 -0
- tune.py +619 -0
- utils/__init__.py +1 -0
- utils/__pycache__/__init__.cpython-37.pyc +0 -0
- utils/__pycache__/__init__.cpython-38.pyc +0 -0
- utils/__pycache__/datasets.cpython-37.pyc +0 -0
- utils/__pycache__/datasets.cpython-38.pyc +0 -0
- utils/__pycache__/general.cpython-37.pyc +0 -0
- utils/__pycache__/google_utils.cpython-37.pyc +0 -0
- utils/__pycache__/google_utils.cpython-38.pyc +0 -0
- utils/__pycache__/layers.cpython-37.pyc +0 -0
- utils/__pycache__/metrics.cpython-37.pyc +0 -0
- utils/__pycache__/parse_config.cpython-37.pyc +0 -0
- utils/__pycache__/plots.cpython-37.pyc +0 -0
- utils/__pycache__/torch_utils.cpython-37.pyc +0 -0
LICENSE
ADDED
@@ -0,0 +1,674 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
GNU GENERAL PUBLIC LICENSE
|
2 |
+
Version 3, 29 June 2007
|
3 |
+
|
4 |
+
Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
|
5 |
+
Everyone is permitted to copy and distribute verbatim copies
|
6 |
+
of this license document, but changing it is not allowed.
|
7 |
+
|
8 |
+
Preamble
|
9 |
+
|
10 |
+
The GNU General Public License is a free, copyleft license for
|
11 |
+
software and other kinds of works.
|
12 |
+
|
13 |
+
The licenses for most software and other practical works are designed
|
14 |
+
to take away your freedom to share and change the works. By contrast,
|
15 |
+
the GNU General Public License is intended to guarantee your freedom to
|
16 |
+
share and change all versions of a program--to make sure it remains free
|
17 |
+
software for all its users. We, the Free Software Foundation, use the
|
18 |
+
GNU General Public License for most of our software; it applies also to
|
19 |
+
any other work released this way by its authors. You can apply it to
|
20 |
+
your programs, too.
|
21 |
+
|
22 |
+
When we speak of free software, we are referring to freedom, not
|
23 |
+
price. Our General Public Licenses are designed to make sure that you
|
24 |
+
have the freedom to distribute copies of free software (and charge for
|
25 |
+
them if you wish), that you receive source code or can get it if you
|
26 |
+
want it, that you can change the software or use pieces of it in new
|
27 |
+
free programs, and that you know you can do these things.
|
28 |
+
|
29 |
+
To protect your rights, we need to prevent others from denying you
|
30 |
+
these rights or asking you to surrender the rights. Therefore, you have
|
31 |
+
certain responsibilities if you distribute copies of the software, or if
|
32 |
+
you modify it: responsibilities to respect the freedom of others.
|
33 |
+
|
34 |
+
For example, if you distribute copies of such a program, whether
|
35 |
+
gratis or for a fee, you must pass on to the recipients the same
|
36 |
+
freedoms that you received. You must make sure that they, too, receive
|
37 |
+
or can get the source code. And you must show them these terms so they
|
38 |
+
know their rights.
|
39 |
+
|
40 |
+
Developers that use the GNU GPL protect your rights with two steps:
|
41 |
+
(1) assert copyright on the software, and (2) offer you this License
|
42 |
+
giving you legal permission to copy, distribute and/or modify it.
|
43 |
+
|
44 |
+
For the developers' and authors' protection, the GPL clearly explains
|
45 |
+
that there is no warranty for this free software. For both users' and
|
46 |
+
authors' sake, the GPL requires that modified versions be marked as
|
47 |
+
changed, so that their problems will not be attributed erroneously to
|
48 |
+
authors of previous versions.
|
49 |
+
|
50 |
+
Some devices are designed to deny users access to install or run
|
51 |
+
modified versions of the software inside them, although the manufacturer
|
52 |
+
can do so. This is fundamentally incompatible with the aim of
|
53 |
+
protecting users' freedom to change the software. The systematic
|
54 |
+
pattern of such abuse occurs in the area of products for individuals to
|
55 |
+
use, which is precisely where it is most unacceptable. Therefore, we
|
56 |
+
have designed this version of the GPL to prohibit the practice for those
|
57 |
+
products. If such problems arise substantially in other domains, we
|
58 |
+
stand ready to extend this provision to those domains in future versions
|
59 |
+
of the GPL, as needed to protect the freedom of users.
|
60 |
+
|
61 |
+
Finally, every program is threatened constantly by software patents.
|
62 |
+
States should not allow patents to restrict development and use of
|
63 |
+
software on general-purpose computers, but in those that do, we wish to
|
64 |
+
avoid the special danger that patents applied to a free program could
|
65 |
+
make it effectively proprietary. To prevent this, the GPL assures that
|
66 |
+
patents cannot be used to render the program non-free.
|
67 |
+
|
68 |
+
The precise terms and conditions for copying, distribution and
|
69 |
+
modification follow.
|
70 |
+
|
71 |
+
TERMS AND CONDITIONS
|
72 |
+
|
73 |
+
0. Definitions.
|
74 |
+
|
75 |
+
"This License" refers to version 3 of the GNU General Public License.
|
76 |
+
|
77 |
+
"Copyright" also means copyright-like laws that apply to other kinds of
|
78 |
+
works, such as semiconductor masks.
|
79 |
+
|
80 |
+
"The Program" refers to any copyrightable work licensed under this
|
81 |
+
License. Each licensee is addressed as "you". "Licensees" and
|
82 |
+
"recipients" may be individuals or organizations.
|
83 |
+
|
84 |
+
To "modify" a work means to copy from or adapt all or part of the work
|
85 |
+
in a fashion requiring copyright permission, other than the making of an
|
86 |
+
exact copy. The resulting work is called a "modified version" of the
|
87 |
+
earlier work or a work "based on" the earlier work.
|
88 |
+
|
89 |
+
A "covered work" means either the unmodified Program or a work based
|
90 |
+
on the Program.
|
91 |
+
|
92 |
+
To "propagate" a work means to do anything with it that, without
|
93 |
+
permission, would make you directly or secondarily liable for
|
94 |
+
infringement under applicable copyright law, except executing it on a
|
95 |
+
computer or modifying a private copy. Propagation includes copying,
|
96 |
+
distribution (with or without modification), making available to the
|
97 |
+
public, and in some countries other activities as well.
|
98 |
+
|
99 |
+
To "convey" a work means any kind of propagation that enables other
|
100 |
+
parties to make or receive copies. Mere interaction with a user through
|
101 |
+
a computer network, with no transfer of a copy, is not conveying.
|
102 |
+
|
103 |
+
An interactive user interface displays "Appropriate Legal Notices"
|
104 |
+
to the extent that it includes a convenient and prominently visible
|
105 |
+
feature that (1) displays an appropriate copyright notice, and (2)
|
106 |
+
tells the user that there is no warranty for the work (except to the
|
107 |
+
extent that warranties are provided), that licensees may convey the
|
108 |
+
work under this License, and how to view a copy of this License. If
|
109 |
+
the interface presents a list of user commands or options, such as a
|
110 |
+
menu, a prominent item in the list meets this criterion.
|
111 |
+
|
112 |
+
1. Source Code.
|
113 |
+
|
114 |
+
The "source code" for a work means the preferred form of the work
|
115 |
+
for making modifications to it. "Object code" means any non-source
|
116 |
+
form of a work.
|
117 |
+
|
118 |
+
A "Standard Interface" means an interface that either is an official
|
119 |
+
standard defined by a recognized standards body, or, in the case of
|
120 |
+
interfaces specified for a particular programming language, one that
|
121 |
+
is widely used among developers working in that language.
|
122 |
+
|
123 |
+
The "System Libraries" of an executable work include anything, other
|
124 |
+
than the work as a whole, that (a) is included in the normal form of
|
125 |
+
packaging a Major Component, but which is not part of that Major
|
126 |
+
Component, and (b) serves only to enable use of the work with that
|
127 |
+
Major Component, or to implement a Standard Interface for which an
|
128 |
+
implementation is available to the public in source code form. A
|
129 |
+
"Major Component", in this context, means a major essential component
|
130 |
+
(kernel, window system, and so on) of the specific operating system
|
131 |
+
(if any) on which the executable work runs, or a compiler used to
|
132 |
+
produce the work, or an object code interpreter used to run it.
|
133 |
+
|
134 |
+
The "Corresponding Source" for a work in object code form means all
|
135 |
+
the source code needed to generate, install, and (for an executable
|
136 |
+
work) run the object code and to modify the work, including scripts to
|
137 |
+
control those activities. However, it does not include the work's
|
138 |
+
System Libraries, or general-purpose tools or generally available free
|
139 |
+
programs which are used unmodified in performing those activities but
|
140 |
+
which are not part of the work. For example, Corresponding Source
|
141 |
+
includes interface definition files associated with source files for
|
142 |
+
the work, and the source code for shared libraries and dynamically
|
143 |
+
linked subprograms that the work is specifically designed to require,
|
144 |
+
such as by intimate data communication or control flow between those
|
145 |
+
subprograms and other parts of the work.
|
146 |
+
|
147 |
+
The Corresponding Source need not include anything that users
|
148 |
+
can regenerate automatically from other parts of the Corresponding
|
149 |
+
Source.
|
150 |
+
|
151 |
+
The Corresponding Source for a work in source code form is that
|
152 |
+
same work.
|
153 |
+
|
154 |
+
2. Basic Permissions.
|
155 |
+
|
156 |
+
All rights granted under this License are granted for the term of
|
157 |
+
copyright on the Program, and are irrevocable provided the stated
|
158 |
+
conditions are met. This License explicitly affirms your unlimited
|
159 |
+
permission to run the unmodified Program. The output from running a
|
160 |
+
covered work is covered by this License only if the output, given its
|
161 |
+
content, constitutes a covered work. This License acknowledges your
|
162 |
+
rights of fair use or other equivalent, as provided by copyright law.
|
163 |
+
|
164 |
+
You may make, run and propagate covered works that you do not
|
165 |
+
convey, without conditions so long as your license otherwise remains
|
166 |
+
in force. You may convey covered works to others for the sole purpose
|
167 |
+
of having them make modifications exclusively for you, or provide you
|
168 |
+
with facilities for running those works, provided that you comply with
|
169 |
+
the terms of this License in conveying all material for which you do
|
170 |
+
not control copyright. Those thus making or running the covered works
|
171 |
+
for you must do so exclusively on your behalf, under your direction
|
172 |
+
and control, on terms that prohibit them from making any copies of
|
173 |
+
your copyrighted material outside their relationship with you.
|
174 |
+
|
175 |
+
Conveying under any other circumstances is permitted solely under
|
176 |
+
the conditions stated below. Sublicensing is not allowed; section 10
|
177 |
+
makes it unnecessary.
|
178 |
+
|
179 |
+
3. Protecting Users' Legal Rights From Anti-Circumvention Law.
|
180 |
+
|
181 |
+
No covered work shall be deemed part of an effective technological
|
182 |
+
measure under any applicable law fulfilling obligations under article
|
183 |
+
11 of the WIPO copyright treaty adopted on 20 December 1996, or
|
184 |
+
similar laws prohibiting or restricting circumvention of such
|
185 |
+
measures.
|
186 |
+
|
187 |
+
When you convey a covered work, you waive any legal power to forbid
|
188 |
+
circumvention of technological measures to the extent such circumvention
|
189 |
+
is effected by exercising rights under this License with respect to
|
190 |
+
the covered work, and you disclaim any intention to limit operation or
|
191 |
+
modification of the work as a means of enforcing, against the work's
|
192 |
+
users, your or third parties' legal rights to forbid circumvention of
|
193 |
+
technological measures.
|
194 |
+
|
195 |
+
4. Conveying Verbatim Copies.
|
196 |
+
|
197 |
+
You may convey verbatim copies of the Program's source code as you
|
198 |
+
receive it, in any medium, provided that you conspicuously and
|
199 |
+
appropriately publish on each copy an appropriate copyright notice;
|
200 |
+
keep intact all notices stating that this License and any
|
201 |
+
non-permissive terms added in accord with section 7 apply to the code;
|
202 |
+
keep intact all notices of the absence of any warranty; and give all
|
203 |
+
recipients a copy of this License along with the Program.
|
204 |
+
|
205 |
+
You may charge any price or no price for each copy that you convey,
|
206 |
+
and you may offer support or warranty protection for a fee.
|
207 |
+
|
208 |
+
5. Conveying Modified Source Versions.
|
209 |
+
|
210 |
+
You may convey a work based on the Program, or the modifications to
|
211 |
+
produce it from the Program, in the form of source code under the
|
212 |
+
terms of section 4, provided that you also meet all of these conditions:
|
213 |
+
|
214 |
+
a) The work must carry prominent notices stating that you modified
|
215 |
+
it, and giving a relevant date.
|
216 |
+
|
217 |
+
b) The work must carry prominent notices stating that it is
|
218 |
+
released under this License and any conditions added under section
|
219 |
+
7. This requirement modifies the requirement in section 4 to
|
220 |
+
"keep intact all notices".
|
221 |
+
|
222 |
+
c) You must license the entire work, as a whole, under this
|
223 |
+
License to anyone who comes into possession of a copy. This
|
224 |
+
License will therefore apply, along with any applicable section 7
|
225 |
+
additional terms, to the whole of the work, and all its parts,
|
226 |
+
regardless of how they are packaged. This License gives no
|
227 |
+
permission to license the work in any other way, but it does not
|
228 |
+
invalidate such permission if you have separately received it.
|
229 |
+
|
230 |
+
d) If the work has interactive user interfaces, each must display
|
231 |
+
Appropriate Legal Notices; however, if the Program has interactive
|
232 |
+
interfaces that do not display Appropriate Legal Notices, your
|
233 |
+
work need not make them do so.
|
234 |
+
|
235 |
+
A compilation of a covered work with other separate and independent
|
236 |
+
works, which are not by their nature extensions of the covered work,
|
237 |
+
and which are not combined with it such as to form a larger program,
|
238 |
+
in or on a volume of a storage or distribution medium, is called an
|
239 |
+
"aggregate" if the compilation and its resulting copyright are not
|
240 |
+
used to limit the access or legal rights of the compilation's users
|
241 |
+
beyond what the individual works permit. Inclusion of a covered work
|
242 |
+
in an aggregate does not cause this License to apply to the other
|
243 |
+
parts of the aggregate.
|
244 |
+
|
245 |
+
6. Conveying Non-Source Forms.
|
246 |
+
|
247 |
+
You may convey a covered work in object code form under the terms
|
248 |
+
of sections 4 and 5, provided that you also convey the
|
249 |
+
machine-readable Corresponding Source under the terms of this License,
|
250 |
+
in one of these ways:
|
251 |
+
|
252 |
+
a) Convey the object code in, or embodied in, a physical product
|
253 |
+
(including a physical distribution medium), accompanied by the
|
254 |
+
Corresponding Source fixed on a durable physical medium
|
255 |
+
customarily used for software interchange.
|
256 |
+
|
257 |
+
b) Convey the object code in, or embodied in, a physical product
|
258 |
+
(including a physical distribution medium), accompanied by a
|
259 |
+
written offer, valid for at least three years and valid for as
|
260 |
+
long as you offer spare parts or customer support for that product
|
261 |
+
model, to give anyone who possesses the object code either (1) a
|
262 |
+
copy of the Corresponding Source for all the software in the
|
263 |
+
product that is covered by this License, on a durable physical
|
264 |
+
medium customarily used for software interchange, for a price no
|
265 |
+
more than your reasonable cost of physically performing this
|
266 |
+
conveying of source, or (2) access to copy the
|
267 |
+
Corresponding Source from a network server at no charge.
|
268 |
+
|
269 |
+
c) Convey individual copies of the object code with a copy of the
|
270 |
+
written offer to provide the Corresponding Source. This
|
271 |
+
alternative is allowed only occasionally and noncommercially, and
|
272 |
+
only if you received the object code with such an offer, in accord
|
273 |
+
with subsection 6b.
|
274 |
+
|
275 |
+
d) Convey the object code by offering access from a designated
|
276 |
+
place (gratis or for a charge), and offer equivalent access to the
|
277 |
+
Corresponding Source in the same way through the same place at no
|
278 |
+
further charge. You need not require recipients to copy the
|
279 |
+
Corresponding Source along with the object code. If the place to
|
280 |
+
copy the object code is a network server, the Corresponding Source
|
281 |
+
may be on a different server (operated by you or a third party)
|
282 |
+
that supports equivalent copying facilities, provided you maintain
|
283 |
+
clear directions next to the object code saying where to find the
|
284 |
+
Corresponding Source. Regardless of what server hosts the
|
285 |
+
Corresponding Source, you remain obligated to ensure that it is
|
286 |
+
available for as long as needed to satisfy these requirements.
|
287 |
+
|
288 |
+
e) Convey the object code using peer-to-peer transmission, provided
|
289 |
+
you inform other peers where the object code and Corresponding
|
290 |
+
Source of the work are being offered to the general public at no
|
291 |
+
charge under subsection 6d.
|
292 |
+
|
293 |
+
A separable portion of the object code, whose source code is excluded
|
294 |
+
from the Corresponding Source as a System Library, need not be
|
295 |
+
included in conveying the object code work.
|
296 |
+
|
297 |
+
A "User Product" is either (1) a "consumer product", which means any
|
298 |
+
tangible personal property which is normally used for personal, family,
|
299 |
+
or household purposes, or (2) anything designed or sold for incorporation
|
300 |
+
into a dwelling. In determining whether a product is a consumer product,
|
301 |
+
doubtful cases shall be resolved in favor of coverage. For a particular
|
302 |
+
product received by a particular user, "normally used" refers to a
|
303 |
+
typical or common use of that class of product, regardless of the status
|
304 |
+
of the particular user or of the way in which the particular user
|
305 |
+
actually uses, or expects or is expected to use, the product. A product
|
306 |
+
is a consumer product regardless of whether the product has substantial
|
307 |
+
commercial, industrial or non-consumer uses, unless such uses represent
|
308 |
+
the only significant mode of use of the product.
|
309 |
+
|
310 |
+
"Installation Information" for a User Product means any methods,
|
311 |
+
procedures, authorization keys, or other information required to install
|
312 |
+
and execute modified versions of a covered work in that User Product from
|
313 |
+
a modified version of its Corresponding Source. The information must
|
314 |
+
suffice to ensure that the continued functioning of the modified object
|
315 |
+
code is in no case prevented or interfered with solely because
|
316 |
+
modification has been made.
|
317 |
+
|
318 |
+
If you convey an object code work under this section in, or with, or
|
319 |
+
specifically for use in, a User Product, and the conveying occurs as
|
320 |
+
part of a transaction in which the right of possession and use of the
|
321 |
+
User Product is transferred to the recipient in perpetuity or for a
|
322 |
+
fixed term (regardless of how the transaction is characterized), the
|
323 |
+
Corresponding Source conveyed under this section must be accompanied
|
324 |
+
by the Installation Information. But this requirement does not apply
|
325 |
+
if neither you nor any third party retains the ability to install
|
326 |
+
modified object code on the User Product (for example, the work has
|
327 |
+
been installed in ROM).
|
328 |
+
|
329 |
+
The requirement to provide Installation Information does not include a
|
330 |
+
requirement to continue to provide support service, warranty, or updates
|
331 |
+
for a work that has been modified or installed by the recipient, or for
|
332 |
+
the User Product in which it has been modified or installed. Access to a
|
333 |
+
network may be denied when the modification itself materially and
|
334 |
+
adversely affects the operation of the network or violates the rules and
|
335 |
+
protocols for communication across the network.
|
336 |
+
|
337 |
+
Corresponding Source conveyed, and Installation Information provided,
|
338 |
+
in accord with this section must be in a format that is publicly
|
339 |
+
documented (and with an implementation available to the public in
|
340 |
+
source code form), and must require no special password or key for
|
341 |
+
unpacking, reading or copying.
|
342 |
+
|
343 |
+
7. Additional Terms.
|
344 |
+
|
345 |
+
"Additional permissions" are terms that supplement the terms of this
|
346 |
+
License by making exceptions from one or more of its conditions.
|
347 |
+
Additional permissions that are applicable to the entire Program shall
|
348 |
+
be treated as though they were included in this License, to the extent
|
349 |
+
that they are valid under applicable law. If additional permissions
|
350 |
+
apply only to part of the Program, that part may be used separately
|
351 |
+
under those permissions, but the entire Program remains governed by
|
352 |
+
this License without regard to the additional permissions.
|
353 |
+
|
354 |
+
When you convey a copy of a covered work, you may at your option
|
355 |
+
remove any additional permissions from that copy, or from any part of
|
356 |
+
it. (Additional permissions may be written to require their own
|
357 |
+
removal in certain cases when you modify the work.) You may place
|
358 |
+
additional permissions on material, added by you to a covered work,
|
359 |
+
for which you have or can give appropriate copyright permission.
|
360 |
+
|
361 |
+
Notwithstanding any other provision of this License, for material you
|
362 |
+
add to a covered work, you may (if authorized by the copyright holders of
|
363 |
+
that material) supplement the terms of this License with terms:
|
364 |
+
|
365 |
+
a) Disclaiming warranty or limiting liability differently from the
|
366 |
+
terms of sections 15 and 16 of this License; or
|
367 |
+
|
368 |
+
b) Requiring preservation of specified reasonable legal notices or
|
369 |
+
author attributions in that material or in the Appropriate Legal
|
370 |
+
Notices displayed by works containing it; or
|
371 |
+
|
372 |
+
c) Prohibiting misrepresentation of the origin of that material, or
|
373 |
+
requiring that modified versions of such material be marked in
|
374 |
+
reasonable ways as different from the original version; or
|
375 |
+
|
376 |
+
d) Limiting the use for publicity purposes of names of licensors or
|
377 |
+
authors of the material; or
|
378 |
+
|
379 |
+
e) Declining to grant rights under trademark law for use of some
|
380 |
+
trade names, trademarks, or service marks; or
|
381 |
+
|
382 |
+
f) Requiring indemnification of licensors and authors of that
|
383 |
+
material by anyone who conveys the material (or modified versions of
|
384 |
+
it) with contractual assumptions of liability to the recipient, for
|
385 |
+
any liability that these contractual assumptions directly impose on
|
386 |
+
those licensors and authors.
|
387 |
+
|
388 |
+
All other non-permissive additional terms are considered "further
|
389 |
+
restrictions" within the meaning of section 10. If the Program as you
|
390 |
+
received it, or any part of it, contains a notice stating that it is
|
391 |
+
governed by this License along with a term that is a further
|
392 |
+
restriction, you may remove that term. If a license document contains
|
393 |
+
a further restriction but permits relicensing or conveying under this
|
394 |
+
License, you may add to a covered work material governed by the terms
|
395 |
+
of that license document, provided that the further restriction does
|
396 |
+
not survive such relicensing or conveying.
|
397 |
+
|
398 |
+
If you add terms to a covered work in accord with this section, you
|
399 |
+
must place, in the relevant source files, a statement of the
|
400 |
+
additional terms that apply to those files, or a notice indicating
|
401 |
+
where to find the applicable terms.
|
402 |
+
|
403 |
+
Additional terms, permissive or non-permissive, may be stated in the
|
404 |
+
form of a separately written license, or stated as exceptions;
|
405 |
+
the above requirements apply either way.
|
406 |
+
|
407 |
+
8. Termination.
|
408 |
+
|
409 |
+
You may not propagate or modify a covered work except as expressly
|
410 |
+
provided under this License. Any attempt otherwise to propagate or
|
411 |
+
modify it is void, and will automatically terminate your rights under
|
412 |
+
this License (including any patent licenses granted under the third
|
413 |
+
paragraph of section 11).
|
414 |
+
|
415 |
+
However, if you cease all violation of this License, then your
|
416 |
+
license from a particular copyright holder is reinstated (a)
|
417 |
+
provisionally, unless and until the copyright holder explicitly and
|
418 |
+
finally terminates your license, and (b) permanently, if the copyright
|
419 |
+
holder fails to notify you of the violation by some reasonable means
|
420 |
+
prior to 60 days after the cessation.
|
421 |
+
|
422 |
+
Moreover, your license from a particular copyright holder is
|
423 |
+
reinstated permanently if the copyright holder notifies you of the
|
424 |
+
violation by some reasonable means, this is the first time you have
|
425 |
+
received notice of violation of this License (for any work) from that
|
426 |
+
copyright holder, and you cure the violation prior to 30 days after
|
427 |
+
your receipt of the notice.
|
428 |
+
|
429 |
+
Termination of your rights under this section does not terminate the
|
430 |
+
licenses of parties who have received copies or rights from you under
|
431 |
+
this License. If your rights have been terminated and not permanently
|
432 |
+
reinstated, you do not qualify to receive new licenses for the same
|
433 |
+
material under section 10.
|
434 |
+
|
435 |
+
9. Acceptance Not Required for Having Copies.
|
436 |
+
|
437 |
+
You are not required to accept this License in order to receive or
|
438 |
+
run a copy of the Program. Ancillary propagation of a covered work
|
439 |
+
occurring solely as a consequence of using peer-to-peer transmission
|
440 |
+
to receive a copy likewise does not require acceptance. However,
|
441 |
+
nothing other than this License grants you permission to propagate or
|
442 |
+
modify any covered work. These actions infringe copyright if you do
|
443 |
+
not accept this License. Therefore, by modifying or propagating a
|
444 |
+
covered work, you indicate your acceptance of this License to do so.
|
445 |
+
|
446 |
+
10. Automatic Licensing of Downstream Recipients.
|
447 |
+
|
448 |
+
Each time you convey a covered work, the recipient automatically
|
449 |
+
receives a license from the original licensors, to run, modify and
|
450 |
+
propagate that work, subject to this License. You are not responsible
|
451 |
+
for enforcing compliance by third parties with this License.
|
452 |
+
|
453 |
+
An "entity transaction" is a transaction transferring control of an
|
454 |
+
organization, or substantially all assets of one, or subdividing an
|
455 |
+
organization, or merging organizations. If propagation of a covered
|
456 |
+
work results from an entity transaction, each party to that
|
457 |
+
transaction who receives a copy of the work also receives whatever
|
458 |
+
licenses to the work the party's predecessor in interest had or could
|
459 |
+
give under the previous paragraph, plus a right to possession of the
|
460 |
+
Corresponding Source of the work from the predecessor in interest, if
|
461 |
+
the predecessor has it or can get it with reasonable efforts.
|
462 |
+
|
463 |
+
You may not impose any further restrictions on the exercise of the
|
464 |
+
rights granted or affirmed under this License. For example, you may
|
465 |
+
not impose a license fee, royalty, or other charge for exercise of
|
466 |
+
rights granted under this License, and you may not initiate litigation
|
467 |
+
(including a cross-claim or counterclaim in a lawsuit) alleging that
|
468 |
+
any patent claim is infringed by making, using, selling, offering for
|
469 |
+
sale, or importing the Program or any portion of it.
|
470 |
+
|
471 |
+
11. Patents.
|
472 |
+
|
473 |
+
A "contributor" is a copyright holder who authorizes use under this
|
474 |
+
License of the Program or a work on which the Program is based. The
|
475 |
+
work thus licensed is called the contributor's "contributor version".
|
476 |
+
|
477 |
+
A contributor's "essential patent claims" are all patent claims
|
478 |
+
owned or controlled by the contributor, whether already acquired or
|
479 |
+
hereafter acquired, that would be infringed by some manner, permitted
|
480 |
+
by this License, of making, using, or selling its contributor version,
|
481 |
+
but do not include claims that would be infringed only as a
|
482 |
+
consequence of further modification of the contributor version. For
|
483 |
+
purposes of this definition, "control" includes the right to grant
|
484 |
+
patent sublicenses in a manner consistent with the requirements of
|
485 |
+
this License.
|
486 |
+
|
487 |
+
Each contributor grants you a non-exclusive, worldwide, royalty-free
|
488 |
+
patent license under the contributor's essential patent claims, to
|
489 |
+
make, use, sell, offer for sale, import and otherwise run, modify and
|
490 |
+
propagate the contents of its contributor version.
|
491 |
+
|
492 |
+
In the following three paragraphs, a "patent license" is any express
|
493 |
+
agreement or commitment, however denominated, not to enforce a patent
|
494 |
+
(such as an express permission to practice a patent or covenant not to
|
495 |
+
sue for patent infringement). To "grant" such a patent license to a
|
496 |
+
party means to make such an agreement or commitment not to enforce a
|
497 |
+
patent against the party.
|
498 |
+
|
499 |
+
If you convey a covered work, knowingly relying on a patent license,
|
500 |
+
and the Corresponding Source of the work is not available for anyone
|
501 |
+
to copy, free of charge and under the terms of this License, through a
|
502 |
+
publicly available network server or other readily accessible means,
|
503 |
+
then you must either (1) cause the Corresponding Source to be so
|
504 |
+
available, or (2) arrange to deprive yourself of the benefit of the
|
505 |
+
patent license for this particular work, or (3) arrange, in a manner
|
506 |
+
consistent with the requirements of this License, to extend the patent
|
507 |
+
license to downstream recipients. "Knowingly relying" means you have
|
508 |
+
actual knowledge that, but for the patent license, your conveying the
|
509 |
+
covered work in a country, or your recipient's use of the covered work
|
510 |
+
in a country, would infringe one or more identifiable patents in that
|
511 |
+
country that you have reason to believe are valid.
|
512 |
+
|
513 |
+
If, pursuant to or in connection with a single transaction or
|
514 |
+
arrangement, you convey, or propagate by procuring conveyance of, a
|
515 |
+
covered work, and grant a patent license to some of the parties
|
516 |
+
receiving the covered work authorizing them to use, propagate, modify
|
517 |
+
or convey a specific copy of the covered work, then the patent license
|
518 |
+
you grant is automatically extended to all recipients of the covered
|
519 |
+
work and works based on it.
|
520 |
+
|
521 |
+
A patent license is "discriminatory" if it does not include within
|
522 |
+
the scope of its coverage, prohibits the exercise of, or is
|
523 |
+
conditioned on the non-exercise of one or more of the rights that are
|
524 |
+
specifically granted under this License. You may not convey a covered
|
525 |
+
work if you are a party to an arrangement with a third party that is
|
526 |
+
in the business of distributing software, under which you make payment
|
527 |
+
to the third party based on the extent of your activity of conveying
|
528 |
+
the work, and under which the third party grants, to any of the
|
529 |
+
parties who would receive the covered work from you, a discriminatory
|
530 |
+
patent license (a) in connection with copies of the covered work
|
531 |
+
conveyed by you (or copies made from those copies), or (b) primarily
|
532 |
+
for and in connection with specific products or compilations that
|
533 |
+
contain the covered work, unless you entered into that arrangement,
|
534 |
+
or that patent license was granted, prior to 28 March 2007.
|
535 |
+
|
536 |
+
Nothing in this License shall be construed as excluding or limiting
|
537 |
+
any implied license or other defenses to infringement that may
|
538 |
+
otherwise be available to you under applicable patent law.
|
539 |
+
|
540 |
+
12. No Surrender of Others' Freedom.
|
541 |
+
|
542 |
+
If conditions are imposed on you (whether by court order, agreement or
|
543 |
+
otherwise) that contradict the conditions of this License, they do not
|
544 |
+
excuse you from the conditions of this License. If you cannot convey a
|
545 |
+
covered work so as to satisfy simultaneously your obligations under this
|
546 |
+
License and any other pertinent obligations, then as a consequence you may
|
547 |
+
not convey it at all. For example, if you agree to terms that obligate you
|
548 |
+
to collect a royalty for further conveying from those to whom you convey
|
549 |
+
the Program, the only way you could satisfy both those terms and this
|
550 |
+
License would be to refrain entirely from conveying the Program.
|
551 |
+
|
552 |
+
13. Use with the GNU Affero General Public License.
|
553 |
+
|
554 |
+
Notwithstanding any other provision of this License, you have
|
555 |
+
permission to link or combine any covered work with a work licensed
|
556 |
+
under version 3 of the GNU Affero General Public License into a single
|
557 |
+
combined work, and to convey the resulting work. The terms of this
|
558 |
+
License will continue to apply to the part which is the covered work,
|
559 |
+
but the special requirements of the GNU Affero General Public License,
|
560 |
+
section 13, concerning interaction through a network will apply to the
|
561 |
+
combination as such.
|
562 |
+
|
563 |
+
14. Revised Versions of this License.
|
564 |
+
|
565 |
+
The Free Software Foundation may publish revised and/or new versions of
|
566 |
+
the GNU General Public License from time to time. Such new versions will
|
567 |
+
be similar in spirit to the present version, but may differ in detail to
|
568 |
+
address new problems or concerns.
|
569 |
+
|
570 |
+
Each version is given a distinguishing version number. If the
|
571 |
+
Program specifies that a certain numbered version of the GNU General
|
572 |
+
Public License "or any later version" applies to it, you have the
|
573 |
+
option of following the terms and conditions either of that numbered
|
574 |
+
version or of any later version published by the Free Software
|
575 |
+
Foundation. If the Program does not specify a version number of the
|
576 |
+
GNU General Public License, you may choose any version ever published
|
577 |
+
by the Free Software Foundation.
|
578 |
+
|
579 |
+
If the Program specifies that a proxy can decide which future
|
580 |
+
versions of the GNU General Public License can be used, that proxy's
|
581 |
+
public statement of acceptance of a version permanently authorizes you
|
582 |
+
to choose that version for the Program.
|
583 |
+
|
584 |
+
Later license versions may give you additional or different
|
585 |
+
permissions. However, no additional obligations are imposed on any
|
586 |
+
author or copyright holder as a result of your choosing to follow a
|
587 |
+
later version.
|
588 |
+
|
589 |
+
15. Disclaimer of Warranty.
|
590 |
+
|
591 |
+
THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
|
592 |
+
APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
|
593 |
+
HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
|
594 |
+
OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
|
595 |
+
THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
596 |
+
PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
|
597 |
+
IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
|
598 |
+
ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
|
599 |
+
|
600 |
+
16. Limitation of Liability.
|
601 |
+
|
602 |
+
IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
|
603 |
+
WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
|
604 |
+
THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
|
605 |
+
GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
|
606 |
+
USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
|
607 |
+
DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
|
608 |
+
PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
|
609 |
+
EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
|
610 |
+
SUCH DAMAGES.
|
611 |
+
|
612 |
+
17. Interpretation of Sections 15 and 16.
|
613 |
+
|
614 |
+
If the disclaimer of warranty and limitation of liability provided
|
615 |
+
above cannot be given local legal effect according to their terms,
|
616 |
+
reviewing courts shall apply local law that most closely approximates
|
617 |
+
an absolute waiver of all civil liability in connection with the
|
618 |
+
Program, unless a warranty or assumption of liability accompanies a
|
619 |
+
copy of the Program in return for a fee.
|
620 |
+
|
621 |
+
END OF TERMS AND CONDITIONS
|
622 |
+
|
623 |
+
How to Apply These Terms to Your New Programs
|
624 |
+
|
625 |
+
If you develop a new program, and you want it to be of the greatest
|
626 |
+
possible use to the public, the best way to achieve this is to make it
|
627 |
+
free software which everyone can redistribute and change under these terms.
|
628 |
+
|
629 |
+
To do so, attach the following notices to the program. It is safest
|
630 |
+
to attach them to the start of each source file to most effectively
|
631 |
+
state the exclusion of warranty; and each file should have at least
|
632 |
+
the "copyright" line and a pointer to where the full notice is found.
|
633 |
+
|
634 |
+
<one line to give the program's name and a brief idea of what it does.>
|
635 |
+
Copyright (C) <year> <name of author>
|
636 |
+
|
637 |
+
This program is free software: you can redistribute it and/or modify
|
638 |
+
it under the terms of the GNU General Public License as published by
|
639 |
+
the Free Software Foundation, either version 3 of the License, or
|
640 |
+
(at your option) any later version.
|
641 |
+
|
642 |
+
This program is distributed in the hope that it will be useful,
|
643 |
+
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
644 |
+
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
645 |
+
GNU General Public License for more details.
|
646 |
+
|
647 |
+
You should have received a copy of the GNU General Public License
|
648 |
+
along with this program. If not, see <https://www.gnu.org/licenses/>.
|
649 |
+
|
650 |
+
Also add information on how to contact you by electronic and paper mail.
|
651 |
+
|
652 |
+
If the program does terminal interaction, make it output a short
|
653 |
+
notice like this when it starts in an interactive mode:
|
654 |
+
|
655 |
+
<program> Copyright (C) <year> <name of author>
|
656 |
+
This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
|
657 |
+
This is free software, and you are welcome to redistribute it
|
658 |
+
under certain conditions; type `show c' for details.
|
659 |
+
|
660 |
+
The hypothetical commands `show w' and `show c' should show the appropriate
|
661 |
+
parts of the General Public License. Of course, your program's commands
|
662 |
+
might be different; for a GUI interface, you would use an "about box".
|
663 |
+
|
664 |
+
You should also get your employer (if you work as a programmer) or school,
|
665 |
+
if any, to sign a "copyright disclaimer" for the program, if necessary.
|
666 |
+
For more information on this, and how to apply and follow the GNU GPL, see
|
667 |
+
<https://www.gnu.org/licenses/>.
|
668 |
+
|
669 |
+
The GNU General Public License does not permit incorporating your program
|
670 |
+
into proprietary programs. If your program is a subroutine library, you
|
671 |
+
may consider it more useful to permit linking proprietary applications with
|
672 |
+
the library. If this is what you want to do, use the GNU Lesser General
|
673 |
+
Public License instead of this License. But first, please read
|
674 |
+
<https://www.gnu.org/licenses/why-not-lgpl.html>.
|
README.md
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
---
|
2 |
title: YOLOR
|
3 |
-
emoji:
|
4 |
colorFrom: gray
|
5 |
colorTo: purple
|
6 |
sdk: gradio
|
@@ -35,3 +35,226 @@ Path is relative to the root of the repository.
|
|
35 |
|
36 |
`pinned`: _boolean_
|
37 |
Whether the Space stays on top of your list.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
title: YOLOR
|
3 |
+
emoji: 🚀
|
4 |
colorFrom: gray
|
5 |
colorTo: purple
|
6 |
sdk: gradio
|
|
|
35 |
|
36 |
`pinned`: _boolean_
|
37 |
Whether the Space stays on top of your list.
|
38 |
+
|
39 |
+
|
40 |
+
# YOLOR
|
41 |
+
implementation of paper - [You Only Learn One Representation: Unified Network for Multiple Tasks](https://arxiv.org/abs/2105.04206)
|
42 |
+
|
43 |
+
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/you-only-learn-one-representation-unified/real-time-object-detection-on-coco)](https://paperswithcode.com/sota/real-time-object-detection-on-coco?p=you-only-learn-one-representation-unified)
|
44 |
+
|
45 |
+
![Unified Network](https://github.com/WongKinYiu/yolor/blob/main/figure/unifued_network.png)
|
46 |
+
|
47 |
+
<img src="https://github.com/WongKinYiu/yolor/blob/main/figure/performance.png" height="480">
|
48 |
+
|
49 |
+
To get the results on the table, please use [this branch](https://github.com/WongKinYiu/yolor/tree/paper).
|
50 |
+
|
51 |
+
| Model | Test Size | AP<sup>test</sup> | AP<sub>50</sub><sup>test</sup> | AP<sub>75</sub><sup>test</sup> | batch1 throughput | batch32 inference |
|
52 |
+
| :-- | :-: | :-: | :-: | :-: | :-: | :-: |
|
53 |
+
| **YOLOR-P6** | 1280 | **54.1%** | **71.8%** | **59.3%** | 49 *fps* | 8.3 *ms* |
|
54 |
+
| **YOLOR-W6** | 1280 | **55.5%** | **73.2%** | **60.6%** | 47 *fps* | 10.7 *ms* |
|
55 |
+
| **YOLOR-E6** | 1280 | **56.4%** | **74.1%** | **61.6%** | 37 *fps* | 17.1 *ms* |
|
56 |
+
| **YOLOR-D6** | 1280 | **57.3%** | **75.0%** | **62.7%** | 30 *fps* | 21.8 *ms* |
|
57 |
+
| **YOLOR-D6*** | 1280 | **57.8%** | **75.5%** | **63.3%** | 30 *fps* | 21.8 *ms* |
|
58 |
+
| | | | | | | |
|
59 |
+
| **YOLOv4-P5** | 896 | **51.8%** | **70.3%** | **56.6%** | 41 *fps* | - |
|
60 |
+
| **YOLOv4-P6** | 1280 | **54.5%** | **72.6%** | **59.8%** | 30 *fps* | - |
|
61 |
+
| **YOLOv4-P7** | 1536 | **55.5%** | **73.4%** | **60.8%** | 16 *fps* | - |
|
62 |
+
| | | | | | | |
|
63 |
+
|
64 |
+
To reproduce the inference speed, please see [darknet](https://github.com/WongKinYiu/yolor/tree/main/darknet).
|
65 |
+
|
66 |
+
| Model | Test Size | AP<sup>val</sup> | AP<sub>50</sub><sup>val</sup> | AP<sub>75</sub><sup>val</sup> | AP<sub>S</sub><sup>val</sup> | AP<sub>M</sub><sup>val</sup> | AP<sub>L</sub><sup>val</sup> | batch1 throughput |
|
67 |
+
| :-- | :-: | :-: | :-: | :-: | :-: | :-: | :-: | :-: |
|
68 |
+
| [**YOLOv4-CSP**](/cfg/yolov4_csp.cfg) | 640 | **49.1%** | **67.7%** | **53.8%** | **32.1%** | **54.4%** | **63.2%** | 76 *fps* |
|
69 |
+
| [**YOLOR-CSP**](/cfg/yolor_csp.cfg) | 640 | **49.2%** | **67.6%** | **53.7%** | **32.9%** | **54.4%** | **63.0%** | [weights](https://drive.google.com/file/d/1ZEqGy4kmZyD-Cj3tEFJcLSZenZBDGiyg/view?usp=sharing) |
|
70 |
+
| [**YOLOR-CSP***](/cfg/yolor_csp.cfg) | 640 | **50.0%** | **68.7%** | **54.3%** | **34.2%** | **55.1%** | **64.3%** | [weights](https://drive.google.com/file/d/1OJKgIasELZYxkIjFoiqyn555bcmixUP2/view?usp=sharing) |
|
71 |
+
| | | | | | | |
|
72 |
+
| [**YOLOv4-CSP-X**](/cfg/yolov4_csp_x.cfg) | 640 | **50.9%** | **69.3%** | **55.4%** | **35.3%** | **55.8%** | **64.8%** | 53 *fps* |
|
73 |
+
| [**YOLOR-CSP-X**](/cfg/yolor_csp_x.cfg) | 640 | **51.1%** | **69.6%** | **55.7%** | **35.7%** | **56.0%** | **65.2%** | [weights](https://drive.google.com/file/d/1L29rfIPNH1n910qQClGftknWpTBgAv6c/view?usp=sharing) |
|
74 |
+
| [**YOLOR-CSP-X***](/cfg/yolor_csp_x.cfg) | 640 | **51.5%** | **69.9%** | **56.1%** | **35.8%** | **56.8%** | **66.1%** | [weights](https://drive.google.com/file/d/1NbMG3ivuBQ4S8kEhFJ0FIqOQXevGje_w/view?usp=sharing) |
|
75 |
+
| | | | | | | |
|
76 |
+
|
77 |
+
Developing...
|
78 |
+
|
79 |
+
| Model | Test Size | AP<sup>test</sup> | AP<sub>50</sub><sup>test</sup> | AP<sub>75</sub><sup>test</sup> | AP<sub>S</sub><sup>test</sup> | AP<sub>M</sub><sup>test</sup> | AP<sub>L</sub><sup>test</sup> |
|
80 |
+
| :-- | :-: | :-: | :-: | :-: | :-: | :-: | :-: |
|
81 |
+
| **YOLOR-CSP** | 640 | **51.1%** | **69.6%** | **55.7%** | **31.7%** | **55.3%** | **64.7%** |
|
82 |
+
| **YOLOR-CSP-X** | 640 | **53.0%** | **71.4%** | **57.9%** | **33.7%** | **57.1%** | **66.8%** |
|
83 |
+
|
84 |
+
Train from scratch for 300 epochs...
|
85 |
+
|
86 |
+
| Model | Info | Test Size | AP |
|
87 |
+
| :-- | :-- | :-: | :-: |
|
88 |
+
| **YOLOR-CSP** | [evolution](https://github.com/ultralytics/yolov3/issues/392) | 640 | **48.0%** |
|
89 |
+
| **YOLOR-CSP** | [strategy](https://openaccess.thecvf.com/content/ICCV2021W/LPCV/html/Wang_Exploring_the_Power_of_Lightweight_YOLOv4_ICCVW_2021_paper.html) | 640 | **50.0%** |
|
90 |
+
| **YOLOR-CSP** | [strategy](https://openaccess.thecvf.com/content/ICCV2021W/LPCV/html/Wang_Exploring_the_Power_of_Lightweight_YOLOv4_ICCVW_2021_paper.html) + [simOTA](https://arxiv.org/abs/2107.08430) | 640 | **51.1%** |
|
91 |
+
| | | | |
|
92 |
+
| **YOLOR-CSP-X** | [strategy](https://openaccess.thecvf.com/content/ICCV2021W/LPCV/html/Wang_Exploring_the_Power_of_Lightweight_YOLOv4_ICCVW_2021_paper.html) | 640 | **51.5%** |
|
93 |
+
| **YOLOR-CSP-X** | [strategy](https://openaccess.thecvf.com/content/ICCV2021W/LPCV/html/Wang_Exploring_the_Power_of_Lightweight_YOLOv4_ICCVW_2021_paper.html) + [simOTA](https://arxiv.org/abs/2107.08430) | 640 | **53.0%** |
|
94 |
+
|
95 |
+
## Installation
|
96 |
+
|
97 |
+
Docker environment (recommended)
|
98 |
+
<details><summary> <b>Expand</b> </summary>
|
99 |
+
|
100 |
+
```
|
101 |
+
# create the docker container, you can change the share memory size if you have more.
|
102 |
+
nvidia-docker run --name yolor -it -v your_coco_path/:/coco/ -v your_code_path/:/yolor --shm-size=64g nvcr.io/nvidia/pytorch:20.11-py3
|
103 |
+
|
104 |
+
# apt install required packages
|
105 |
+
apt update
|
106 |
+
apt install -y zip htop screen libgl1-mesa-glx
|
107 |
+
|
108 |
+
# pip install required packages
|
109 |
+
pip install seaborn thop
|
110 |
+
|
111 |
+
# install mish-cuda if you want to use mish activation
|
112 |
+
# https://github.com/thomasbrandon/mish-cuda
|
113 |
+
# https://github.com/JunnYu/mish-cuda
|
114 |
+
cd /
|
115 |
+
git clone https://github.com/JunnYu/mish-cuda
|
116 |
+
cd mish-cuda
|
117 |
+
python setup.py build install
|
118 |
+
|
119 |
+
# install pytorch_wavelets if you want to use dwt down-sampling module
|
120 |
+
# https://github.com/fbcotter/pytorch_wavelets
|
121 |
+
cd /
|
122 |
+
git clone https://github.com/fbcotter/pytorch_wavelets
|
123 |
+
cd pytorch_wavelets
|
124 |
+
pip install .
|
125 |
+
|
126 |
+
# go to code folder
|
127 |
+
cd /yolor
|
128 |
+
```
|
129 |
+
|
130 |
+
</details>
|
131 |
+
|
132 |
+
Colab environment
|
133 |
+
<details><summary> <b>Expand</b> </summary>
|
134 |
+
|
135 |
+
```
|
136 |
+
git clone https://github.com/WongKinYiu/yolor
|
137 |
+
cd yolor
|
138 |
+
|
139 |
+
# pip install required packages
|
140 |
+
pip install -qr requirements.txt
|
141 |
+
|
142 |
+
# install mish-cuda if you want to use mish activation
|
143 |
+
# https://github.com/thomasbrandon/mish-cuda
|
144 |
+
# https://github.com/JunnYu/mish-cuda
|
145 |
+
git clone https://github.com/JunnYu/mish-cuda
|
146 |
+
cd mish-cuda
|
147 |
+
python setup.py build install
|
148 |
+
cd ..
|
149 |
+
|
150 |
+
# install pytorch_wavelets if you want to use dwt down-sampling module
|
151 |
+
# https://github.com/fbcotter/pytorch_wavelets
|
152 |
+
git clone https://github.com/fbcotter/pytorch_wavelets
|
153 |
+
cd pytorch_wavelets
|
154 |
+
pip install .
|
155 |
+
cd ..
|
156 |
+
```
|
157 |
+
|
158 |
+
</details>
|
159 |
+
|
160 |
+
Prepare COCO dataset
|
161 |
+
<details><summary> <b>Expand</b> </summary>
|
162 |
+
|
163 |
+
```
|
164 |
+
cd /yolor
|
165 |
+
bash scripts/get_coco.sh
|
166 |
+
```
|
167 |
+
|
168 |
+
</details>
|
169 |
+
|
170 |
+
Prepare pretrained weight
|
171 |
+
<details><summary> <b>Expand</b> </summary>
|
172 |
+
|
173 |
+
```
|
174 |
+
cd /yolor
|
175 |
+
bash scripts/get_pretrain.sh
|
176 |
+
```
|
177 |
+
|
178 |
+
</details>
|
179 |
+
|
180 |
+
## Testing
|
181 |
+
|
182 |
+
[`yolor_p6.pt`](https://drive.google.com/file/d/1Tdn3yqpZ79X7R1Ql0zNlNScB1Dv9Fp76/view?usp=sharing)
|
183 |
+
|
184 |
+
```
|
185 |
+
python test.py --data data/coco.yaml --img 1280 --batch 32 --conf 0.001 --iou 0.65 --device 0 --cfg cfg/yolor_p6.cfg --weights yolor_p6.pt --name yolor_p6_val
|
186 |
+
```
|
187 |
+
|
188 |
+
You will get the results:
|
189 |
+
|
190 |
+
```
|
191 |
+
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.52510
|
192 |
+
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.70718
|
193 |
+
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.57520
|
194 |
+
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.37058
|
195 |
+
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.56878
|
196 |
+
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.66102
|
197 |
+
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.39181
|
198 |
+
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.65229
|
199 |
+
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.71441
|
200 |
+
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.57755
|
201 |
+
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.75337
|
202 |
+
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.84013
|
203 |
+
```
|
204 |
+
|
205 |
+
## Training
|
206 |
+
|
207 |
+
Single GPU training:
|
208 |
+
|
209 |
+
```
|
210 |
+
python train.py --batch-size 8 --img 1280 1280 --data coco.yaml --cfg cfg/yolor_p6.cfg --weights '' --device 0 --name yolor_p6 --hyp hyp.scratch.1280.yaml --epochs 300
|
211 |
+
```
|
212 |
+
|
213 |
+
Multiple GPU training:
|
214 |
+
|
215 |
+
```
|
216 |
+
python -m torch.distributed.launch --nproc_per_node 2 --master_port 9527 train.py --batch-size 16 --img 1280 1280 --data coco.yaml --cfg cfg/yolor_p6.cfg --weights '' --device 0,1 --sync-bn --name yolor_p6 --hyp hyp.scratch.1280.yaml --epochs 300
|
217 |
+
```
|
218 |
+
|
219 |
+
Training schedule in the paper:
|
220 |
+
|
221 |
+
```
|
222 |
+
python -m torch.distributed.launch --nproc_per_node 8 --master_port 9527 train.py --batch-size 64 --img 1280 1280 --data data/coco.yaml --cfg cfg/yolor_p6.cfg --weights '' --device 0,1,2,3,4,5,6,7 --sync-bn --name yolor_p6 --hyp hyp.scratch.1280.yaml --epochs 300
|
223 |
+
python -m torch.distributed.launch --nproc_per_node 8 --master_port 9527 tune.py --batch-size 64 --img 1280 1280 --data data/coco.yaml --cfg cfg/yolor_p6.cfg --weights 'runs/train/yolor_p6/weights/last_298.pt' --device 0,1,2,3,4,5,6,7 --sync-bn --name yolor_p6-tune --hyp hyp.finetune.1280.yaml --epochs 450
|
224 |
+
python -m torch.distributed.launch --nproc_per_node 8 --master_port 9527 train.py --batch-size 64 --img 1280 1280 --data data/coco.yaml --cfg cfg/yolor_p6.cfg --weights 'runs/train/yolor_p6-tune/weights/epoch_424.pt' --device 0,1,2,3,4,5,6,7 --sync-bn --name yolor_p6-fine --hyp hyp.finetune.1280.yaml --epochs 450
|
225 |
+
```
|
226 |
+
|
227 |
+
## Inference
|
228 |
+
|
229 |
+
[`yolor_p6.pt`](https://drive.google.com/file/d/1Tdn3yqpZ79X7R1Ql0zNlNScB1Dv9Fp76/view?usp=sharing)
|
230 |
+
|
231 |
+
```
|
232 |
+
python detect.py --source inference/images/horses.jpg --cfg cfg/yolor_p6.cfg --weights yolor_p6.pt --conf 0.25 --img-size 1280 --device 0
|
233 |
+
```
|
234 |
+
|
235 |
+
You will get the results:
|
236 |
+
|
237 |
+
![horses](https://github.com/WongKinYiu/yolor/blob/main/inference/output/horses.jpg)
|
238 |
+
|
239 |
+
## Citation
|
240 |
+
|
241 |
+
```
|
242 |
+
@article{wang2021you,
|
243 |
+
title={You Only Learn One Representation: Unified Network for Multiple Tasks},
|
244 |
+
author={Wang, Chien-Yao and Yeh, I-Hau and Liao, Hong-Yuan Mark},
|
245 |
+
journal={arXiv preprint arXiv:2105.04206},
|
246 |
+
year={2021}
|
247 |
+
}
|
248 |
+
```
|
249 |
+
|
250 |
+
## Acknowledgements
|
251 |
+
|
252 |
+
<details><summary> <b>Expand</b> </summary>
|
253 |
+
|
254 |
+
* [https://github.com/AlexeyAB/darknet](https://github.com/AlexeyAB/darknet)
|
255 |
+
* [https://github.com/WongKinYiu/PyTorch_YOLOv4](https://github.com/WongKinYiu/PyTorch_YOLOv4)
|
256 |
+
* [https://github.com/WongKinYiu/ScaledYOLOv4](https://github.com/WongKinYiu/ScaledYOLOv4)
|
257 |
+
* [https://github.com/ultralytics/yolov3](https://github.com/ultralytics/yolov3)
|
258 |
+
* [https://github.com/ultralytics/yolov5](https://github.com/ultralytics/yolov5)
|
259 |
+
|
260 |
+
</details>
|
app.py
ADDED
@@ -0,0 +1,104 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from PIL import Image
|
2 |
+
import cv2
|
3 |
+
import torch
|
4 |
+
from numpy import random
|
5 |
+
|
6 |
+
from utils.general import (non_max_suppression, scale_coords)
|
7 |
+
from utils.plots import plot_one_box
|
8 |
+
|
9 |
+
from models.models import *
|
10 |
+
from utils.datasets import *
|
11 |
+
from utils.general import *
|
12 |
+
|
13 |
+
import gradio as gr
|
14 |
+
import requests
|
15 |
+
|
16 |
+
import gdown
|
17 |
+
|
18 |
+
|
19 |
+
url = 'https://drive.google.com/u/0/uc?id=1Tdn3yqpZ79X7R1Ql0zNlNScB1Dv9Fp76&export=download'
|
20 |
+
output = 'yolor_p6.pt'
|
21 |
+
gdown.download(url, output, quiet=False)
|
22 |
+
|
23 |
+
url1 = 'https://cdn.pixabay.com/photo/2014/09/07/21/52/city-438393_1280.jpg'
|
24 |
+
r = requests.get(url1, allow_redirects=True)
|
25 |
+
open("city1.jpg", 'wb').write(r.content)
|
26 |
+
url2 = 'https://cdn.pixabay.com/photo/2016/02/19/11/36/canal-1209808_1280.jpg'
|
27 |
+
r = requests.get(url2, allow_redirects=True)
|
28 |
+
open("city2.jpg", 'wb').write(r.content)
|
29 |
+
|
30 |
+
|
31 |
+
conf_thres = 0.4
|
32 |
+
iou_thres = 0.5
|
33 |
+
|
34 |
+
|
35 |
+
def load_classes(path):
|
36 |
+
# Loads *.names file at 'path'
|
37 |
+
with open(path, 'r') as f:
|
38 |
+
names = f.read().split('\n')
|
39 |
+
return list(filter(None, names)) # filter removes empty strings (such as last line)
|
40 |
+
|
41 |
+
def detect(pil_img,names):
|
42 |
+
img_np = np.array(pil_img)
|
43 |
+
img = torch.from_numpy(img_np)
|
44 |
+
img = img.float()
|
45 |
+
img /= 255.0 # 0 - 255 to 0.0 - 1.0
|
46 |
+
|
47 |
+
# Inference
|
48 |
+
pred = model(img.unsqueeze(0).permute(0,3,1,2), augment=False)[0]
|
49 |
+
|
50 |
+
# Apply NMS
|
51 |
+
pred = non_max_suppression(pred, conf_thres, iou_thres, classes=None, agnostic=False)
|
52 |
+
|
53 |
+
# Process detections
|
54 |
+
for i, det in enumerate(pred): # detections per image
|
55 |
+
if det is not None and len(det):
|
56 |
+
# Rescale boxes from img_size to im0 size
|
57 |
+
det[:, :4] = scale_coords(img_np.shape, det[:, :4], img_np.shape).round()
|
58 |
+
|
59 |
+
# Print results
|
60 |
+
for c in det[:, -1].unique():
|
61 |
+
n = (det[:, -1] == c).sum() # detections per class
|
62 |
+
|
63 |
+
# Write results
|
64 |
+
for *xyxy, conf, cls in det:
|
65 |
+
label = '%s %.2f' % (names[int(cls)], conf)
|
66 |
+
plot_one_box(xyxy, img_np, label=label, color=colors[int(cls)], line_thickness=3)
|
67 |
+
cv2.imwrite('/tmp/aaa.jpg',img_np[:,:,::-1])
|
68 |
+
return Image.fromarray(img_np)
|
69 |
+
|
70 |
+
|
71 |
+
with torch.no_grad():
|
72 |
+
cfg = 'cfg/yolor_p6.cfg'
|
73 |
+
imgsz = 1280
|
74 |
+
names = 'data/coco.names'
|
75 |
+
weights = 'yolor_p6.pt'
|
76 |
+
|
77 |
+
# Load model
|
78 |
+
model = Darknet(cfg, imgsz)
|
79 |
+
model.load_state_dict(torch.load(weights)['model'])
|
80 |
+
model.eval()
|
81 |
+
|
82 |
+
# Get names and colors
|
83 |
+
names = load_classes(names)
|
84 |
+
colors = [[random.randint(0, 255) for _ in range(3)] for _ in range(len(names))]
|
85 |
+
|
86 |
+
def inference(image):
|
87 |
+
image = image.resize(size=(imgsz, imgsz))
|
88 |
+
return detect(image, names)
|
89 |
+
|
90 |
+
title = "YOLOR P6"
|
91 |
+
description = "demo for YOLOR. To use it, simply upload your image, or click one of the examples to load them. Read more at the links below.\nModel: YOLOR-P6"
|
92 |
+
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2105.04206'>You Only Learn One Representation: Unified Network for Multiple Tasks</a> | <a href='https://github.com/WongKinYiu/yolor'>Github Repo</a></p>"
|
93 |
+
|
94 |
+
gr.Interface(
|
95 |
+
inference,
|
96 |
+
[gr.inputs.Image(type="pil", label="Input")],
|
97 |
+
gr.outputs.Image(type="numpy", label="Output"),
|
98 |
+
title=title,
|
99 |
+
description=description,
|
100 |
+
article=article,
|
101 |
+
examples=[
|
102 |
+
["city1.jpg"],
|
103 |
+
["city2.jpg"]
|
104 |
+
]).launch()
|
cfg/yolor_csp.cfg
ADDED
@@ -0,0 +1,1376 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[net]
|
2 |
+
# Testing
|
3 |
+
#batch=1
|
4 |
+
#subdivisions=1
|
5 |
+
# Training
|
6 |
+
batch=64
|
7 |
+
subdivisions=8
|
8 |
+
width=512
|
9 |
+
height=512
|
10 |
+
channels=3
|
11 |
+
momentum=0.949
|
12 |
+
decay=0.0005
|
13 |
+
angle=0
|
14 |
+
saturation = 1.5
|
15 |
+
exposure = 1.5
|
16 |
+
hue=.1
|
17 |
+
|
18 |
+
learning_rate=0.00261
|
19 |
+
burn_in=1000
|
20 |
+
max_batches = 500500
|
21 |
+
policy=steps
|
22 |
+
steps=400000,450000
|
23 |
+
scales=.1,.1
|
24 |
+
|
25 |
+
#cutmix=1
|
26 |
+
mosaic=1
|
27 |
+
|
28 |
+
|
29 |
+
# ============ Backbone ============ #
|
30 |
+
|
31 |
+
# Stem
|
32 |
+
|
33 |
+
# 0
|
34 |
+
[convolutional]
|
35 |
+
batch_normalize=1
|
36 |
+
filters=32
|
37 |
+
size=3
|
38 |
+
stride=1
|
39 |
+
pad=1
|
40 |
+
activation=silu
|
41 |
+
|
42 |
+
# P1
|
43 |
+
|
44 |
+
# Downsample
|
45 |
+
|
46 |
+
[convolutional]
|
47 |
+
batch_normalize=1
|
48 |
+
filters=64
|
49 |
+
size=3
|
50 |
+
stride=2
|
51 |
+
pad=1
|
52 |
+
activation=silu
|
53 |
+
|
54 |
+
# Residual Block
|
55 |
+
|
56 |
+
[convolutional]
|
57 |
+
batch_normalize=1
|
58 |
+
filters=32
|
59 |
+
size=1
|
60 |
+
stride=1
|
61 |
+
pad=1
|
62 |
+
activation=silu
|
63 |
+
|
64 |
+
[convolutional]
|
65 |
+
batch_normalize=1
|
66 |
+
filters=64
|
67 |
+
size=3
|
68 |
+
stride=1
|
69 |
+
pad=1
|
70 |
+
activation=silu
|
71 |
+
|
72 |
+
# 4 (previous+1+3k)
|
73 |
+
[shortcut]
|
74 |
+
from=-3
|
75 |
+
activation=linear
|
76 |
+
|
77 |
+
# P2
|
78 |
+
|
79 |
+
# Downsample
|
80 |
+
|
81 |
+
[convolutional]
|
82 |
+
batch_normalize=1
|
83 |
+
filters=128
|
84 |
+
size=3
|
85 |
+
stride=2
|
86 |
+
pad=1
|
87 |
+
activation=silu
|
88 |
+
|
89 |
+
# Split
|
90 |
+
|
91 |
+
[convolutional]
|
92 |
+
batch_normalize=1
|
93 |
+
filters=64
|
94 |
+
size=1
|
95 |
+
stride=1
|
96 |
+
pad=1
|
97 |
+
activation=silu
|
98 |
+
|
99 |
+
[route]
|
100 |
+
layers = -2
|
101 |
+
|
102 |
+
[convolutional]
|
103 |
+
batch_normalize=1
|
104 |
+
filters=64
|
105 |
+
size=1
|
106 |
+
stride=1
|
107 |
+
pad=1
|
108 |
+
activation=silu
|
109 |
+
|
110 |
+
# Residual Block
|
111 |
+
|
112 |
+
[convolutional]
|
113 |
+
batch_normalize=1
|
114 |
+
filters=64
|
115 |
+
size=1
|
116 |
+
stride=1
|
117 |
+
pad=1
|
118 |
+
activation=silu
|
119 |
+
|
120 |
+
[convolutional]
|
121 |
+
batch_normalize=1
|
122 |
+
filters=64
|
123 |
+
size=3
|
124 |
+
stride=1
|
125 |
+
pad=1
|
126 |
+
activation=silu
|
127 |
+
|
128 |
+
[shortcut]
|
129 |
+
from=-3
|
130 |
+
activation=linear
|
131 |
+
|
132 |
+
[convolutional]
|
133 |
+
batch_normalize=1
|
134 |
+
filters=64
|
135 |
+
size=1
|
136 |
+
stride=1
|
137 |
+
pad=1
|
138 |
+
activation=silu
|
139 |
+
|
140 |
+
[convolutional]
|
141 |
+
batch_normalize=1
|
142 |
+
filters=64
|
143 |
+
size=3
|
144 |
+
stride=1
|
145 |
+
pad=1
|
146 |
+
activation=silu
|
147 |
+
|
148 |
+
[shortcut]
|
149 |
+
from=-3
|
150 |
+
activation=linear
|
151 |
+
|
152 |
+
# Transition first
|
153 |
+
|
154 |
+
[convolutional]
|
155 |
+
batch_normalize=1
|
156 |
+
filters=64
|
157 |
+
size=1
|
158 |
+
stride=1
|
159 |
+
pad=1
|
160 |
+
activation=silu
|
161 |
+
|
162 |
+
# Merge [-1, -(3k+4)]
|
163 |
+
|
164 |
+
[route]
|
165 |
+
layers = -1,-10
|
166 |
+
|
167 |
+
# Transition last
|
168 |
+
|
169 |
+
# 17 (previous+7+3k)
|
170 |
+
[convolutional]
|
171 |
+
batch_normalize=1
|
172 |
+
filters=128
|
173 |
+
size=1
|
174 |
+
stride=1
|
175 |
+
pad=1
|
176 |
+
activation=silu
|
177 |
+
|
178 |
+
# P3
|
179 |
+
|
180 |
+
# Downsample
|
181 |
+
|
182 |
+
[convolutional]
|
183 |
+
batch_normalize=1
|
184 |
+
filters=256
|
185 |
+
size=3
|
186 |
+
stride=2
|
187 |
+
pad=1
|
188 |
+
activation=silu
|
189 |
+
|
190 |
+
# Split
|
191 |
+
|
192 |
+
[convolutional]
|
193 |
+
batch_normalize=1
|
194 |
+
filters=128
|
195 |
+
size=1
|
196 |
+
stride=1
|
197 |
+
pad=1
|
198 |
+
activation=silu
|
199 |
+
|
200 |
+
[route]
|
201 |
+
layers = -2
|
202 |
+
|
203 |
+
[convolutional]
|
204 |
+
batch_normalize=1
|
205 |
+
filters=128
|
206 |
+
size=1
|
207 |
+
stride=1
|
208 |
+
pad=1
|
209 |
+
activation=silu
|
210 |
+
|
211 |
+
# Residual Block
|
212 |
+
|
213 |
+
[convolutional]
|
214 |
+
batch_normalize=1
|
215 |
+
filters=128
|
216 |
+
size=1
|
217 |
+
stride=1
|
218 |
+
pad=1
|
219 |
+
activation=silu
|
220 |
+
|
221 |
+
[convolutional]
|
222 |
+
batch_normalize=1
|
223 |
+
filters=128
|
224 |
+
size=3
|
225 |
+
stride=1
|
226 |
+
pad=1
|
227 |
+
activation=silu
|
228 |
+
|
229 |
+
[shortcut]
|
230 |
+
from=-3
|
231 |
+
activation=linear
|
232 |
+
|
233 |
+
[convolutional]
|
234 |
+
batch_normalize=1
|
235 |
+
filters=128
|
236 |
+
size=1
|
237 |
+
stride=1
|
238 |
+
pad=1
|
239 |
+
activation=silu
|
240 |
+
|
241 |
+
[convolutional]
|
242 |
+
batch_normalize=1
|
243 |
+
filters=128
|
244 |
+
size=3
|
245 |
+
stride=1
|
246 |
+
pad=1
|
247 |
+
activation=silu
|
248 |
+
|
249 |
+
[shortcut]
|
250 |
+
from=-3
|
251 |
+
activation=linear
|
252 |
+
|
253 |
+
[convolutional]
|
254 |
+
batch_normalize=1
|
255 |
+
filters=128
|
256 |
+
size=1
|
257 |
+
stride=1
|
258 |
+
pad=1
|
259 |
+
activation=silu
|
260 |
+
|
261 |
+
[convolutional]
|
262 |
+
batch_normalize=1
|
263 |
+
filters=128
|
264 |
+
size=3
|
265 |
+
stride=1
|
266 |
+
pad=1
|
267 |
+
activation=silu
|
268 |
+
|
269 |
+
[shortcut]
|
270 |
+
from=-3
|
271 |
+
activation=linear
|
272 |
+
|
273 |
+
[convolutional]
|
274 |
+
batch_normalize=1
|
275 |
+
filters=128
|
276 |
+
size=1
|
277 |
+
stride=1
|
278 |
+
pad=1
|
279 |
+
activation=silu
|
280 |
+
|
281 |
+
[convolutional]
|
282 |
+
batch_normalize=1
|
283 |
+
filters=128
|
284 |
+
size=3
|
285 |
+
stride=1
|
286 |
+
pad=1
|
287 |
+
activation=silu
|
288 |
+
|
289 |
+
[shortcut]
|
290 |
+
from=-3
|
291 |
+
activation=linear
|
292 |
+
|
293 |
+
[convolutional]
|
294 |
+
batch_normalize=1
|
295 |
+
filters=128
|
296 |
+
size=1
|
297 |
+
stride=1
|
298 |
+
pad=1
|
299 |
+
activation=silu
|
300 |
+
|
301 |
+
[convolutional]
|
302 |
+
batch_normalize=1
|
303 |
+
filters=128
|
304 |
+
size=3
|
305 |
+
stride=1
|
306 |
+
pad=1
|
307 |
+
activation=silu
|
308 |
+
|
309 |
+
[shortcut]
|
310 |
+
from=-3
|
311 |
+
activation=linear
|
312 |
+
|
313 |
+
[convolutional]
|
314 |
+
batch_normalize=1
|
315 |
+
filters=128
|
316 |
+
size=1
|
317 |
+
stride=1
|
318 |
+
pad=1
|
319 |
+
activation=silu
|
320 |
+
|
321 |
+
[convolutional]
|
322 |
+
batch_normalize=1
|
323 |
+
filters=128
|
324 |
+
size=3
|
325 |
+
stride=1
|
326 |
+
pad=1
|
327 |
+
activation=silu
|
328 |
+
|
329 |
+
[shortcut]
|
330 |
+
from=-3
|
331 |
+
activation=linear
|
332 |
+
|
333 |
+
[convolutional]
|
334 |
+
batch_normalize=1
|
335 |
+
filters=128
|
336 |
+
size=1
|
337 |
+
stride=1
|
338 |
+
pad=1
|
339 |
+
activation=silu
|
340 |
+
|
341 |
+
[convolutional]
|
342 |
+
batch_normalize=1
|
343 |
+
filters=128
|
344 |
+
size=3
|
345 |
+
stride=1
|
346 |
+
pad=1
|
347 |
+
activation=silu
|
348 |
+
|
349 |
+
[shortcut]
|
350 |
+
from=-3
|
351 |
+
activation=linear
|
352 |
+
|
353 |
+
[convolutional]
|
354 |
+
batch_normalize=1
|
355 |
+
filters=128
|
356 |
+
size=1
|
357 |
+
stride=1
|
358 |
+
pad=1
|
359 |
+
activation=silu
|
360 |
+
|
361 |
+
[convolutional]
|
362 |
+
batch_normalize=1
|
363 |
+
filters=128
|
364 |
+
size=3
|
365 |
+
stride=1
|
366 |
+
pad=1
|
367 |
+
activation=silu
|
368 |
+
|
369 |
+
[shortcut]
|
370 |
+
from=-3
|
371 |
+
activation=linear
|
372 |
+
|
373 |
+
# Transition first
|
374 |
+
|
375 |
+
[convolutional]
|
376 |
+
batch_normalize=1
|
377 |
+
filters=128
|
378 |
+
size=1
|
379 |
+
stride=1
|
380 |
+
pad=1
|
381 |
+
activation=silu
|
382 |
+
|
383 |
+
# Merge [-1 -(4+3k)]
|
384 |
+
|
385 |
+
[route]
|
386 |
+
layers = -1,-28
|
387 |
+
|
388 |
+
# Transition last
|
389 |
+
|
390 |
+
# 48 (previous+7+3k)
|
391 |
+
[convolutional]
|
392 |
+
batch_normalize=1
|
393 |
+
filters=256
|
394 |
+
size=1
|
395 |
+
stride=1
|
396 |
+
pad=1
|
397 |
+
activation=silu
|
398 |
+
|
399 |
+
# P4
|
400 |
+
|
401 |
+
# Downsample
|
402 |
+
|
403 |
+
[convolutional]
|
404 |
+
batch_normalize=1
|
405 |
+
filters=512
|
406 |
+
size=3
|
407 |
+
stride=2
|
408 |
+
pad=1
|
409 |
+
activation=silu
|
410 |
+
|
411 |
+
# Split
|
412 |
+
|
413 |
+
[convolutional]
|
414 |
+
batch_normalize=1
|
415 |
+
filters=256
|
416 |
+
size=1
|
417 |
+
stride=1
|
418 |
+
pad=1
|
419 |
+
activation=silu
|
420 |
+
|
421 |
+
[route]
|
422 |
+
layers = -2
|
423 |
+
|
424 |
+
[convolutional]
|
425 |
+
batch_normalize=1
|
426 |
+
filters=256
|
427 |
+
size=1
|
428 |
+
stride=1
|
429 |
+
pad=1
|
430 |
+
activation=silu
|
431 |
+
|
432 |
+
# Residual Block
|
433 |
+
|
434 |
+
[convolutional]
|
435 |
+
batch_normalize=1
|
436 |
+
filters=256
|
437 |
+
size=1
|
438 |
+
stride=1
|
439 |
+
pad=1
|
440 |
+
activation=silu
|
441 |
+
|
442 |
+
[convolutional]
|
443 |
+
batch_normalize=1
|
444 |
+
filters=256
|
445 |
+
size=3
|
446 |
+
stride=1
|
447 |
+
pad=1
|
448 |
+
activation=silu
|
449 |
+
|
450 |
+
[shortcut]
|
451 |
+
from=-3
|
452 |
+
activation=linear
|
453 |
+
|
454 |
+
[convolutional]
|
455 |
+
batch_normalize=1
|
456 |
+
filters=256
|
457 |
+
size=1
|
458 |
+
stride=1
|
459 |
+
pad=1
|
460 |
+
activation=silu
|
461 |
+
|
462 |
+
[convolutional]
|
463 |
+
batch_normalize=1
|
464 |
+
filters=256
|
465 |
+
size=3
|
466 |
+
stride=1
|
467 |
+
pad=1
|
468 |
+
activation=silu
|
469 |
+
|
470 |
+
[shortcut]
|
471 |
+
from=-3
|
472 |
+
activation=linear
|
473 |
+
|
474 |
+
[convolutional]
|
475 |
+
batch_normalize=1
|
476 |
+
filters=256
|
477 |
+
size=1
|
478 |
+
stride=1
|
479 |
+
pad=1
|
480 |
+
activation=silu
|
481 |
+
|
482 |
+
[convolutional]
|
483 |
+
batch_normalize=1
|
484 |
+
filters=256
|
485 |
+
size=3
|
486 |
+
stride=1
|
487 |
+
pad=1
|
488 |
+
activation=silu
|
489 |
+
|
490 |
+
[shortcut]
|
491 |
+
from=-3
|
492 |
+
activation=linear
|
493 |
+
|
494 |
+
[convolutional]
|
495 |
+
batch_normalize=1
|
496 |
+
filters=256
|
497 |
+
size=1
|
498 |
+
stride=1
|
499 |
+
pad=1
|
500 |
+
activation=silu
|
501 |
+
|
502 |
+
[convolutional]
|
503 |
+
batch_normalize=1
|
504 |
+
filters=256
|
505 |
+
size=3
|
506 |
+
stride=1
|
507 |
+
pad=1
|
508 |
+
activation=silu
|
509 |
+
|
510 |
+
[shortcut]
|
511 |
+
from=-3
|
512 |
+
activation=linear
|
513 |
+
|
514 |
+
[convolutional]
|
515 |
+
batch_normalize=1
|
516 |
+
filters=256
|
517 |
+
size=1
|
518 |
+
stride=1
|
519 |
+
pad=1
|
520 |
+
activation=silu
|
521 |
+
|
522 |
+
[convolutional]
|
523 |
+
batch_normalize=1
|
524 |
+
filters=256
|
525 |
+
size=3
|
526 |
+
stride=1
|
527 |
+
pad=1
|
528 |
+
activation=silu
|
529 |
+
|
530 |
+
[shortcut]
|
531 |
+
from=-3
|
532 |
+
activation=linear
|
533 |
+
|
534 |
+
[convolutional]
|
535 |
+
batch_normalize=1
|
536 |
+
filters=256
|
537 |
+
size=1
|
538 |
+
stride=1
|
539 |
+
pad=1
|
540 |
+
activation=silu
|
541 |
+
|
542 |
+
[convolutional]
|
543 |
+
batch_normalize=1
|
544 |
+
filters=256
|
545 |
+
size=3
|
546 |
+
stride=1
|
547 |
+
pad=1
|
548 |
+
activation=silu
|
549 |
+
|
550 |
+
[shortcut]
|
551 |
+
from=-3
|
552 |
+
activation=linear
|
553 |
+
|
554 |
+
[convolutional]
|
555 |
+
batch_normalize=1
|
556 |
+
filters=256
|
557 |
+
size=1
|
558 |
+
stride=1
|
559 |
+
pad=1
|
560 |
+
activation=silu
|
561 |
+
|
562 |
+
[convolutional]
|
563 |
+
batch_normalize=1
|
564 |
+
filters=256
|
565 |
+
size=3
|
566 |
+
stride=1
|
567 |
+
pad=1
|
568 |
+
activation=silu
|
569 |
+
|
570 |
+
[shortcut]
|
571 |
+
from=-3
|
572 |
+
activation=linear
|
573 |
+
|
574 |
+
[convolutional]
|
575 |
+
batch_normalize=1
|
576 |
+
filters=256
|
577 |
+
size=1
|
578 |
+
stride=1
|
579 |
+
pad=1
|
580 |
+
activation=silu
|
581 |
+
|
582 |
+
[convolutional]
|
583 |
+
batch_normalize=1
|
584 |
+
filters=256
|
585 |
+
size=3
|
586 |
+
stride=1
|
587 |
+
pad=1
|
588 |
+
activation=silu
|
589 |
+
|
590 |
+
[shortcut]
|
591 |
+
from=-3
|
592 |
+
activation=linear
|
593 |
+
|
594 |
+
# Transition first
|
595 |
+
|
596 |
+
[convolutional]
|
597 |
+
batch_normalize=1
|
598 |
+
filters=256
|
599 |
+
size=1
|
600 |
+
stride=1
|
601 |
+
pad=1
|
602 |
+
activation=silu
|
603 |
+
|
604 |
+
# Merge [-1 -(3k+4)]
|
605 |
+
|
606 |
+
[route]
|
607 |
+
layers = -1,-28
|
608 |
+
|
609 |
+
# Transition last
|
610 |
+
|
611 |
+
# 79 (previous+7+3k)
|
612 |
+
[convolutional]
|
613 |
+
batch_normalize=1
|
614 |
+
filters=512
|
615 |
+
size=1
|
616 |
+
stride=1
|
617 |
+
pad=1
|
618 |
+
activation=silu
|
619 |
+
|
620 |
+
# P5
|
621 |
+
|
622 |
+
# Downsample
|
623 |
+
|
624 |
+
[convolutional]
|
625 |
+
batch_normalize=1
|
626 |
+
filters=1024
|
627 |
+
size=3
|
628 |
+
stride=2
|
629 |
+
pad=1
|
630 |
+
activation=silu
|
631 |
+
|
632 |
+
# Split
|
633 |
+
|
634 |
+
[convolutional]
|
635 |
+
batch_normalize=1
|
636 |
+
filters=512
|
637 |
+
size=1
|
638 |
+
stride=1
|
639 |
+
pad=1
|
640 |
+
activation=silu
|
641 |
+
|
642 |
+
[route]
|
643 |
+
layers = -2
|
644 |
+
|
645 |
+
[convolutional]
|
646 |
+
batch_normalize=1
|
647 |
+
filters=512
|
648 |
+
size=1
|
649 |
+
stride=1
|
650 |
+
pad=1
|
651 |
+
activation=silu
|
652 |
+
|
653 |
+
# Residual Block
|
654 |
+
|
655 |
+
[convolutional]
|
656 |
+
batch_normalize=1
|
657 |
+
filters=512
|
658 |
+
size=1
|
659 |
+
stride=1
|
660 |
+
pad=1
|
661 |
+
activation=silu
|
662 |
+
|
663 |
+
[convolutional]
|
664 |
+
batch_normalize=1
|
665 |
+
filters=512
|
666 |
+
size=3
|
667 |
+
stride=1
|
668 |
+
pad=1
|
669 |
+
activation=silu
|
670 |
+
|
671 |
+
[shortcut]
|
672 |
+
from=-3
|
673 |
+
activation=linear
|
674 |
+
|
675 |
+
[convolutional]
|
676 |
+
batch_normalize=1
|
677 |
+
filters=512
|
678 |
+
size=1
|
679 |
+
stride=1
|
680 |
+
pad=1
|
681 |
+
activation=silu
|
682 |
+
|
683 |
+
[convolutional]
|
684 |
+
batch_normalize=1
|
685 |
+
filters=512
|
686 |
+
size=3
|
687 |
+
stride=1
|
688 |
+
pad=1
|
689 |
+
activation=silu
|
690 |
+
|
691 |
+
[shortcut]
|
692 |
+
from=-3
|
693 |
+
activation=linear
|
694 |
+
|
695 |
+
[convolutional]
|
696 |
+
batch_normalize=1
|
697 |
+
filters=512
|
698 |
+
size=1
|
699 |
+
stride=1
|
700 |
+
pad=1
|
701 |
+
activation=silu
|
702 |
+
|
703 |
+
[convolutional]
|
704 |
+
batch_normalize=1
|
705 |
+
filters=512
|
706 |
+
size=3
|
707 |
+
stride=1
|
708 |
+
pad=1
|
709 |
+
activation=silu
|
710 |
+
|
711 |
+
[shortcut]
|
712 |
+
from=-3
|
713 |
+
activation=linear
|
714 |
+
|
715 |
+
[convolutional]
|
716 |
+
batch_normalize=1
|
717 |
+
filters=512
|
718 |
+
size=1
|
719 |
+
stride=1
|
720 |
+
pad=1
|
721 |
+
activation=silu
|
722 |
+
|
723 |
+
[convolutional]
|
724 |
+
batch_normalize=1
|
725 |
+
filters=512
|
726 |
+
size=3
|
727 |
+
stride=1
|
728 |
+
pad=1
|
729 |
+
activation=silu
|
730 |
+
|
731 |
+
[shortcut]
|
732 |
+
from=-3
|
733 |
+
activation=linear
|
734 |
+
|
735 |
+
# Transition first
|
736 |
+
|
737 |
+
[convolutional]
|
738 |
+
batch_normalize=1
|
739 |
+
filters=512
|
740 |
+
size=1
|
741 |
+
stride=1
|
742 |
+
pad=1
|
743 |
+
activation=silu
|
744 |
+
|
745 |
+
# Merge [-1 -(3k+4)]
|
746 |
+
|
747 |
+
[route]
|
748 |
+
layers = -1,-16
|
749 |
+
|
750 |
+
# Transition last
|
751 |
+
|
752 |
+
# 98 (previous+7+3k)
|
753 |
+
[convolutional]
|
754 |
+
batch_normalize=1
|
755 |
+
filters=1024
|
756 |
+
size=1
|
757 |
+
stride=1
|
758 |
+
pad=1
|
759 |
+
activation=silu
|
760 |
+
|
761 |
+
# ============ End of Backbone ============ #
|
762 |
+
|
763 |
+
# ============ Neck ============ #
|
764 |
+
|
765 |
+
# CSPSPP
|
766 |
+
|
767 |
+
[convolutional]
|
768 |
+
batch_normalize=1
|
769 |
+
filters=512
|
770 |
+
size=1
|
771 |
+
stride=1
|
772 |
+
pad=1
|
773 |
+
activation=silu
|
774 |
+
|
775 |
+
[route]
|
776 |
+
layers = -2
|
777 |
+
|
778 |
+
[convolutional]
|
779 |
+
batch_normalize=1
|
780 |
+
filters=512
|
781 |
+
size=1
|
782 |
+
stride=1
|
783 |
+
pad=1
|
784 |
+
activation=silu
|
785 |
+
|
786 |
+
[convolutional]
|
787 |
+
batch_normalize=1
|
788 |
+
size=3
|
789 |
+
stride=1
|
790 |
+
pad=1
|
791 |
+
filters=512
|
792 |
+
activation=silu
|
793 |
+
|
794 |
+
[convolutional]
|
795 |
+
batch_normalize=1
|
796 |
+
filters=512
|
797 |
+
size=1
|
798 |
+
stride=1
|
799 |
+
pad=1
|
800 |
+
activation=silu
|
801 |
+
|
802 |
+
### SPP ###
|
803 |
+
[maxpool]
|
804 |
+
stride=1
|
805 |
+
size=5
|
806 |
+
|
807 |
+
[route]
|
808 |
+
layers=-2
|
809 |
+
|
810 |
+
[maxpool]
|
811 |
+
stride=1
|
812 |
+
size=9
|
813 |
+
|
814 |
+
[route]
|
815 |
+
layers=-4
|
816 |
+
|
817 |
+
[maxpool]
|
818 |
+
stride=1
|
819 |
+
size=13
|
820 |
+
|
821 |
+
[route]
|
822 |
+
layers=-1,-3,-5,-6
|
823 |
+
### End SPP ###
|
824 |
+
|
825 |
+
[convolutional]
|
826 |
+
batch_normalize=1
|
827 |
+
filters=512
|
828 |
+
size=1
|
829 |
+
stride=1
|
830 |
+
pad=1
|
831 |
+
activation=silu
|
832 |
+
|
833 |
+
[convolutional]
|
834 |
+
batch_normalize=1
|
835 |
+
size=3
|
836 |
+
stride=1
|
837 |
+
pad=1
|
838 |
+
filters=512
|
839 |
+
activation=silu
|
840 |
+
|
841 |
+
[route]
|
842 |
+
layers = -1, -13
|
843 |
+
|
844 |
+
# 113 (previous+6+5+2k)
|
845 |
+
[convolutional]
|
846 |
+
batch_normalize=1
|
847 |
+
filters=512
|
848 |
+
size=1
|
849 |
+
stride=1
|
850 |
+
pad=1
|
851 |
+
activation=silu
|
852 |
+
|
853 |
+
# End of CSPSPP
|
854 |
+
|
855 |
+
|
856 |
+
# FPN-4
|
857 |
+
|
858 |
+
[convolutional]
|
859 |
+
batch_normalize=1
|
860 |
+
filters=256
|
861 |
+
size=1
|
862 |
+
stride=1
|
863 |
+
pad=1
|
864 |
+
activation=silu
|
865 |
+
|
866 |
+
[upsample]
|
867 |
+
stride=2
|
868 |
+
|
869 |
+
[route]
|
870 |
+
layers = 79
|
871 |
+
|
872 |
+
[convolutional]
|
873 |
+
batch_normalize=1
|
874 |
+
filters=256
|
875 |
+
size=1
|
876 |
+
stride=1
|
877 |
+
pad=1
|
878 |
+
activation=silu
|
879 |
+
|
880 |
+
[route]
|
881 |
+
layers = -1, -3
|
882 |
+
|
883 |
+
[convolutional]
|
884 |
+
batch_normalize=1
|
885 |
+
filters=256
|
886 |
+
size=1
|
887 |
+
stride=1
|
888 |
+
pad=1
|
889 |
+
activation=silu
|
890 |
+
|
891 |
+
# Split
|
892 |
+
|
893 |
+
[convolutional]
|
894 |
+
batch_normalize=1
|
895 |
+
filters=256
|
896 |
+
size=1
|
897 |
+
stride=1
|
898 |
+
pad=1
|
899 |
+
activation=silu
|
900 |
+
|
901 |
+
[route]
|
902 |
+
layers = -2
|
903 |
+
|
904 |
+
# Plain Block
|
905 |
+
|
906 |
+
[convolutional]
|
907 |
+
batch_normalize=1
|
908 |
+
filters=256
|
909 |
+
size=1
|
910 |
+
stride=1
|
911 |
+
pad=1
|
912 |
+
activation=silu
|
913 |
+
|
914 |
+
[convolutional]
|
915 |
+
batch_normalize=1
|
916 |
+
size=3
|
917 |
+
stride=1
|
918 |
+
pad=1
|
919 |
+
filters=256
|
920 |
+
activation=silu
|
921 |
+
|
922 |
+
[convolutional]
|
923 |
+
batch_normalize=1
|
924 |
+
filters=256
|
925 |
+
size=1
|
926 |
+
stride=1
|
927 |
+
pad=1
|
928 |
+
activation=silu
|
929 |
+
|
930 |
+
[convolutional]
|
931 |
+
batch_normalize=1
|
932 |
+
size=3
|
933 |
+
stride=1
|
934 |
+
pad=1
|
935 |
+
filters=256
|
936 |
+
activation=silu
|
937 |
+
|
938 |
+
# Merge [-1, -(2k+2)]
|
939 |
+
|
940 |
+
[route]
|
941 |
+
layers = -1, -6
|
942 |
+
|
943 |
+
# Transition last
|
944 |
+
|
945 |
+
# 127 (previous+6+4+2k)
|
946 |
+
[convolutional]
|
947 |
+
batch_normalize=1
|
948 |
+
filters=256
|
949 |
+
size=1
|
950 |
+
stride=1
|
951 |
+
pad=1
|
952 |
+
activation=silu
|
953 |
+
|
954 |
+
|
955 |
+
# FPN-3
|
956 |
+
|
957 |
+
[convolutional]
|
958 |
+
batch_normalize=1
|
959 |
+
filters=128
|
960 |
+
size=1
|
961 |
+
stride=1
|
962 |
+
pad=1
|
963 |
+
activation=silu
|
964 |
+
|
965 |
+
[upsample]
|
966 |
+
stride=2
|
967 |
+
|
968 |
+
[route]
|
969 |
+
layers = 48
|
970 |
+
|
971 |
+
[convolutional]
|
972 |
+
batch_normalize=1
|
973 |
+
filters=128
|
974 |
+
size=1
|
975 |
+
stride=1
|
976 |
+
pad=1
|
977 |
+
activation=silu
|
978 |
+
|
979 |
+
[route]
|
980 |
+
layers = -1, -3
|
981 |
+
|
982 |
+
[convolutional]
|
983 |
+
batch_normalize=1
|
984 |
+
filters=128
|
985 |
+
size=1
|
986 |
+
stride=1
|
987 |
+
pad=1
|
988 |
+
activation=silu
|
989 |
+
|
990 |
+
# Split
|
991 |
+
|
992 |
+
[convolutional]
|
993 |
+
batch_normalize=1
|
994 |
+
filters=128
|
995 |
+
size=1
|
996 |
+
stride=1
|
997 |
+
pad=1
|
998 |
+
activation=silu
|
999 |
+
|
1000 |
+
[route]
|
1001 |
+
layers = -2
|
1002 |
+
|
1003 |
+
# Plain Block
|
1004 |
+
|
1005 |
+
[convolutional]
|
1006 |
+
batch_normalize=1
|
1007 |
+
filters=128
|
1008 |
+
size=1
|
1009 |
+
stride=1
|
1010 |
+
pad=1
|
1011 |
+
activation=silu
|
1012 |
+
|
1013 |
+
[convolutional]
|
1014 |
+
batch_normalize=1
|
1015 |
+
size=3
|
1016 |
+
stride=1
|
1017 |
+
pad=1
|
1018 |
+
filters=128
|
1019 |
+
activation=silu
|
1020 |
+
|
1021 |
+
[convolutional]
|
1022 |
+
batch_normalize=1
|
1023 |
+
filters=128
|
1024 |
+
size=1
|
1025 |
+
stride=1
|
1026 |
+
pad=1
|
1027 |
+
activation=silu
|
1028 |
+
|
1029 |
+
[convolutional]
|
1030 |
+
batch_normalize=1
|
1031 |
+
size=3
|
1032 |
+
stride=1
|
1033 |
+
pad=1
|
1034 |
+
filters=128
|
1035 |
+
activation=silu
|
1036 |
+
|
1037 |
+
# Merge [-1, -(2k+2)]
|
1038 |
+
|
1039 |
+
[route]
|
1040 |
+
layers = -1, -6
|
1041 |
+
|
1042 |
+
# Transition last
|
1043 |
+
|
1044 |
+
# 141 (previous+6+4+2k)
|
1045 |
+
[convolutional]
|
1046 |
+
batch_normalize=1
|
1047 |
+
filters=128
|
1048 |
+
size=1
|
1049 |
+
stride=1
|
1050 |
+
pad=1
|
1051 |
+
activation=silu
|
1052 |
+
|
1053 |
+
|
1054 |
+
# PAN-4
|
1055 |
+
|
1056 |
+
[convolutional]
|
1057 |
+
batch_normalize=1
|
1058 |
+
size=3
|
1059 |
+
stride=2
|
1060 |
+
pad=1
|
1061 |
+
filters=256
|
1062 |
+
activation=silu
|
1063 |
+
|
1064 |
+
[route]
|
1065 |
+
layers = -1, 127
|
1066 |
+
|
1067 |
+
[convolutional]
|
1068 |
+
batch_normalize=1
|
1069 |
+
filters=256
|
1070 |
+
size=1
|
1071 |
+
stride=1
|
1072 |
+
pad=1
|
1073 |
+
activation=silu
|
1074 |
+
|
1075 |
+
# Split
|
1076 |
+
|
1077 |
+
[convolutional]
|
1078 |
+
batch_normalize=1
|
1079 |
+
filters=256
|
1080 |
+
size=1
|
1081 |
+
stride=1
|
1082 |
+
pad=1
|
1083 |
+
activation=silu
|
1084 |
+
|
1085 |
+
[route]
|
1086 |
+
layers = -2
|
1087 |
+
|
1088 |
+
# Plain Block
|
1089 |
+
|
1090 |
+
[convolutional]
|
1091 |
+
batch_normalize=1
|
1092 |
+
filters=256
|
1093 |
+
size=1
|
1094 |
+
stride=1
|
1095 |
+
pad=1
|
1096 |
+
activation=silu
|
1097 |
+
|
1098 |
+
[convolutional]
|
1099 |
+
batch_normalize=1
|
1100 |
+
size=3
|
1101 |
+
stride=1
|
1102 |
+
pad=1
|
1103 |
+
filters=256
|
1104 |
+
activation=silu
|
1105 |
+
|
1106 |
+
[convolutional]
|
1107 |
+
batch_normalize=1
|
1108 |
+
filters=256
|
1109 |
+
size=1
|
1110 |
+
stride=1
|
1111 |
+
pad=1
|
1112 |
+
activation=silu
|
1113 |
+
|
1114 |
+
[convolutional]
|
1115 |
+
batch_normalize=1
|
1116 |
+
size=3
|
1117 |
+
stride=1
|
1118 |
+
pad=1
|
1119 |
+
filters=256
|
1120 |
+
activation=silu
|
1121 |
+
|
1122 |
+
[route]
|
1123 |
+
layers = -1,-6
|
1124 |
+
|
1125 |
+
# Transition last
|
1126 |
+
|
1127 |
+
# 152 (previous+3+4+2k)
|
1128 |
+
[convolutional]
|
1129 |
+
batch_normalize=1
|
1130 |
+
filters=256
|
1131 |
+
size=1
|
1132 |
+
stride=1
|
1133 |
+
pad=1
|
1134 |
+
activation=silu
|
1135 |
+
|
1136 |
+
|
1137 |
+
# PAN-5
|
1138 |
+
|
1139 |
+
[convolutional]
|
1140 |
+
batch_normalize=1
|
1141 |
+
size=3
|
1142 |
+
stride=2
|
1143 |
+
pad=1
|
1144 |
+
filters=512
|
1145 |
+
activation=silu
|
1146 |
+
|
1147 |
+
[route]
|
1148 |
+
layers = -1, 113
|
1149 |
+
|
1150 |
+
[convolutional]
|
1151 |
+
batch_normalize=1
|
1152 |
+
filters=512
|
1153 |
+
size=1
|
1154 |
+
stride=1
|
1155 |
+
pad=1
|
1156 |
+
activation=silu
|
1157 |
+
|
1158 |
+
# Split
|
1159 |
+
|
1160 |
+
[convolutional]
|
1161 |
+
batch_normalize=1
|
1162 |
+
filters=512
|
1163 |
+
size=1
|
1164 |
+
stride=1
|
1165 |
+
pad=1
|
1166 |
+
activation=silu
|
1167 |
+
|
1168 |
+
[route]
|
1169 |
+
layers = -2
|
1170 |
+
|
1171 |
+
# Plain Block
|
1172 |
+
|
1173 |
+
[convolutional]
|
1174 |
+
batch_normalize=1
|
1175 |
+
filters=512
|
1176 |
+
size=1
|
1177 |
+
stride=1
|
1178 |
+
pad=1
|
1179 |
+
activation=silu
|
1180 |
+
|
1181 |
+
[convolutional]
|
1182 |
+
batch_normalize=1
|
1183 |
+
size=3
|
1184 |
+
stride=1
|
1185 |
+
pad=1
|
1186 |
+
filters=512
|
1187 |
+
activation=silu
|
1188 |
+
|
1189 |
+
[convolutional]
|
1190 |
+
batch_normalize=1
|
1191 |
+
filters=512
|
1192 |
+
size=1
|
1193 |
+
stride=1
|
1194 |
+
pad=1
|
1195 |
+
activation=silu
|
1196 |
+
|
1197 |
+
[convolutional]
|
1198 |
+
batch_normalize=1
|
1199 |
+
size=3
|
1200 |
+
stride=1
|
1201 |
+
pad=1
|
1202 |
+
filters=512
|
1203 |
+
activation=silu
|
1204 |
+
|
1205 |
+
[route]
|
1206 |
+
layers = -1,-6
|
1207 |
+
|
1208 |
+
# Transition last
|
1209 |
+
|
1210 |
+
# 163 (previous+3+4+2k)
|
1211 |
+
[convolutional]
|
1212 |
+
batch_normalize=1
|
1213 |
+
filters=512
|
1214 |
+
size=1
|
1215 |
+
stride=1
|
1216 |
+
pad=1
|
1217 |
+
activation=silu
|
1218 |
+
|
1219 |
+
# ============ End of Neck ============ #
|
1220 |
+
|
1221 |
+
# 164
|
1222 |
+
[implicit_add]
|
1223 |
+
filters=256
|
1224 |
+
|
1225 |
+
# 165
|
1226 |
+
[implicit_add]
|
1227 |
+
filters=512
|
1228 |
+
|
1229 |
+
# 166
|
1230 |
+
[implicit_add]
|
1231 |
+
filters=1024
|
1232 |
+
|
1233 |
+
# 167
|
1234 |
+
[implicit_mul]
|
1235 |
+
filters=255
|
1236 |
+
|
1237 |
+
# 168
|
1238 |
+
[implicit_mul]
|
1239 |
+
filters=255
|
1240 |
+
|
1241 |
+
# 169
|
1242 |
+
[implicit_mul]
|
1243 |
+
filters=255
|
1244 |
+
|
1245 |
+
# ============ Head ============ #
|
1246 |
+
|
1247 |
+
# YOLO-3
|
1248 |
+
|
1249 |
+
[route]
|
1250 |
+
layers = 141
|
1251 |
+
|
1252 |
+
[convolutional]
|
1253 |
+
batch_normalize=1
|
1254 |
+
size=3
|
1255 |
+
stride=1
|
1256 |
+
pad=1
|
1257 |
+
filters=256
|
1258 |
+
activation=silu
|
1259 |
+
|
1260 |
+
[shift_channels]
|
1261 |
+
from=164
|
1262 |
+
|
1263 |
+
[convolutional]
|
1264 |
+
size=1
|
1265 |
+
stride=1
|
1266 |
+
pad=1
|
1267 |
+
filters=255
|
1268 |
+
activation=linear
|
1269 |
+
|
1270 |
+
[control_channels]
|
1271 |
+
from=167
|
1272 |
+
|
1273 |
+
[yolo]
|
1274 |
+
mask = 0,1,2
|
1275 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1276 |
+
classes=80
|
1277 |
+
num=9
|
1278 |
+
jitter=.3
|
1279 |
+
ignore_thresh = .7
|
1280 |
+
truth_thresh = 1
|
1281 |
+
random=1
|
1282 |
+
scale_x_y = 1.05
|
1283 |
+
iou_thresh=0.213
|
1284 |
+
cls_normalizer=1.0
|
1285 |
+
iou_normalizer=0.07
|
1286 |
+
iou_loss=ciou
|
1287 |
+
nms_kind=greedynms
|
1288 |
+
beta_nms=0.6
|
1289 |
+
|
1290 |
+
|
1291 |
+
# YOLO-4
|
1292 |
+
|
1293 |
+
[route]
|
1294 |
+
layers = 152
|
1295 |
+
|
1296 |
+
[convolutional]
|
1297 |
+
batch_normalize=1
|
1298 |
+
size=3
|
1299 |
+
stride=1
|
1300 |
+
pad=1
|
1301 |
+
filters=512
|
1302 |
+
activation=silu
|
1303 |
+
|
1304 |
+
[shift_channels]
|
1305 |
+
from=165
|
1306 |
+
|
1307 |
+
[convolutional]
|
1308 |
+
size=1
|
1309 |
+
stride=1
|
1310 |
+
pad=1
|
1311 |
+
filters=255
|
1312 |
+
activation=linear
|
1313 |
+
|
1314 |
+
[control_channels]
|
1315 |
+
from=168
|
1316 |
+
|
1317 |
+
[yolo]
|
1318 |
+
mask = 3,4,5
|
1319 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1320 |
+
classes=80
|
1321 |
+
num=9
|
1322 |
+
jitter=.3
|
1323 |
+
ignore_thresh = .7
|
1324 |
+
truth_thresh = 1
|
1325 |
+
random=1
|
1326 |
+
scale_x_y = 1.05
|
1327 |
+
iou_thresh=0.213
|
1328 |
+
cls_normalizer=1.0
|
1329 |
+
iou_normalizer=0.07
|
1330 |
+
iou_loss=ciou
|
1331 |
+
nms_kind=greedynms
|
1332 |
+
beta_nms=0.6
|
1333 |
+
|
1334 |
+
|
1335 |
+
# YOLO-5
|
1336 |
+
|
1337 |
+
[route]
|
1338 |
+
layers = 163
|
1339 |
+
|
1340 |
+
[convolutional]
|
1341 |
+
batch_normalize=1
|
1342 |
+
size=3
|
1343 |
+
stride=1
|
1344 |
+
pad=1
|
1345 |
+
filters=1024
|
1346 |
+
activation=silu
|
1347 |
+
|
1348 |
+
[shift_channels]
|
1349 |
+
from=166
|
1350 |
+
|
1351 |
+
[convolutional]
|
1352 |
+
size=1
|
1353 |
+
stride=1
|
1354 |
+
pad=1
|
1355 |
+
filters=255
|
1356 |
+
activation=linear
|
1357 |
+
|
1358 |
+
[control_channels]
|
1359 |
+
from=169
|
1360 |
+
|
1361 |
+
[yolo]
|
1362 |
+
mask = 6,7,8
|
1363 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1364 |
+
classes=80
|
1365 |
+
num=9
|
1366 |
+
jitter=.3
|
1367 |
+
ignore_thresh = .7
|
1368 |
+
truth_thresh = 1
|
1369 |
+
random=1
|
1370 |
+
scale_x_y = 1.05
|
1371 |
+
iou_thresh=0.213
|
1372 |
+
cls_normalizer=1.0
|
1373 |
+
iou_normalizer=0.07
|
1374 |
+
iou_loss=ciou
|
1375 |
+
nms_kind=greedynms
|
1376 |
+
beta_nms=0.6
|
cfg/yolor_csp_x.cfg
ADDED
@@ -0,0 +1,1576 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[net]
|
2 |
+
# Testing
|
3 |
+
#batch=1
|
4 |
+
#subdivisions=1
|
5 |
+
# Training
|
6 |
+
batch=64
|
7 |
+
subdivisions=8
|
8 |
+
width=512
|
9 |
+
height=512
|
10 |
+
channels=3
|
11 |
+
momentum=0.949
|
12 |
+
decay=0.0005
|
13 |
+
angle=0
|
14 |
+
saturation = 1.5
|
15 |
+
exposure = 1.5
|
16 |
+
hue=.1
|
17 |
+
|
18 |
+
learning_rate=0.00261
|
19 |
+
burn_in=1000
|
20 |
+
max_batches = 500500
|
21 |
+
policy=steps
|
22 |
+
steps=400000,450000
|
23 |
+
scales=.1,.1
|
24 |
+
|
25 |
+
#cutmix=1
|
26 |
+
mosaic=1
|
27 |
+
|
28 |
+
|
29 |
+
# ============ Backbone ============ #
|
30 |
+
|
31 |
+
# Stem
|
32 |
+
|
33 |
+
# 0
|
34 |
+
[convolutional]
|
35 |
+
batch_normalize=1
|
36 |
+
filters=32
|
37 |
+
size=3
|
38 |
+
stride=1
|
39 |
+
pad=1
|
40 |
+
activation=silu
|
41 |
+
|
42 |
+
# P1
|
43 |
+
|
44 |
+
# Downsample
|
45 |
+
|
46 |
+
[convolutional]
|
47 |
+
batch_normalize=1
|
48 |
+
filters=80
|
49 |
+
size=3
|
50 |
+
stride=2
|
51 |
+
pad=1
|
52 |
+
activation=silu
|
53 |
+
|
54 |
+
# Residual Block
|
55 |
+
|
56 |
+
[convolutional]
|
57 |
+
batch_normalize=1
|
58 |
+
filters=40
|
59 |
+
size=1
|
60 |
+
stride=1
|
61 |
+
pad=1
|
62 |
+
activation=silu
|
63 |
+
|
64 |
+
[convolutional]
|
65 |
+
batch_normalize=1
|
66 |
+
filters=80
|
67 |
+
size=3
|
68 |
+
stride=1
|
69 |
+
pad=1
|
70 |
+
activation=silu
|
71 |
+
|
72 |
+
# 4 (previous+1+3k)
|
73 |
+
[shortcut]
|
74 |
+
from=-3
|
75 |
+
activation=linear
|
76 |
+
|
77 |
+
# P2
|
78 |
+
|
79 |
+
# Downsample
|
80 |
+
|
81 |
+
[convolutional]
|
82 |
+
batch_normalize=1
|
83 |
+
filters=160
|
84 |
+
size=3
|
85 |
+
stride=2
|
86 |
+
pad=1
|
87 |
+
activation=silu
|
88 |
+
|
89 |
+
# Split
|
90 |
+
|
91 |
+
[convolutional]
|
92 |
+
batch_normalize=1
|
93 |
+
filters=80
|
94 |
+
size=1
|
95 |
+
stride=1
|
96 |
+
pad=1
|
97 |
+
activation=silu
|
98 |
+
|
99 |
+
[route]
|
100 |
+
layers = -2
|
101 |
+
|
102 |
+
[convolutional]
|
103 |
+
batch_normalize=1
|
104 |
+
filters=80
|
105 |
+
size=1
|
106 |
+
stride=1
|
107 |
+
pad=1
|
108 |
+
activation=silu
|
109 |
+
|
110 |
+
# Residual Block
|
111 |
+
|
112 |
+
[convolutional]
|
113 |
+
batch_normalize=1
|
114 |
+
filters=80
|
115 |
+
size=1
|
116 |
+
stride=1
|
117 |
+
pad=1
|
118 |
+
activation=silu
|
119 |
+
|
120 |
+
[convolutional]
|
121 |
+
batch_normalize=1
|
122 |
+
filters=80
|
123 |
+
size=3
|
124 |
+
stride=1
|
125 |
+
pad=1
|
126 |
+
activation=silu
|
127 |
+
|
128 |
+
[shortcut]
|
129 |
+
from=-3
|
130 |
+
activation=linear
|
131 |
+
|
132 |
+
[convolutional]
|
133 |
+
batch_normalize=1
|
134 |
+
filters=80
|
135 |
+
size=1
|
136 |
+
stride=1
|
137 |
+
pad=1
|
138 |
+
activation=silu
|
139 |
+
|
140 |
+
[convolutional]
|
141 |
+
batch_normalize=1
|
142 |
+
filters=80
|
143 |
+
size=3
|
144 |
+
stride=1
|
145 |
+
pad=1
|
146 |
+
activation=silu
|
147 |
+
|
148 |
+
[shortcut]
|
149 |
+
from=-3
|
150 |
+
activation=linear
|
151 |
+
|
152 |
+
[convolutional]
|
153 |
+
batch_normalize=1
|
154 |
+
filters=80
|
155 |
+
size=1
|
156 |
+
stride=1
|
157 |
+
pad=1
|
158 |
+
activation=silu
|
159 |
+
|
160 |
+
[convolutional]
|
161 |
+
batch_normalize=1
|
162 |
+
filters=80
|
163 |
+
size=3
|
164 |
+
stride=1
|
165 |
+
pad=1
|
166 |
+
activation=silu
|
167 |
+
|
168 |
+
[shortcut]
|
169 |
+
from=-3
|
170 |
+
activation=linear
|
171 |
+
|
172 |
+
# Transition first
|
173 |
+
|
174 |
+
[convolutional]
|
175 |
+
batch_normalize=1
|
176 |
+
filters=80
|
177 |
+
size=1
|
178 |
+
stride=1
|
179 |
+
pad=1
|
180 |
+
activation=silu
|
181 |
+
|
182 |
+
# Merge [-1, -(3k+4)]
|
183 |
+
|
184 |
+
[route]
|
185 |
+
layers = -1,-13
|
186 |
+
|
187 |
+
# Transition last
|
188 |
+
|
189 |
+
# 20 (previous+7+3k)
|
190 |
+
[convolutional]
|
191 |
+
batch_normalize=1
|
192 |
+
filters=160
|
193 |
+
size=1
|
194 |
+
stride=1
|
195 |
+
pad=1
|
196 |
+
activation=silu
|
197 |
+
|
198 |
+
# P3
|
199 |
+
|
200 |
+
# Downsample
|
201 |
+
|
202 |
+
[convolutional]
|
203 |
+
batch_normalize=1
|
204 |
+
filters=320
|
205 |
+
size=3
|
206 |
+
stride=2
|
207 |
+
pad=1
|
208 |
+
activation=silu
|
209 |
+
|
210 |
+
# Split
|
211 |
+
|
212 |
+
[convolutional]
|
213 |
+
batch_normalize=1
|
214 |
+
filters=160
|
215 |
+
size=1
|
216 |
+
stride=1
|
217 |
+
pad=1
|
218 |
+
activation=silu
|
219 |
+
|
220 |
+
[route]
|
221 |
+
layers = -2
|
222 |
+
|
223 |
+
[convolutional]
|
224 |
+
batch_normalize=1
|
225 |
+
filters=160
|
226 |
+
size=1
|
227 |
+
stride=1
|
228 |
+
pad=1
|
229 |
+
activation=silu
|
230 |
+
|
231 |
+
# Residual Block
|
232 |
+
|
233 |
+
[convolutional]
|
234 |
+
batch_normalize=1
|
235 |
+
filters=160
|
236 |
+
size=1
|
237 |
+
stride=1
|
238 |
+
pad=1
|
239 |
+
activation=silu
|
240 |
+
|
241 |
+
[convolutional]
|
242 |
+
batch_normalize=1
|
243 |
+
filters=160
|
244 |
+
size=3
|
245 |
+
stride=1
|
246 |
+
pad=1
|
247 |
+
activation=silu
|
248 |
+
|
249 |
+
[shortcut]
|
250 |
+
from=-3
|
251 |
+
activation=linear
|
252 |
+
|
253 |
+
[convolutional]
|
254 |
+
batch_normalize=1
|
255 |
+
filters=160
|
256 |
+
size=1
|
257 |
+
stride=1
|
258 |
+
pad=1
|
259 |
+
activation=silu
|
260 |
+
|
261 |
+
[convolutional]
|
262 |
+
batch_normalize=1
|
263 |
+
filters=160
|
264 |
+
size=3
|
265 |
+
stride=1
|
266 |
+
pad=1
|
267 |
+
activation=silu
|
268 |
+
|
269 |
+
[shortcut]
|
270 |
+
from=-3
|
271 |
+
activation=linear
|
272 |
+
|
273 |
+
[convolutional]
|
274 |
+
batch_normalize=1
|
275 |
+
filters=160
|
276 |
+
size=1
|
277 |
+
stride=1
|
278 |
+
pad=1
|
279 |
+
activation=silu
|
280 |
+
|
281 |
+
[convolutional]
|
282 |
+
batch_normalize=1
|
283 |
+
filters=160
|
284 |
+
size=3
|
285 |
+
stride=1
|
286 |
+
pad=1
|
287 |
+
activation=silu
|
288 |
+
|
289 |
+
[shortcut]
|
290 |
+
from=-3
|
291 |
+
activation=linear
|
292 |
+
|
293 |
+
[convolutional]
|
294 |
+
batch_normalize=1
|
295 |
+
filters=160
|
296 |
+
size=1
|
297 |
+
stride=1
|
298 |
+
pad=1
|
299 |
+
activation=silu
|
300 |
+
|
301 |
+
[convolutional]
|
302 |
+
batch_normalize=1
|
303 |
+
filters=160
|
304 |
+
size=3
|
305 |
+
stride=1
|
306 |
+
pad=1
|
307 |
+
activation=silu
|
308 |
+
|
309 |
+
[shortcut]
|
310 |
+
from=-3
|
311 |
+
activation=linear
|
312 |
+
|
313 |
+
[convolutional]
|
314 |
+
batch_normalize=1
|
315 |
+
filters=160
|
316 |
+
size=1
|
317 |
+
stride=1
|
318 |
+
pad=1
|
319 |
+
activation=silu
|
320 |
+
|
321 |
+
[convolutional]
|
322 |
+
batch_normalize=1
|
323 |
+
filters=160
|
324 |
+
size=3
|
325 |
+
stride=1
|
326 |
+
pad=1
|
327 |
+
activation=silu
|
328 |
+
|
329 |
+
[shortcut]
|
330 |
+
from=-3
|
331 |
+
activation=linear
|
332 |
+
|
333 |
+
[convolutional]
|
334 |
+
batch_normalize=1
|
335 |
+
filters=160
|
336 |
+
size=1
|
337 |
+
stride=1
|
338 |
+
pad=1
|
339 |
+
activation=silu
|
340 |
+
|
341 |
+
[convolutional]
|
342 |
+
batch_normalize=1
|
343 |
+
filters=160
|
344 |
+
size=3
|
345 |
+
stride=1
|
346 |
+
pad=1
|
347 |
+
activation=silu
|
348 |
+
|
349 |
+
[shortcut]
|
350 |
+
from=-3
|
351 |
+
activation=linear
|
352 |
+
|
353 |
+
[convolutional]
|
354 |
+
batch_normalize=1
|
355 |
+
filters=160
|
356 |
+
size=1
|
357 |
+
stride=1
|
358 |
+
pad=1
|
359 |
+
activation=silu
|
360 |
+
|
361 |
+
[convolutional]
|
362 |
+
batch_normalize=1
|
363 |
+
filters=160
|
364 |
+
size=3
|
365 |
+
stride=1
|
366 |
+
pad=1
|
367 |
+
activation=silu
|
368 |
+
|
369 |
+
[shortcut]
|
370 |
+
from=-3
|
371 |
+
activation=linear
|
372 |
+
|
373 |
+
[convolutional]
|
374 |
+
batch_normalize=1
|
375 |
+
filters=160
|
376 |
+
size=1
|
377 |
+
stride=1
|
378 |
+
pad=1
|
379 |
+
activation=silu
|
380 |
+
|
381 |
+
[convolutional]
|
382 |
+
batch_normalize=1
|
383 |
+
filters=160
|
384 |
+
size=3
|
385 |
+
stride=1
|
386 |
+
pad=1
|
387 |
+
activation=silu
|
388 |
+
|
389 |
+
[shortcut]
|
390 |
+
from=-3
|
391 |
+
activation=linear
|
392 |
+
|
393 |
+
[convolutional]
|
394 |
+
batch_normalize=1
|
395 |
+
filters=160
|
396 |
+
size=1
|
397 |
+
stride=1
|
398 |
+
pad=1
|
399 |
+
activation=silu
|
400 |
+
|
401 |
+
[convolutional]
|
402 |
+
batch_normalize=1
|
403 |
+
filters=160
|
404 |
+
size=3
|
405 |
+
stride=1
|
406 |
+
pad=1
|
407 |
+
activation=silu
|
408 |
+
|
409 |
+
[shortcut]
|
410 |
+
from=-3
|
411 |
+
activation=linear
|
412 |
+
|
413 |
+
[convolutional]
|
414 |
+
batch_normalize=1
|
415 |
+
filters=160
|
416 |
+
size=1
|
417 |
+
stride=1
|
418 |
+
pad=1
|
419 |
+
activation=silu
|
420 |
+
|
421 |
+
[convolutional]
|
422 |
+
batch_normalize=1
|
423 |
+
filters=160
|
424 |
+
size=3
|
425 |
+
stride=1
|
426 |
+
pad=1
|
427 |
+
activation=silu
|
428 |
+
|
429 |
+
[shortcut]
|
430 |
+
from=-3
|
431 |
+
activation=linear
|
432 |
+
|
433 |
+
# Transition first
|
434 |
+
|
435 |
+
[convolutional]
|
436 |
+
batch_normalize=1
|
437 |
+
filters=160
|
438 |
+
size=1
|
439 |
+
stride=1
|
440 |
+
pad=1
|
441 |
+
activation=silu
|
442 |
+
|
443 |
+
# Merge [-1 -(4+3k)]
|
444 |
+
|
445 |
+
[route]
|
446 |
+
layers = -1,-34
|
447 |
+
|
448 |
+
# Transition last
|
449 |
+
|
450 |
+
# 57 (previous+7+3k)
|
451 |
+
[convolutional]
|
452 |
+
batch_normalize=1
|
453 |
+
filters=320
|
454 |
+
size=1
|
455 |
+
stride=1
|
456 |
+
pad=1
|
457 |
+
activation=silu
|
458 |
+
|
459 |
+
# P4
|
460 |
+
|
461 |
+
# Downsample
|
462 |
+
|
463 |
+
[convolutional]
|
464 |
+
batch_normalize=1
|
465 |
+
filters=640
|
466 |
+
size=3
|
467 |
+
stride=2
|
468 |
+
pad=1
|
469 |
+
activation=silu
|
470 |
+
|
471 |
+
# Split
|
472 |
+
|
473 |
+
[convolutional]
|
474 |
+
batch_normalize=1
|
475 |
+
filters=320
|
476 |
+
size=1
|
477 |
+
stride=1
|
478 |
+
pad=1
|
479 |
+
activation=silu
|
480 |
+
|
481 |
+
[route]
|
482 |
+
layers = -2
|
483 |
+
|
484 |
+
[convolutional]
|
485 |
+
batch_normalize=1
|
486 |
+
filters=320
|
487 |
+
size=1
|
488 |
+
stride=1
|
489 |
+
pad=1
|
490 |
+
activation=silu
|
491 |
+
|
492 |
+
# Residual Block
|
493 |
+
|
494 |
+
[convolutional]
|
495 |
+
batch_normalize=1
|
496 |
+
filters=320
|
497 |
+
size=1
|
498 |
+
stride=1
|
499 |
+
pad=1
|
500 |
+
activation=silu
|
501 |
+
|
502 |
+
[convolutional]
|
503 |
+
batch_normalize=1
|
504 |
+
filters=320
|
505 |
+
size=3
|
506 |
+
stride=1
|
507 |
+
pad=1
|
508 |
+
activation=silu
|
509 |
+
|
510 |
+
[shortcut]
|
511 |
+
from=-3
|
512 |
+
activation=linear
|
513 |
+
|
514 |
+
[convolutional]
|
515 |
+
batch_normalize=1
|
516 |
+
filters=320
|
517 |
+
size=1
|
518 |
+
stride=1
|
519 |
+
pad=1
|
520 |
+
activation=silu
|
521 |
+
|
522 |
+
[convolutional]
|
523 |
+
batch_normalize=1
|
524 |
+
filters=320
|
525 |
+
size=3
|
526 |
+
stride=1
|
527 |
+
pad=1
|
528 |
+
activation=silu
|
529 |
+
|
530 |
+
[shortcut]
|
531 |
+
from=-3
|
532 |
+
activation=linear
|
533 |
+
|
534 |
+
[convolutional]
|
535 |
+
batch_normalize=1
|
536 |
+
filters=320
|
537 |
+
size=1
|
538 |
+
stride=1
|
539 |
+
pad=1
|
540 |
+
activation=silu
|
541 |
+
|
542 |
+
[convolutional]
|
543 |
+
batch_normalize=1
|
544 |
+
filters=320
|
545 |
+
size=3
|
546 |
+
stride=1
|
547 |
+
pad=1
|
548 |
+
activation=silu
|
549 |
+
|
550 |
+
[shortcut]
|
551 |
+
from=-3
|
552 |
+
activation=linear
|
553 |
+
|
554 |
+
[convolutional]
|
555 |
+
batch_normalize=1
|
556 |
+
filters=320
|
557 |
+
size=1
|
558 |
+
stride=1
|
559 |
+
pad=1
|
560 |
+
activation=silu
|
561 |
+
|
562 |
+
[convolutional]
|
563 |
+
batch_normalize=1
|
564 |
+
filters=320
|
565 |
+
size=3
|
566 |
+
stride=1
|
567 |
+
pad=1
|
568 |
+
activation=silu
|
569 |
+
|
570 |
+
[shortcut]
|
571 |
+
from=-3
|
572 |
+
activation=linear
|
573 |
+
|
574 |
+
[convolutional]
|
575 |
+
batch_normalize=1
|
576 |
+
filters=320
|
577 |
+
size=1
|
578 |
+
stride=1
|
579 |
+
pad=1
|
580 |
+
activation=silu
|
581 |
+
|
582 |
+
[convolutional]
|
583 |
+
batch_normalize=1
|
584 |
+
filters=320
|
585 |
+
size=3
|
586 |
+
stride=1
|
587 |
+
pad=1
|
588 |
+
activation=silu
|
589 |
+
|
590 |
+
[shortcut]
|
591 |
+
from=-3
|
592 |
+
activation=linear
|
593 |
+
|
594 |
+
[convolutional]
|
595 |
+
batch_normalize=1
|
596 |
+
filters=320
|
597 |
+
size=1
|
598 |
+
stride=1
|
599 |
+
pad=1
|
600 |
+
activation=silu
|
601 |
+
|
602 |
+
[convolutional]
|
603 |
+
batch_normalize=1
|
604 |
+
filters=320
|
605 |
+
size=3
|
606 |
+
stride=1
|
607 |
+
pad=1
|
608 |
+
activation=silu
|
609 |
+
|
610 |
+
[shortcut]
|
611 |
+
from=-3
|
612 |
+
activation=linear
|
613 |
+
|
614 |
+
[convolutional]
|
615 |
+
batch_normalize=1
|
616 |
+
filters=320
|
617 |
+
size=1
|
618 |
+
stride=1
|
619 |
+
pad=1
|
620 |
+
activation=silu
|
621 |
+
|
622 |
+
[convolutional]
|
623 |
+
batch_normalize=1
|
624 |
+
filters=320
|
625 |
+
size=3
|
626 |
+
stride=1
|
627 |
+
pad=1
|
628 |
+
activation=silu
|
629 |
+
|
630 |
+
[shortcut]
|
631 |
+
from=-3
|
632 |
+
activation=linear
|
633 |
+
|
634 |
+
[convolutional]
|
635 |
+
batch_normalize=1
|
636 |
+
filters=320
|
637 |
+
size=1
|
638 |
+
stride=1
|
639 |
+
pad=1
|
640 |
+
activation=silu
|
641 |
+
|
642 |
+
[convolutional]
|
643 |
+
batch_normalize=1
|
644 |
+
filters=320
|
645 |
+
size=3
|
646 |
+
stride=1
|
647 |
+
pad=1
|
648 |
+
activation=silu
|
649 |
+
|
650 |
+
[shortcut]
|
651 |
+
from=-3
|
652 |
+
activation=linear
|
653 |
+
|
654 |
+
[convolutional]
|
655 |
+
batch_normalize=1
|
656 |
+
filters=320
|
657 |
+
size=1
|
658 |
+
stride=1
|
659 |
+
pad=1
|
660 |
+
activation=silu
|
661 |
+
|
662 |
+
[convolutional]
|
663 |
+
batch_normalize=1
|
664 |
+
filters=320
|
665 |
+
size=3
|
666 |
+
stride=1
|
667 |
+
pad=1
|
668 |
+
activation=silu
|
669 |
+
|
670 |
+
[shortcut]
|
671 |
+
from=-3
|
672 |
+
activation=linear
|
673 |
+
|
674 |
+
[convolutional]
|
675 |
+
batch_normalize=1
|
676 |
+
filters=320
|
677 |
+
size=1
|
678 |
+
stride=1
|
679 |
+
pad=1
|
680 |
+
activation=silu
|
681 |
+
|
682 |
+
[convolutional]
|
683 |
+
batch_normalize=1
|
684 |
+
filters=320
|
685 |
+
size=3
|
686 |
+
stride=1
|
687 |
+
pad=1
|
688 |
+
activation=silu
|
689 |
+
|
690 |
+
[shortcut]
|
691 |
+
from=-3
|
692 |
+
activation=linear
|
693 |
+
|
694 |
+
# Transition first
|
695 |
+
|
696 |
+
[convolutional]
|
697 |
+
batch_normalize=1
|
698 |
+
filters=320
|
699 |
+
size=1
|
700 |
+
stride=1
|
701 |
+
pad=1
|
702 |
+
activation=silu
|
703 |
+
|
704 |
+
# Merge [-1 -(3k+4)]
|
705 |
+
|
706 |
+
[route]
|
707 |
+
layers = -1,-34
|
708 |
+
|
709 |
+
# Transition last
|
710 |
+
|
711 |
+
# 94 (previous+7+3k)
|
712 |
+
[convolutional]
|
713 |
+
batch_normalize=1
|
714 |
+
filters=640
|
715 |
+
size=1
|
716 |
+
stride=1
|
717 |
+
pad=1
|
718 |
+
activation=silu
|
719 |
+
|
720 |
+
# P5
|
721 |
+
|
722 |
+
# Downsample
|
723 |
+
|
724 |
+
[convolutional]
|
725 |
+
batch_normalize=1
|
726 |
+
filters=1280
|
727 |
+
size=3
|
728 |
+
stride=2
|
729 |
+
pad=1
|
730 |
+
activation=silu
|
731 |
+
|
732 |
+
# Split
|
733 |
+
|
734 |
+
[convolutional]
|
735 |
+
batch_normalize=1
|
736 |
+
filters=640
|
737 |
+
size=1
|
738 |
+
stride=1
|
739 |
+
pad=1
|
740 |
+
activation=silu
|
741 |
+
|
742 |
+
[route]
|
743 |
+
layers = -2
|
744 |
+
|
745 |
+
[convolutional]
|
746 |
+
batch_normalize=1
|
747 |
+
filters=640
|
748 |
+
size=1
|
749 |
+
stride=1
|
750 |
+
pad=1
|
751 |
+
activation=silu
|
752 |
+
|
753 |
+
# Residual Block
|
754 |
+
|
755 |
+
[convolutional]
|
756 |
+
batch_normalize=1
|
757 |
+
filters=640
|
758 |
+
size=1
|
759 |
+
stride=1
|
760 |
+
pad=1
|
761 |
+
activation=silu
|
762 |
+
|
763 |
+
[convolutional]
|
764 |
+
batch_normalize=1
|
765 |
+
filters=640
|
766 |
+
size=3
|
767 |
+
stride=1
|
768 |
+
pad=1
|
769 |
+
activation=silu
|
770 |
+
|
771 |
+
[shortcut]
|
772 |
+
from=-3
|
773 |
+
activation=linear
|
774 |
+
|
775 |
+
[convolutional]
|
776 |
+
batch_normalize=1
|
777 |
+
filters=640
|
778 |
+
size=1
|
779 |
+
stride=1
|
780 |
+
pad=1
|
781 |
+
activation=silu
|
782 |
+
|
783 |
+
[convolutional]
|
784 |
+
batch_normalize=1
|
785 |
+
filters=640
|
786 |
+
size=3
|
787 |
+
stride=1
|
788 |
+
pad=1
|
789 |
+
activation=silu
|
790 |
+
|
791 |
+
[shortcut]
|
792 |
+
from=-3
|
793 |
+
activation=linear
|
794 |
+
|
795 |
+
[convolutional]
|
796 |
+
batch_normalize=1
|
797 |
+
filters=640
|
798 |
+
size=1
|
799 |
+
stride=1
|
800 |
+
pad=1
|
801 |
+
activation=silu
|
802 |
+
|
803 |
+
[convolutional]
|
804 |
+
batch_normalize=1
|
805 |
+
filters=640
|
806 |
+
size=3
|
807 |
+
stride=1
|
808 |
+
pad=1
|
809 |
+
activation=silu
|
810 |
+
|
811 |
+
[shortcut]
|
812 |
+
from=-3
|
813 |
+
activation=linear
|
814 |
+
|
815 |
+
[convolutional]
|
816 |
+
batch_normalize=1
|
817 |
+
filters=640
|
818 |
+
size=1
|
819 |
+
stride=1
|
820 |
+
pad=1
|
821 |
+
activation=silu
|
822 |
+
|
823 |
+
[convolutional]
|
824 |
+
batch_normalize=1
|
825 |
+
filters=640
|
826 |
+
size=3
|
827 |
+
stride=1
|
828 |
+
pad=1
|
829 |
+
activation=silu
|
830 |
+
|
831 |
+
[shortcut]
|
832 |
+
from=-3
|
833 |
+
activation=linear
|
834 |
+
|
835 |
+
[convolutional]
|
836 |
+
batch_normalize=1
|
837 |
+
filters=640
|
838 |
+
size=1
|
839 |
+
stride=1
|
840 |
+
pad=1
|
841 |
+
activation=silu
|
842 |
+
|
843 |
+
[convolutional]
|
844 |
+
batch_normalize=1
|
845 |
+
filters=640
|
846 |
+
size=3
|
847 |
+
stride=1
|
848 |
+
pad=1
|
849 |
+
activation=silu
|
850 |
+
|
851 |
+
[shortcut]
|
852 |
+
from=-3
|
853 |
+
activation=linear
|
854 |
+
|
855 |
+
# Transition first
|
856 |
+
|
857 |
+
[convolutional]
|
858 |
+
batch_normalize=1
|
859 |
+
filters=640
|
860 |
+
size=1
|
861 |
+
stride=1
|
862 |
+
pad=1
|
863 |
+
activation=silu
|
864 |
+
|
865 |
+
# Merge [-1 -(3k+4)]
|
866 |
+
|
867 |
+
[route]
|
868 |
+
layers = -1,-19
|
869 |
+
|
870 |
+
# Transition last
|
871 |
+
|
872 |
+
# 116 (previous+7+3k)
|
873 |
+
[convolutional]
|
874 |
+
batch_normalize=1
|
875 |
+
filters=1280
|
876 |
+
size=1
|
877 |
+
stride=1
|
878 |
+
pad=1
|
879 |
+
activation=silu
|
880 |
+
|
881 |
+
# ============ End of Backbone ============ #
|
882 |
+
|
883 |
+
# ============ Neck ============ #
|
884 |
+
|
885 |
+
# CSPSPP
|
886 |
+
|
887 |
+
[convolutional]
|
888 |
+
batch_normalize=1
|
889 |
+
filters=640
|
890 |
+
size=1
|
891 |
+
stride=1
|
892 |
+
pad=1
|
893 |
+
activation=silu
|
894 |
+
|
895 |
+
[route]
|
896 |
+
layers = -2
|
897 |
+
|
898 |
+
[convolutional]
|
899 |
+
batch_normalize=1
|
900 |
+
filters=640
|
901 |
+
size=1
|
902 |
+
stride=1
|
903 |
+
pad=1
|
904 |
+
activation=silu
|
905 |
+
|
906 |
+
[convolutional]
|
907 |
+
batch_normalize=1
|
908 |
+
size=3
|
909 |
+
stride=1
|
910 |
+
pad=1
|
911 |
+
filters=640
|
912 |
+
activation=silu
|
913 |
+
|
914 |
+
[convolutional]
|
915 |
+
batch_normalize=1
|
916 |
+
filters=640
|
917 |
+
size=1
|
918 |
+
stride=1
|
919 |
+
pad=1
|
920 |
+
activation=silu
|
921 |
+
|
922 |
+
### SPP ###
|
923 |
+
[maxpool]
|
924 |
+
stride=1
|
925 |
+
size=5
|
926 |
+
|
927 |
+
[route]
|
928 |
+
layers=-2
|
929 |
+
|
930 |
+
[maxpool]
|
931 |
+
stride=1
|
932 |
+
size=9
|
933 |
+
|
934 |
+
[route]
|
935 |
+
layers=-4
|
936 |
+
|
937 |
+
[maxpool]
|
938 |
+
stride=1
|
939 |
+
size=13
|
940 |
+
|
941 |
+
[route]
|
942 |
+
layers=-1,-3,-5,-6
|
943 |
+
### End SPP ###
|
944 |
+
|
945 |
+
[convolutional]
|
946 |
+
batch_normalize=1
|
947 |
+
filters=640
|
948 |
+
size=1
|
949 |
+
stride=1
|
950 |
+
pad=1
|
951 |
+
activation=silu
|
952 |
+
|
953 |
+
[convolutional]
|
954 |
+
batch_normalize=1
|
955 |
+
size=3
|
956 |
+
stride=1
|
957 |
+
pad=1
|
958 |
+
filters=640
|
959 |
+
activation=silu
|
960 |
+
|
961 |
+
[convolutional]
|
962 |
+
batch_normalize=1
|
963 |
+
filters=640
|
964 |
+
size=1
|
965 |
+
stride=1
|
966 |
+
pad=1
|
967 |
+
activation=silu
|
968 |
+
|
969 |
+
[convolutional]
|
970 |
+
batch_normalize=1
|
971 |
+
size=3
|
972 |
+
stride=1
|
973 |
+
pad=1
|
974 |
+
filters=640
|
975 |
+
activation=silu
|
976 |
+
|
977 |
+
[route]
|
978 |
+
layers = -1, -15
|
979 |
+
|
980 |
+
# 133 (previous+6+5+2k)
|
981 |
+
[convolutional]
|
982 |
+
batch_normalize=1
|
983 |
+
filters=640
|
984 |
+
size=1
|
985 |
+
stride=1
|
986 |
+
pad=1
|
987 |
+
activation=silu
|
988 |
+
|
989 |
+
# End of CSPSPP
|
990 |
+
|
991 |
+
|
992 |
+
# FPN-4
|
993 |
+
|
994 |
+
[convolutional]
|
995 |
+
batch_normalize=1
|
996 |
+
filters=320
|
997 |
+
size=1
|
998 |
+
stride=1
|
999 |
+
pad=1
|
1000 |
+
activation=silu
|
1001 |
+
|
1002 |
+
[upsample]
|
1003 |
+
stride=2
|
1004 |
+
|
1005 |
+
[route]
|
1006 |
+
layers = 94
|
1007 |
+
|
1008 |
+
[convolutional]
|
1009 |
+
batch_normalize=1
|
1010 |
+
filters=320
|
1011 |
+
size=1
|
1012 |
+
stride=1
|
1013 |
+
pad=1
|
1014 |
+
activation=silu
|
1015 |
+
|
1016 |
+
[route]
|
1017 |
+
layers = -1, -3
|
1018 |
+
|
1019 |
+
[convolutional]
|
1020 |
+
batch_normalize=1
|
1021 |
+
filters=320
|
1022 |
+
size=1
|
1023 |
+
stride=1
|
1024 |
+
pad=1
|
1025 |
+
activation=silu
|
1026 |
+
|
1027 |
+
# Split
|
1028 |
+
|
1029 |
+
[convolutional]
|
1030 |
+
batch_normalize=1
|
1031 |
+
filters=320
|
1032 |
+
size=1
|
1033 |
+
stride=1
|
1034 |
+
pad=1
|
1035 |
+
activation=silu
|
1036 |
+
|
1037 |
+
[route]
|
1038 |
+
layers = -2
|
1039 |
+
|
1040 |
+
# Plain Block
|
1041 |
+
|
1042 |
+
[convolutional]
|
1043 |
+
batch_normalize=1
|
1044 |
+
filters=320
|
1045 |
+
size=1
|
1046 |
+
stride=1
|
1047 |
+
pad=1
|
1048 |
+
activation=silu
|
1049 |
+
|
1050 |
+
[convolutional]
|
1051 |
+
batch_normalize=1
|
1052 |
+
size=3
|
1053 |
+
stride=1
|
1054 |
+
pad=1
|
1055 |
+
filters=320
|
1056 |
+
activation=silu
|
1057 |
+
|
1058 |
+
[convolutional]
|
1059 |
+
batch_normalize=1
|
1060 |
+
filters=320
|
1061 |
+
size=1
|
1062 |
+
stride=1
|
1063 |
+
pad=1
|
1064 |
+
activation=silu
|
1065 |
+
|
1066 |
+
[convolutional]
|
1067 |
+
batch_normalize=1
|
1068 |
+
size=3
|
1069 |
+
stride=1
|
1070 |
+
pad=1
|
1071 |
+
filters=320
|
1072 |
+
activation=silu
|
1073 |
+
|
1074 |
+
[convolutional]
|
1075 |
+
batch_normalize=1
|
1076 |
+
filters=320
|
1077 |
+
size=1
|
1078 |
+
stride=1
|
1079 |
+
pad=1
|
1080 |
+
activation=silu
|
1081 |
+
|
1082 |
+
[convolutional]
|
1083 |
+
batch_normalize=1
|
1084 |
+
size=3
|
1085 |
+
stride=1
|
1086 |
+
pad=1
|
1087 |
+
filters=320
|
1088 |
+
activation=silu
|
1089 |
+
|
1090 |
+
# Merge [-1, -(2k+2)]
|
1091 |
+
|
1092 |
+
[route]
|
1093 |
+
layers = -1, -8
|
1094 |
+
|
1095 |
+
# Transition last
|
1096 |
+
|
1097 |
+
# 149 (previous+6+4+2k)
|
1098 |
+
[convolutional]
|
1099 |
+
batch_normalize=1
|
1100 |
+
filters=320
|
1101 |
+
size=1
|
1102 |
+
stride=1
|
1103 |
+
pad=1
|
1104 |
+
activation=silu
|
1105 |
+
|
1106 |
+
|
1107 |
+
# FPN-3
|
1108 |
+
|
1109 |
+
[convolutional]
|
1110 |
+
batch_normalize=1
|
1111 |
+
filters=160
|
1112 |
+
size=1
|
1113 |
+
stride=1
|
1114 |
+
pad=1
|
1115 |
+
activation=silu
|
1116 |
+
|
1117 |
+
[upsample]
|
1118 |
+
stride=2
|
1119 |
+
|
1120 |
+
[route]
|
1121 |
+
layers = 57
|
1122 |
+
|
1123 |
+
[convolutional]
|
1124 |
+
batch_normalize=1
|
1125 |
+
filters=160
|
1126 |
+
size=1
|
1127 |
+
stride=1
|
1128 |
+
pad=1
|
1129 |
+
activation=silu
|
1130 |
+
|
1131 |
+
[route]
|
1132 |
+
layers = -1, -3
|
1133 |
+
|
1134 |
+
[convolutional]
|
1135 |
+
batch_normalize=1
|
1136 |
+
filters=160
|
1137 |
+
size=1
|
1138 |
+
stride=1
|
1139 |
+
pad=1
|
1140 |
+
activation=silu
|
1141 |
+
|
1142 |
+
# Split
|
1143 |
+
|
1144 |
+
[convolutional]
|
1145 |
+
batch_normalize=1
|
1146 |
+
filters=160
|
1147 |
+
size=1
|
1148 |
+
stride=1
|
1149 |
+
pad=1
|
1150 |
+
activation=silu
|
1151 |
+
|
1152 |
+
[route]
|
1153 |
+
layers = -2
|
1154 |
+
|
1155 |
+
# Plain Block
|
1156 |
+
|
1157 |
+
[convolutional]
|
1158 |
+
batch_normalize=1
|
1159 |
+
filters=160
|
1160 |
+
size=1
|
1161 |
+
stride=1
|
1162 |
+
pad=1
|
1163 |
+
activation=silu
|
1164 |
+
|
1165 |
+
[convolutional]
|
1166 |
+
batch_normalize=1
|
1167 |
+
size=3
|
1168 |
+
stride=1
|
1169 |
+
pad=1
|
1170 |
+
filters=160
|
1171 |
+
activation=silu
|
1172 |
+
|
1173 |
+
[convolutional]
|
1174 |
+
batch_normalize=1
|
1175 |
+
filters=160
|
1176 |
+
size=1
|
1177 |
+
stride=1
|
1178 |
+
pad=1
|
1179 |
+
activation=silu
|
1180 |
+
|
1181 |
+
[convolutional]
|
1182 |
+
batch_normalize=1
|
1183 |
+
size=3
|
1184 |
+
stride=1
|
1185 |
+
pad=1
|
1186 |
+
filters=160
|
1187 |
+
activation=silu
|
1188 |
+
|
1189 |
+
[convolutional]
|
1190 |
+
batch_normalize=1
|
1191 |
+
filters=160
|
1192 |
+
size=1
|
1193 |
+
stride=1
|
1194 |
+
pad=1
|
1195 |
+
activation=silu
|
1196 |
+
|
1197 |
+
[convolutional]
|
1198 |
+
batch_normalize=1
|
1199 |
+
size=3
|
1200 |
+
stride=1
|
1201 |
+
pad=1
|
1202 |
+
filters=160
|
1203 |
+
activation=silu
|
1204 |
+
|
1205 |
+
# Merge [-1, -(2k+2)]
|
1206 |
+
|
1207 |
+
[route]
|
1208 |
+
layers = -1, -8
|
1209 |
+
|
1210 |
+
# Transition last
|
1211 |
+
|
1212 |
+
# 165 (previous+6+4+2k)
|
1213 |
+
[convolutional]
|
1214 |
+
batch_normalize=1
|
1215 |
+
filters=160
|
1216 |
+
size=1
|
1217 |
+
stride=1
|
1218 |
+
pad=1
|
1219 |
+
activation=silu
|
1220 |
+
|
1221 |
+
|
1222 |
+
# PAN-4
|
1223 |
+
|
1224 |
+
[convolutional]
|
1225 |
+
batch_normalize=1
|
1226 |
+
size=3
|
1227 |
+
stride=2
|
1228 |
+
pad=1
|
1229 |
+
filters=320
|
1230 |
+
activation=silu
|
1231 |
+
|
1232 |
+
[route]
|
1233 |
+
layers = -1, 149
|
1234 |
+
|
1235 |
+
[convolutional]
|
1236 |
+
batch_normalize=1
|
1237 |
+
filters=320
|
1238 |
+
size=1
|
1239 |
+
stride=1
|
1240 |
+
pad=1
|
1241 |
+
activation=silu
|
1242 |
+
|
1243 |
+
# Split
|
1244 |
+
|
1245 |
+
[convolutional]
|
1246 |
+
batch_normalize=1
|
1247 |
+
filters=320
|
1248 |
+
size=1
|
1249 |
+
stride=1
|
1250 |
+
pad=1
|
1251 |
+
activation=silu
|
1252 |
+
|
1253 |
+
[route]
|
1254 |
+
layers = -2
|
1255 |
+
|
1256 |
+
# Plain Block
|
1257 |
+
|
1258 |
+
[convolutional]
|
1259 |
+
batch_normalize=1
|
1260 |
+
filters=320
|
1261 |
+
size=1
|
1262 |
+
stride=1
|
1263 |
+
pad=1
|
1264 |
+
activation=silu
|
1265 |
+
|
1266 |
+
[convolutional]
|
1267 |
+
batch_normalize=1
|
1268 |
+
size=3
|
1269 |
+
stride=1
|
1270 |
+
pad=1
|
1271 |
+
filters=320
|
1272 |
+
activation=silu
|
1273 |
+
|
1274 |
+
[convolutional]
|
1275 |
+
batch_normalize=1
|
1276 |
+
filters=320
|
1277 |
+
size=1
|
1278 |
+
stride=1
|
1279 |
+
pad=1
|
1280 |
+
activation=silu
|
1281 |
+
|
1282 |
+
[convolutional]
|
1283 |
+
batch_normalize=1
|
1284 |
+
size=3
|
1285 |
+
stride=1
|
1286 |
+
pad=1
|
1287 |
+
filters=320
|
1288 |
+
activation=silu
|
1289 |
+
|
1290 |
+
[convolutional]
|
1291 |
+
batch_normalize=1
|
1292 |
+
filters=320
|
1293 |
+
size=1
|
1294 |
+
stride=1
|
1295 |
+
pad=1
|
1296 |
+
activation=silu
|
1297 |
+
|
1298 |
+
[convolutional]
|
1299 |
+
batch_normalize=1
|
1300 |
+
size=3
|
1301 |
+
stride=1
|
1302 |
+
pad=1
|
1303 |
+
filters=320
|
1304 |
+
activation=silu
|
1305 |
+
|
1306 |
+
[route]
|
1307 |
+
layers = -1,-8
|
1308 |
+
|
1309 |
+
# Transition last
|
1310 |
+
|
1311 |
+
# 178 (previous+3+4+2k)
|
1312 |
+
[convolutional]
|
1313 |
+
batch_normalize=1
|
1314 |
+
filters=320
|
1315 |
+
size=1
|
1316 |
+
stride=1
|
1317 |
+
pad=1
|
1318 |
+
activation=silu
|
1319 |
+
|
1320 |
+
|
1321 |
+
# PAN-5
|
1322 |
+
|
1323 |
+
[convolutional]
|
1324 |
+
batch_normalize=1
|
1325 |
+
size=3
|
1326 |
+
stride=2
|
1327 |
+
pad=1
|
1328 |
+
filters=640
|
1329 |
+
activation=silu
|
1330 |
+
|
1331 |
+
[route]
|
1332 |
+
layers = -1, 133
|
1333 |
+
|
1334 |
+
[convolutional]
|
1335 |
+
batch_normalize=1
|
1336 |
+
filters=640
|
1337 |
+
size=1
|
1338 |
+
stride=1
|
1339 |
+
pad=1
|
1340 |
+
activation=silu
|
1341 |
+
|
1342 |
+
# Split
|
1343 |
+
|
1344 |
+
[convolutional]
|
1345 |
+
batch_normalize=1
|
1346 |
+
filters=640
|
1347 |
+
size=1
|
1348 |
+
stride=1
|
1349 |
+
pad=1
|
1350 |
+
activation=silu
|
1351 |
+
|
1352 |
+
[route]
|
1353 |
+
layers = -2
|
1354 |
+
|
1355 |
+
# Plain Block
|
1356 |
+
|
1357 |
+
[convolutional]
|
1358 |
+
batch_normalize=1
|
1359 |
+
filters=640
|
1360 |
+
size=1
|
1361 |
+
stride=1
|
1362 |
+
pad=1
|
1363 |
+
activation=silu
|
1364 |
+
|
1365 |
+
[convolutional]
|
1366 |
+
batch_normalize=1
|
1367 |
+
size=3
|
1368 |
+
stride=1
|
1369 |
+
pad=1
|
1370 |
+
filters=640
|
1371 |
+
activation=silu
|
1372 |
+
|
1373 |
+
[convolutional]
|
1374 |
+
batch_normalize=1
|
1375 |
+
filters=640
|
1376 |
+
size=1
|
1377 |
+
stride=1
|
1378 |
+
pad=1
|
1379 |
+
activation=silu
|
1380 |
+
|
1381 |
+
[convolutional]
|
1382 |
+
batch_normalize=1
|
1383 |
+
size=3
|
1384 |
+
stride=1
|
1385 |
+
pad=1
|
1386 |
+
filters=640
|
1387 |
+
activation=silu
|
1388 |
+
|
1389 |
+
[convolutional]
|
1390 |
+
batch_normalize=1
|
1391 |
+
filters=640
|
1392 |
+
size=1
|
1393 |
+
stride=1
|
1394 |
+
pad=1
|
1395 |
+
activation=silu
|
1396 |
+
|
1397 |
+
[convolutional]
|
1398 |
+
batch_normalize=1
|
1399 |
+
size=3
|
1400 |
+
stride=1
|
1401 |
+
pad=1
|
1402 |
+
filters=640
|
1403 |
+
activation=silu
|
1404 |
+
|
1405 |
+
[route]
|
1406 |
+
layers = -1,-8
|
1407 |
+
|
1408 |
+
# Transition last
|
1409 |
+
|
1410 |
+
# 191 (previous+3+4+2k)
|
1411 |
+
[convolutional]
|
1412 |
+
batch_normalize=1
|
1413 |
+
filters=640
|
1414 |
+
size=1
|
1415 |
+
stride=1
|
1416 |
+
pad=1
|
1417 |
+
activation=silu
|
1418 |
+
|
1419 |
+
# ============ End of Neck ============ #
|
1420 |
+
|
1421 |
+
# 192
|
1422 |
+
[implicit_add]
|
1423 |
+
filters=320
|
1424 |
+
|
1425 |
+
# 193
|
1426 |
+
[implicit_add]
|
1427 |
+
filters=640
|
1428 |
+
|
1429 |
+
# 194
|
1430 |
+
[implicit_add]
|
1431 |
+
filters=1280
|
1432 |
+
|
1433 |
+
# 195
|
1434 |
+
[implicit_mul]
|
1435 |
+
filters=255
|
1436 |
+
|
1437 |
+
# 196
|
1438 |
+
[implicit_mul]
|
1439 |
+
filters=255
|
1440 |
+
|
1441 |
+
# 197
|
1442 |
+
[implicit_mul]
|
1443 |
+
filters=255
|
1444 |
+
|
1445 |
+
# ============ Head ============ #
|
1446 |
+
|
1447 |
+
# YOLO-3
|
1448 |
+
|
1449 |
+
[route]
|
1450 |
+
layers = 165
|
1451 |
+
|
1452 |
+
[convolutional]
|
1453 |
+
batch_normalize=1
|
1454 |
+
size=3
|
1455 |
+
stride=1
|
1456 |
+
pad=1
|
1457 |
+
filters=320
|
1458 |
+
activation=silu
|
1459 |
+
|
1460 |
+
[shift_channels]
|
1461 |
+
from=192
|
1462 |
+
|
1463 |
+
[convolutional]
|
1464 |
+
size=1
|
1465 |
+
stride=1
|
1466 |
+
pad=1
|
1467 |
+
filters=255
|
1468 |
+
activation=linear
|
1469 |
+
|
1470 |
+
[control_channels]
|
1471 |
+
from=195
|
1472 |
+
|
1473 |
+
[yolo]
|
1474 |
+
mask = 0,1,2
|
1475 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1476 |
+
classes=80
|
1477 |
+
num=9
|
1478 |
+
jitter=.3
|
1479 |
+
ignore_thresh = .7
|
1480 |
+
truth_thresh = 1
|
1481 |
+
random=1
|
1482 |
+
scale_x_y = 1.05
|
1483 |
+
iou_thresh=0.213
|
1484 |
+
cls_normalizer=1.0
|
1485 |
+
iou_normalizer=0.07
|
1486 |
+
iou_loss=ciou
|
1487 |
+
nms_kind=greedynms
|
1488 |
+
beta_nms=0.6
|
1489 |
+
|
1490 |
+
|
1491 |
+
# YOLO-4
|
1492 |
+
|
1493 |
+
[route]
|
1494 |
+
layers = 178
|
1495 |
+
|
1496 |
+
[convolutional]
|
1497 |
+
batch_normalize=1
|
1498 |
+
size=3
|
1499 |
+
stride=1
|
1500 |
+
pad=1
|
1501 |
+
filters=640
|
1502 |
+
activation=silu
|
1503 |
+
|
1504 |
+
[shift_channels]
|
1505 |
+
from=193
|
1506 |
+
|
1507 |
+
[convolutional]
|
1508 |
+
size=1
|
1509 |
+
stride=1
|
1510 |
+
pad=1
|
1511 |
+
filters=255
|
1512 |
+
activation=linear
|
1513 |
+
|
1514 |
+
[control_channels]
|
1515 |
+
from=196
|
1516 |
+
|
1517 |
+
[yolo]
|
1518 |
+
mask = 3,4,5
|
1519 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1520 |
+
classes=80
|
1521 |
+
num=9
|
1522 |
+
jitter=.3
|
1523 |
+
ignore_thresh = .7
|
1524 |
+
truth_thresh = 1
|
1525 |
+
random=1
|
1526 |
+
scale_x_y = 1.05
|
1527 |
+
iou_thresh=0.213
|
1528 |
+
cls_normalizer=1.0
|
1529 |
+
iou_normalizer=0.07
|
1530 |
+
iou_loss=ciou
|
1531 |
+
nms_kind=greedynms
|
1532 |
+
beta_nms=0.6
|
1533 |
+
|
1534 |
+
|
1535 |
+
# YOLO-5
|
1536 |
+
|
1537 |
+
[route]
|
1538 |
+
layers = 191
|
1539 |
+
|
1540 |
+
[convolutional]
|
1541 |
+
batch_normalize=1
|
1542 |
+
size=3
|
1543 |
+
stride=1
|
1544 |
+
pad=1
|
1545 |
+
filters=1280
|
1546 |
+
activation=silu
|
1547 |
+
|
1548 |
+
[shift_channels]
|
1549 |
+
from=194
|
1550 |
+
|
1551 |
+
[convolutional]
|
1552 |
+
size=1
|
1553 |
+
stride=1
|
1554 |
+
pad=1
|
1555 |
+
filters=255
|
1556 |
+
activation=linear
|
1557 |
+
|
1558 |
+
[control_channels]
|
1559 |
+
from=197
|
1560 |
+
|
1561 |
+
[yolo]
|
1562 |
+
mask = 6,7,8
|
1563 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1564 |
+
classes=80
|
1565 |
+
num=9
|
1566 |
+
jitter=.3
|
1567 |
+
ignore_thresh = .7
|
1568 |
+
truth_thresh = 1
|
1569 |
+
random=1
|
1570 |
+
scale_x_y = 1.05
|
1571 |
+
iou_thresh=0.213
|
1572 |
+
cls_normalizer=1.0
|
1573 |
+
iou_normalizer=0.07
|
1574 |
+
iou_loss=ciou
|
1575 |
+
nms_kind=greedynms
|
1576 |
+
beta_nms=0.6
|
cfg/yolor_p6.cfg
ADDED
@@ -0,0 +1,1760 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[net]
|
2 |
+
batch=64
|
3 |
+
subdivisions=8
|
4 |
+
width=1280
|
5 |
+
height=1280
|
6 |
+
channels=3
|
7 |
+
momentum=0.949
|
8 |
+
decay=0.0005
|
9 |
+
angle=0
|
10 |
+
saturation = 1.5
|
11 |
+
exposure = 1.5
|
12 |
+
hue=.1
|
13 |
+
|
14 |
+
learning_rate=0.00261
|
15 |
+
burn_in=1000
|
16 |
+
max_batches = 500500
|
17 |
+
policy=steps
|
18 |
+
steps=400000,450000
|
19 |
+
scales=.1,.1
|
20 |
+
|
21 |
+
mosaic=1
|
22 |
+
|
23 |
+
|
24 |
+
# ============ Backbone ============ #
|
25 |
+
|
26 |
+
# Stem
|
27 |
+
|
28 |
+
# P1
|
29 |
+
|
30 |
+
# Downsample
|
31 |
+
|
32 |
+
# 0
|
33 |
+
[reorg]
|
34 |
+
|
35 |
+
[convolutional]
|
36 |
+
batch_normalize=1
|
37 |
+
filters=64
|
38 |
+
size=3
|
39 |
+
stride=1
|
40 |
+
pad=1
|
41 |
+
activation=silu
|
42 |
+
|
43 |
+
|
44 |
+
# P2
|
45 |
+
|
46 |
+
# Downsample
|
47 |
+
|
48 |
+
[convolutional]
|
49 |
+
batch_normalize=1
|
50 |
+
filters=128
|
51 |
+
size=3
|
52 |
+
stride=2
|
53 |
+
pad=1
|
54 |
+
activation=silu
|
55 |
+
|
56 |
+
# Split
|
57 |
+
|
58 |
+
[convolutional]
|
59 |
+
batch_normalize=1
|
60 |
+
filters=64
|
61 |
+
size=1
|
62 |
+
stride=1
|
63 |
+
pad=1
|
64 |
+
activation=silu
|
65 |
+
|
66 |
+
[route]
|
67 |
+
layers = -2
|
68 |
+
|
69 |
+
[convolutional]
|
70 |
+
batch_normalize=1
|
71 |
+
filters=64
|
72 |
+
size=1
|
73 |
+
stride=1
|
74 |
+
pad=1
|
75 |
+
activation=silu
|
76 |
+
|
77 |
+
# Residual Block
|
78 |
+
|
79 |
+
[convolutional]
|
80 |
+
batch_normalize=1
|
81 |
+
filters=64
|
82 |
+
size=1
|
83 |
+
stride=1
|
84 |
+
pad=1
|
85 |
+
activation=silu
|
86 |
+
|
87 |
+
[convolutional]
|
88 |
+
batch_normalize=1
|
89 |
+
filters=64
|
90 |
+
size=3
|
91 |
+
stride=1
|
92 |
+
pad=1
|
93 |
+
activation=silu
|
94 |
+
|
95 |
+
[shortcut]
|
96 |
+
from=-3
|
97 |
+
activation=linear
|
98 |
+
|
99 |
+
[convolutional]
|
100 |
+
batch_normalize=1
|
101 |
+
filters=64
|
102 |
+
size=1
|
103 |
+
stride=1
|
104 |
+
pad=1
|
105 |
+
activation=silu
|
106 |
+
|
107 |
+
[convolutional]
|
108 |
+
batch_normalize=1
|
109 |
+
filters=64
|
110 |
+
size=3
|
111 |
+
stride=1
|
112 |
+
pad=1
|
113 |
+
activation=silu
|
114 |
+
|
115 |
+
[shortcut]
|
116 |
+
from=-3
|
117 |
+
activation=linear
|
118 |
+
|
119 |
+
[convolutional]
|
120 |
+
batch_normalize=1
|
121 |
+
filters=64
|
122 |
+
size=1
|
123 |
+
stride=1
|
124 |
+
pad=1
|
125 |
+
activation=silu
|
126 |
+
|
127 |
+
[convolutional]
|
128 |
+
batch_normalize=1
|
129 |
+
filters=64
|
130 |
+
size=3
|
131 |
+
stride=1
|
132 |
+
pad=1
|
133 |
+
activation=silu
|
134 |
+
|
135 |
+
[shortcut]
|
136 |
+
from=-3
|
137 |
+
activation=linear
|
138 |
+
|
139 |
+
# Transition first
|
140 |
+
#
|
141 |
+
#[convolutional]
|
142 |
+
#batch_normalize=1
|
143 |
+
#filters=64
|
144 |
+
#size=1
|
145 |
+
#stride=1
|
146 |
+
#pad=1
|
147 |
+
#activation=silu
|
148 |
+
|
149 |
+
# Merge [-1, -(3k+3)]
|
150 |
+
|
151 |
+
[route]
|
152 |
+
layers = -1,-12
|
153 |
+
|
154 |
+
# Transition last
|
155 |
+
|
156 |
+
# 16 (previous+6+3k)
|
157 |
+
[convolutional]
|
158 |
+
batch_normalize=1
|
159 |
+
filters=128
|
160 |
+
size=1
|
161 |
+
stride=1
|
162 |
+
pad=1
|
163 |
+
activation=silu
|
164 |
+
|
165 |
+
|
166 |
+
# P3
|
167 |
+
|
168 |
+
# Downsample
|
169 |
+
|
170 |
+
[convolutional]
|
171 |
+
batch_normalize=1
|
172 |
+
filters=256
|
173 |
+
size=3
|
174 |
+
stride=2
|
175 |
+
pad=1
|
176 |
+
activation=silu
|
177 |
+
|
178 |
+
# Split
|
179 |
+
|
180 |
+
[convolutional]
|
181 |
+
batch_normalize=1
|
182 |
+
filters=128
|
183 |
+
size=1
|
184 |
+
stride=1
|
185 |
+
pad=1
|
186 |
+
activation=silu
|
187 |
+
|
188 |
+
[route]
|
189 |
+
layers = -2
|
190 |
+
|
191 |
+
[convolutional]
|
192 |
+
batch_normalize=1
|
193 |
+
filters=128
|
194 |
+
size=1
|
195 |
+
stride=1
|
196 |
+
pad=1
|
197 |
+
activation=silu
|
198 |
+
|
199 |
+
# Residual Block
|
200 |
+
|
201 |
+
[convolutional]
|
202 |
+
batch_normalize=1
|
203 |
+
filters=128
|
204 |
+
size=1
|
205 |
+
stride=1
|
206 |
+
pad=1
|
207 |
+
activation=silu
|
208 |
+
|
209 |
+
[convolutional]
|
210 |
+
batch_normalize=1
|
211 |
+
filters=128
|
212 |
+
size=3
|
213 |
+
stride=1
|
214 |
+
pad=1
|
215 |
+
activation=silu
|
216 |
+
|
217 |
+
[shortcut]
|
218 |
+
from=-3
|
219 |
+
activation=linear
|
220 |
+
|
221 |
+
[convolutional]
|
222 |
+
batch_normalize=1
|
223 |
+
filters=128
|
224 |
+
size=1
|
225 |
+
stride=1
|
226 |
+
pad=1
|
227 |
+
activation=silu
|
228 |
+
|
229 |
+
[convolutional]
|
230 |
+
batch_normalize=1
|
231 |
+
filters=128
|
232 |
+
size=3
|
233 |
+
stride=1
|
234 |
+
pad=1
|
235 |
+
activation=silu
|
236 |
+
|
237 |
+
[shortcut]
|
238 |
+
from=-3
|
239 |
+
activation=linear
|
240 |
+
|
241 |
+
[convolutional]
|
242 |
+
batch_normalize=1
|
243 |
+
filters=128
|
244 |
+
size=1
|
245 |
+
stride=1
|
246 |
+
pad=1
|
247 |
+
activation=silu
|
248 |
+
|
249 |
+
[convolutional]
|
250 |
+
batch_normalize=1
|
251 |
+
filters=128
|
252 |
+
size=3
|
253 |
+
stride=1
|
254 |
+
pad=1
|
255 |
+
activation=silu
|
256 |
+
|
257 |
+
[shortcut]
|
258 |
+
from=-3
|
259 |
+
activation=linear
|
260 |
+
|
261 |
+
[convolutional]
|
262 |
+
batch_normalize=1
|
263 |
+
filters=128
|
264 |
+
size=1
|
265 |
+
stride=1
|
266 |
+
pad=1
|
267 |
+
activation=silu
|
268 |
+
|
269 |
+
[convolutional]
|
270 |
+
batch_normalize=1
|
271 |
+
filters=128
|
272 |
+
size=3
|
273 |
+
stride=1
|
274 |
+
pad=1
|
275 |
+
activation=silu
|
276 |
+
|
277 |
+
[shortcut]
|
278 |
+
from=-3
|
279 |
+
activation=linear
|
280 |
+
|
281 |
+
[convolutional]
|
282 |
+
batch_normalize=1
|
283 |
+
filters=128
|
284 |
+
size=1
|
285 |
+
stride=1
|
286 |
+
pad=1
|
287 |
+
activation=silu
|
288 |
+
|
289 |
+
[convolutional]
|
290 |
+
batch_normalize=1
|
291 |
+
filters=128
|
292 |
+
size=3
|
293 |
+
stride=1
|
294 |
+
pad=1
|
295 |
+
activation=silu
|
296 |
+
|
297 |
+
[shortcut]
|
298 |
+
from=-3
|
299 |
+
activation=linear
|
300 |
+
|
301 |
+
[convolutional]
|
302 |
+
batch_normalize=1
|
303 |
+
filters=128
|
304 |
+
size=1
|
305 |
+
stride=1
|
306 |
+
pad=1
|
307 |
+
activation=silu
|
308 |
+
|
309 |
+
[convolutional]
|
310 |
+
batch_normalize=1
|
311 |
+
filters=128
|
312 |
+
size=3
|
313 |
+
stride=1
|
314 |
+
pad=1
|
315 |
+
activation=silu
|
316 |
+
|
317 |
+
[shortcut]
|
318 |
+
from=-3
|
319 |
+
activation=linear
|
320 |
+
|
321 |
+
[convolutional]
|
322 |
+
batch_normalize=1
|
323 |
+
filters=128
|
324 |
+
size=1
|
325 |
+
stride=1
|
326 |
+
pad=1
|
327 |
+
activation=silu
|
328 |
+
|
329 |
+
[convolutional]
|
330 |
+
batch_normalize=1
|
331 |
+
filters=128
|
332 |
+
size=3
|
333 |
+
stride=1
|
334 |
+
pad=1
|
335 |
+
activation=silu
|
336 |
+
|
337 |
+
[shortcut]
|
338 |
+
from=-3
|
339 |
+
activation=linear
|
340 |
+
|
341 |
+
# Transition first
|
342 |
+
#
|
343 |
+
#[convolutional]
|
344 |
+
#batch_normalize=1
|
345 |
+
#filters=128
|
346 |
+
#size=1
|
347 |
+
#stride=1
|
348 |
+
#pad=1
|
349 |
+
#activation=silu
|
350 |
+
|
351 |
+
# Merge [-1, -(3k+3)]
|
352 |
+
|
353 |
+
[route]
|
354 |
+
layers = -1,-24
|
355 |
+
|
356 |
+
# Transition last
|
357 |
+
|
358 |
+
# 43 (previous+6+3k)
|
359 |
+
[convolutional]
|
360 |
+
batch_normalize=1
|
361 |
+
filters=256
|
362 |
+
size=1
|
363 |
+
stride=1
|
364 |
+
pad=1
|
365 |
+
activation=silu
|
366 |
+
|
367 |
+
|
368 |
+
# P4
|
369 |
+
|
370 |
+
# Downsample
|
371 |
+
|
372 |
+
[convolutional]
|
373 |
+
batch_normalize=1
|
374 |
+
filters=384
|
375 |
+
size=3
|
376 |
+
stride=2
|
377 |
+
pad=1
|
378 |
+
activation=silu
|
379 |
+
|
380 |
+
# Split
|
381 |
+
|
382 |
+
[convolutional]
|
383 |
+
batch_normalize=1
|
384 |
+
filters=192
|
385 |
+
size=1
|
386 |
+
stride=1
|
387 |
+
pad=1
|
388 |
+
activation=silu
|
389 |
+
|
390 |
+
[route]
|
391 |
+
layers = -2
|
392 |
+
|
393 |
+
[convolutional]
|
394 |
+
batch_normalize=1
|
395 |
+
filters=192
|
396 |
+
size=1
|
397 |
+
stride=1
|
398 |
+
pad=1
|
399 |
+
activation=silu
|
400 |
+
|
401 |
+
# Residual Block
|
402 |
+
|
403 |
+
[convolutional]
|
404 |
+
batch_normalize=1
|
405 |
+
filters=192
|
406 |
+
size=1
|
407 |
+
stride=1
|
408 |
+
pad=1
|
409 |
+
activation=silu
|
410 |
+
|
411 |
+
[convolutional]
|
412 |
+
batch_normalize=1
|
413 |
+
filters=192
|
414 |
+
size=3
|
415 |
+
stride=1
|
416 |
+
pad=1
|
417 |
+
activation=silu
|
418 |
+
|
419 |
+
[shortcut]
|
420 |
+
from=-3
|
421 |
+
activation=linear
|
422 |
+
|
423 |
+
[convolutional]
|
424 |
+
batch_normalize=1
|
425 |
+
filters=192
|
426 |
+
size=1
|
427 |
+
stride=1
|
428 |
+
pad=1
|
429 |
+
activation=silu
|
430 |
+
|
431 |
+
[convolutional]
|
432 |
+
batch_normalize=1
|
433 |
+
filters=192
|
434 |
+
size=3
|
435 |
+
stride=1
|
436 |
+
pad=1
|
437 |
+
activation=silu
|
438 |
+
|
439 |
+
[shortcut]
|
440 |
+
from=-3
|
441 |
+
activation=linear
|
442 |
+
|
443 |
+
[convolutional]
|
444 |
+
batch_normalize=1
|
445 |
+
filters=192
|
446 |
+
size=1
|
447 |
+
stride=1
|
448 |
+
pad=1
|
449 |
+
activation=silu
|
450 |
+
|
451 |
+
[convolutional]
|
452 |
+
batch_normalize=1
|
453 |
+
filters=192
|
454 |
+
size=3
|
455 |
+
stride=1
|
456 |
+
pad=1
|
457 |
+
activation=silu
|
458 |
+
|
459 |
+
[shortcut]
|
460 |
+
from=-3
|
461 |
+
activation=linear
|
462 |
+
|
463 |
+
[convolutional]
|
464 |
+
batch_normalize=1
|
465 |
+
filters=192
|
466 |
+
size=1
|
467 |
+
stride=1
|
468 |
+
pad=1
|
469 |
+
activation=silu
|
470 |
+
|
471 |
+
[convolutional]
|
472 |
+
batch_normalize=1
|
473 |
+
filters=192
|
474 |
+
size=3
|
475 |
+
stride=1
|
476 |
+
pad=1
|
477 |
+
activation=silu
|
478 |
+
|
479 |
+
[shortcut]
|
480 |
+
from=-3
|
481 |
+
activation=linear
|
482 |
+
|
483 |
+
[convolutional]
|
484 |
+
batch_normalize=1
|
485 |
+
filters=192
|
486 |
+
size=1
|
487 |
+
stride=1
|
488 |
+
pad=1
|
489 |
+
activation=silu
|
490 |
+
|
491 |
+
[convolutional]
|
492 |
+
batch_normalize=1
|
493 |
+
filters=192
|
494 |
+
size=3
|
495 |
+
stride=1
|
496 |
+
pad=1
|
497 |
+
activation=silu
|
498 |
+
|
499 |
+
[shortcut]
|
500 |
+
from=-3
|
501 |
+
activation=linear
|
502 |
+
|
503 |
+
[convolutional]
|
504 |
+
batch_normalize=1
|
505 |
+
filters=192
|
506 |
+
size=1
|
507 |
+
stride=1
|
508 |
+
pad=1
|
509 |
+
activation=silu
|
510 |
+
|
511 |
+
[convolutional]
|
512 |
+
batch_normalize=1
|
513 |
+
filters=192
|
514 |
+
size=3
|
515 |
+
stride=1
|
516 |
+
pad=1
|
517 |
+
activation=silu
|
518 |
+
|
519 |
+
[shortcut]
|
520 |
+
from=-3
|
521 |
+
activation=linear
|
522 |
+
|
523 |
+
[convolutional]
|
524 |
+
batch_normalize=1
|
525 |
+
filters=192
|
526 |
+
size=1
|
527 |
+
stride=1
|
528 |
+
pad=1
|
529 |
+
activation=silu
|
530 |
+
|
531 |
+
[convolutional]
|
532 |
+
batch_normalize=1
|
533 |
+
filters=192
|
534 |
+
size=3
|
535 |
+
stride=1
|
536 |
+
pad=1
|
537 |
+
activation=silu
|
538 |
+
|
539 |
+
[shortcut]
|
540 |
+
from=-3
|
541 |
+
activation=linear
|
542 |
+
|
543 |
+
# Transition first
|
544 |
+
#
|
545 |
+
#[convolutional]
|
546 |
+
#batch_normalize=1
|
547 |
+
#filters=192
|
548 |
+
#size=1
|
549 |
+
#stride=1
|
550 |
+
#pad=1
|
551 |
+
#activation=silu
|
552 |
+
|
553 |
+
# Merge [-1, -(3k+3)]
|
554 |
+
|
555 |
+
[route]
|
556 |
+
layers = -1,-24
|
557 |
+
|
558 |
+
# Transition last
|
559 |
+
|
560 |
+
# 70 (previous+6+3k)
|
561 |
+
[convolutional]
|
562 |
+
batch_normalize=1
|
563 |
+
filters=384
|
564 |
+
size=1
|
565 |
+
stride=1
|
566 |
+
pad=1
|
567 |
+
activation=silu
|
568 |
+
|
569 |
+
|
570 |
+
# P5
|
571 |
+
|
572 |
+
# Downsample
|
573 |
+
|
574 |
+
[convolutional]
|
575 |
+
batch_normalize=1
|
576 |
+
filters=512
|
577 |
+
size=3
|
578 |
+
stride=2
|
579 |
+
pad=1
|
580 |
+
activation=silu
|
581 |
+
|
582 |
+
# Split
|
583 |
+
|
584 |
+
[convolutional]
|
585 |
+
batch_normalize=1
|
586 |
+
filters=256
|
587 |
+
size=1
|
588 |
+
stride=1
|
589 |
+
pad=1
|
590 |
+
activation=silu
|
591 |
+
|
592 |
+
[route]
|
593 |
+
layers = -2
|
594 |
+
|
595 |
+
[convolutional]
|
596 |
+
batch_normalize=1
|
597 |
+
filters=256
|
598 |
+
size=1
|
599 |
+
stride=1
|
600 |
+
pad=1
|
601 |
+
activation=silu
|
602 |
+
|
603 |
+
# Residual Block
|
604 |
+
|
605 |
+
[convolutional]
|
606 |
+
batch_normalize=1
|
607 |
+
filters=256
|
608 |
+
size=1
|
609 |
+
stride=1
|
610 |
+
pad=1
|
611 |
+
activation=silu
|
612 |
+
|
613 |
+
[convolutional]
|
614 |
+
batch_normalize=1
|
615 |
+
filters=256
|
616 |
+
size=3
|
617 |
+
stride=1
|
618 |
+
pad=1
|
619 |
+
activation=silu
|
620 |
+
|
621 |
+
[shortcut]
|
622 |
+
from=-3
|
623 |
+
activation=linear
|
624 |
+
|
625 |
+
[convolutional]
|
626 |
+
batch_normalize=1
|
627 |
+
filters=256
|
628 |
+
size=1
|
629 |
+
stride=1
|
630 |
+
pad=1
|
631 |
+
activation=silu
|
632 |
+
|
633 |
+
[convolutional]
|
634 |
+
batch_normalize=1
|
635 |
+
filters=256
|
636 |
+
size=3
|
637 |
+
stride=1
|
638 |
+
pad=1
|
639 |
+
activation=silu
|
640 |
+
|
641 |
+
[shortcut]
|
642 |
+
from=-3
|
643 |
+
activation=linear
|
644 |
+
|
645 |
+
[convolutional]
|
646 |
+
batch_normalize=1
|
647 |
+
filters=256
|
648 |
+
size=1
|
649 |
+
stride=1
|
650 |
+
pad=1
|
651 |
+
activation=silu
|
652 |
+
|
653 |
+
[convolutional]
|
654 |
+
batch_normalize=1
|
655 |
+
filters=256
|
656 |
+
size=3
|
657 |
+
stride=1
|
658 |
+
pad=1
|
659 |
+
activation=silu
|
660 |
+
|
661 |
+
[shortcut]
|
662 |
+
from=-3
|
663 |
+
activation=linear
|
664 |
+
|
665 |
+
# Transition first
|
666 |
+
#
|
667 |
+
#[convolutional]
|
668 |
+
#batch_normalize=1
|
669 |
+
#filters=256
|
670 |
+
#size=1
|
671 |
+
#stride=1
|
672 |
+
#pad=1
|
673 |
+
#activation=silu
|
674 |
+
|
675 |
+
# Merge [-1, -(3k+3)]
|
676 |
+
|
677 |
+
[route]
|
678 |
+
layers = -1,-12
|
679 |
+
|
680 |
+
# Transition last
|
681 |
+
|
682 |
+
# 85 (previous+6+3k)
|
683 |
+
[convolutional]
|
684 |
+
batch_normalize=1
|
685 |
+
filters=512
|
686 |
+
size=1
|
687 |
+
stride=1
|
688 |
+
pad=1
|
689 |
+
activation=silu
|
690 |
+
|
691 |
+
|
692 |
+
# P6
|
693 |
+
|
694 |
+
# Downsample
|
695 |
+
|
696 |
+
[convolutional]
|
697 |
+
batch_normalize=1
|
698 |
+
filters=640
|
699 |
+
size=3
|
700 |
+
stride=2
|
701 |
+
pad=1
|
702 |
+
activation=silu
|
703 |
+
|
704 |
+
# Split
|
705 |
+
|
706 |
+
[convolutional]
|
707 |
+
batch_normalize=1
|
708 |
+
filters=320
|
709 |
+
size=1
|
710 |
+
stride=1
|
711 |
+
pad=1
|
712 |
+
activation=silu
|
713 |
+
|
714 |
+
[route]
|
715 |
+
layers = -2
|
716 |
+
|
717 |
+
[convolutional]
|
718 |
+
batch_normalize=1
|
719 |
+
filters=320
|
720 |
+
size=1
|
721 |
+
stride=1
|
722 |
+
pad=1
|
723 |
+
activation=silu
|
724 |
+
|
725 |
+
# Residual Block
|
726 |
+
|
727 |
+
[convolutional]
|
728 |
+
batch_normalize=1
|
729 |
+
filters=320
|
730 |
+
size=1
|
731 |
+
stride=1
|
732 |
+
pad=1
|
733 |
+
activation=silu
|
734 |
+
|
735 |
+
[convolutional]
|
736 |
+
batch_normalize=1
|
737 |
+
filters=320
|
738 |
+
size=3
|
739 |
+
stride=1
|
740 |
+
pad=1
|
741 |
+
activation=silu
|
742 |
+
|
743 |
+
[shortcut]
|
744 |
+
from=-3
|
745 |
+
activation=linear
|
746 |
+
|
747 |
+
[convolutional]
|
748 |
+
batch_normalize=1
|
749 |
+
filters=320
|
750 |
+
size=1
|
751 |
+
stride=1
|
752 |
+
pad=1
|
753 |
+
activation=silu
|
754 |
+
|
755 |
+
[convolutional]
|
756 |
+
batch_normalize=1
|
757 |
+
filters=320
|
758 |
+
size=3
|
759 |
+
stride=1
|
760 |
+
pad=1
|
761 |
+
activation=silu
|
762 |
+
|
763 |
+
[shortcut]
|
764 |
+
from=-3
|
765 |
+
activation=linear
|
766 |
+
|
767 |
+
[convolutional]
|
768 |
+
batch_normalize=1
|
769 |
+
filters=320
|
770 |
+
size=1
|
771 |
+
stride=1
|
772 |
+
pad=1
|
773 |
+
activation=silu
|
774 |
+
|
775 |
+
[convolutional]
|
776 |
+
batch_normalize=1
|
777 |
+
filters=320
|
778 |
+
size=3
|
779 |
+
stride=1
|
780 |
+
pad=1
|
781 |
+
activation=silu
|
782 |
+
|
783 |
+
[shortcut]
|
784 |
+
from=-3
|
785 |
+
activation=linear
|
786 |
+
|
787 |
+
# Transition first
|
788 |
+
#
|
789 |
+
#[convolutional]
|
790 |
+
#batch_normalize=1
|
791 |
+
#filters=320
|
792 |
+
#size=1
|
793 |
+
#stride=1
|
794 |
+
#pad=1
|
795 |
+
#activation=silu
|
796 |
+
|
797 |
+
# Merge [-1, -(3k+3)]
|
798 |
+
|
799 |
+
[route]
|
800 |
+
layers = -1,-12
|
801 |
+
|
802 |
+
# Transition last
|
803 |
+
|
804 |
+
# 100 (previous+6+3k)
|
805 |
+
[convolutional]
|
806 |
+
batch_normalize=1
|
807 |
+
filters=640
|
808 |
+
size=1
|
809 |
+
stride=1
|
810 |
+
pad=1
|
811 |
+
activation=silu
|
812 |
+
|
813 |
+
# ============ End of Backbone ============ #
|
814 |
+
|
815 |
+
# ============ Neck ============ #
|
816 |
+
|
817 |
+
# CSPSPP
|
818 |
+
|
819 |
+
[convolutional]
|
820 |
+
batch_normalize=1
|
821 |
+
filters=320
|
822 |
+
size=1
|
823 |
+
stride=1
|
824 |
+
pad=1
|
825 |
+
activation=silu
|
826 |
+
|
827 |
+
[route]
|
828 |
+
layers = -2
|
829 |
+
|
830 |
+
[convolutional]
|
831 |
+
batch_normalize=1
|
832 |
+
filters=320
|
833 |
+
size=1
|
834 |
+
stride=1
|
835 |
+
pad=1
|
836 |
+
activation=silu
|
837 |
+
|
838 |
+
[convolutional]
|
839 |
+
batch_normalize=1
|
840 |
+
size=3
|
841 |
+
stride=1
|
842 |
+
pad=1
|
843 |
+
filters=320
|
844 |
+
activation=silu
|
845 |
+
|
846 |
+
[convolutional]
|
847 |
+
batch_normalize=1
|
848 |
+
filters=320
|
849 |
+
size=1
|
850 |
+
stride=1
|
851 |
+
pad=1
|
852 |
+
activation=silu
|
853 |
+
|
854 |
+
### SPP ###
|
855 |
+
[maxpool]
|
856 |
+
stride=1
|
857 |
+
size=5
|
858 |
+
|
859 |
+
[route]
|
860 |
+
layers=-2
|
861 |
+
|
862 |
+
[maxpool]
|
863 |
+
stride=1
|
864 |
+
size=9
|
865 |
+
|
866 |
+
[route]
|
867 |
+
layers=-4
|
868 |
+
|
869 |
+
[maxpool]
|
870 |
+
stride=1
|
871 |
+
size=13
|
872 |
+
|
873 |
+
[route]
|
874 |
+
layers=-1,-3,-5,-6
|
875 |
+
### End SPP ###
|
876 |
+
|
877 |
+
[convolutional]
|
878 |
+
batch_normalize=1
|
879 |
+
filters=320
|
880 |
+
size=1
|
881 |
+
stride=1
|
882 |
+
pad=1
|
883 |
+
activation=silu
|
884 |
+
|
885 |
+
[convolutional]
|
886 |
+
batch_normalize=1
|
887 |
+
size=3
|
888 |
+
stride=1
|
889 |
+
pad=1
|
890 |
+
filters=320
|
891 |
+
activation=silu
|
892 |
+
|
893 |
+
[route]
|
894 |
+
layers = -1, -13
|
895 |
+
|
896 |
+
# 115 (previous+6+5+2k)
|
897 |
+
[convolutional]
|
898 |
+
batch_normalize=1
|
899 |
+
filters=320
|
900 |
+
size=1
|
901 |
+
stride=1
|
902 |
+
pad=1
|
903 |
+
activation=silu
|
904 |
+
|
905 |
+
# End of CSPSPP
|
906 |
+
|
907 |
+
|
908 |
+
# FPN-5
|
909 |
+
|
910 |
+
[convolutional]
|
911 |
+
batch_normalize=1
|
912 |
+
filters=256
|
913 |
+
size=1
|
914 |
+
stride=1
|
915 |
+
pad=1
|
916 |
+
activation=silu
|
917 |
+
|
918 |
+
[upsample]
|
919 |
+
stride=2
|
920 |
+
|
921 |
+
[route]
|
922 |
+
layers = 85
|
923 |
+
|
924 |
+
[convolutional]
|
925 |
+
batch_normalize=1
|
926 |
+
filters=256
|
927 |
+
size=1
|
928 |
+
stride=1
|
929 |
+
pad=1
|
930 |
+
activation=silu
|
931 |
+
|
932 |
+
[route]
|
933 |
+
layers = -1, -3
|
934 |
+
|
935 |
+
[convolutional]
|
936 |
+
batch_normalize=1
|
937 |
+
filters=256
|
938 |
+
size=1
|
939 |
+
stride=1
|
940 |
+
pad=1
|
941 |
+
activation=silu
|
942 |
+
|
943 |
+
# Split
|
944 |
+
|
945 |
+
[convolutional]
|
946 |
+
batch_normalize=1
|
947 |
+
filters=256
|
948 |
+
size=1
|
949 |
+
stride=1
|
950 |
+
pad=1
|
951 |
+
activation=silu
|
952 |
+
|
953 |
+
[route]
|
954 |
+
layers = -2
|
955 |
+
|
956 |
+
# Plain Block
|
957 |
+
|
958 |
+
[convolutional]
|
959 |
+
batch_normalize=1
|
960 |
+
filters=256
|
961 |
+
size=1
|
962 |
+
stride=1
|
963 |
+
pad=1
|
964 |
+
activation=silu
|
965 |
+
|
966 |
+
[convolutional]
|
967 |
+
batch_normalize=1
|
968 |
+
size=3
|
969 |
+
stride=1
|
970 |
+
pad=1
|
971 |
+
filters=256
|
972 |
+
activation=silu
|
973 |
+
|
974 |
+
[convolutional]
|
975 |
+
batch_normalize=1
|
976 |
+
filters=256
|
977 |
+
size=1
|
978 |
+
stride=1
|
979 |
+
pad=1
|
980 |
+
activation=silu
|
981 |
+
|
982 |
+
[convolutional]
|
983 |
+
batch_normalize=1
|
984 |
+
size=3
|
985 |
+
stride=1
|
986 |
+
pad=1
|
987 |
+
filters=256
|
988 |
+
activation=silu
|
989 |
+
|
990 |
+
[convolutional]
|
991 |
+
batch_normalize=1
|
992 |
+
filters=256
|
993 |
+
size=1
|
994 |
+
stride=1
|
995 |
+
pad=1
|
996 |
+
activation=silu
|
997 |
+
|
998 |
+
[convolutional]
|
999 |
+
batch_normalize=1
|
1000 |
+
size=3
|
1001 |
+
stride=1
|
1002 |
+
pad=1
|
1003 |
+
filters=256
|
1004 |
+
activation=silu
|
1005 |
+
|
1006 |
+
# Merge [-1, -(2k+2)]
|
1007 |
+
|
1008 |
+
[route]
|
1009 |
+
layers = -1, -8
|
1010 |
+
|
1011 |
+
# Transition last
|
1012 |
+
|
1013 |
+
# 131 (previous+6+4+2k)
|
1014 |
+
[convolutional]
|
1015 |
+
batch_normalize=1
|
1016 |
+
filters=256
|
1017 |
+
size=1
|
1018 |
+
stride=1
|
1019 |
+
pad=1
|
1020 |
+
activation=silu
|
1021 |
+
|
1022 |
+
|
1023 |
+
# FPN-4
|
1024 |
+
|
1025 |
+
[convolutional]
|
1026 |
+
batch_normalize=1
|
1027 |
+
filters=192
|
1028 |
+
size=1
|
1029 |
+
stride=1
|
1030 |
+
pad=1
|
1031 |
+
activation=silu
|
1032 |
+
|
1033 |
+
[upsample]
|
1034 |
+
stride=2
|
1035 |
+
|
1036 |
+
[route]
|
1037 |
+
layers = 70
|
1038 |
+
|
1039 |
+
[convolutional]
|
1040 |
+
batch_normalize=1
|
1041 |
+
filters=192
|
1042 |
+
size=1
|
1043 |
+
stride=1
|
1044 |
+
pad=1
|
1045 |
+
activation=silu
|
1046 |
+
|
1047 |
+
[route]
|
1048 |
+
layers = -1, -3
|
1049 |
+
|
1050 |
+
[convolutional]
|
1051 |
+
batch_normalize=1
|
1052 |
+
filters=192
|
1053 |
+
size=1
|
1054 |
+
stride=1
|
1055 |
+
pad=1
|
1056 |
+
activation=silu
|
1057 |
+
|
1058 |
+
# Split
|
1059 |
+
|
1060 |
+
[convolutional]
|
1061 |
+
batch_normalize=1
|
1062 |
+
filters=192
|
1063 |
+
size=1
|
1064 |
+
stride=1
|
1065 |
+
pad=1
|
1066 |
+
activation=silu
|
1067 |
+
|
1068 |
+
[route]
|
1069 |
+
layers = -2
|
1070 |
+
|
1071 |
+
# Plain Block
|
1072 |
+
|
1073 |
+
[convolutional]
|
1074 |
+
batch_normalize=1
|
1075 |
+
filters=192
|
1076 |
+
size=1
|
1077 |
+
stride=1
|
1078 |
+
pad=1
|
1079 |
+
activation=silu
|
1080 |
+
|
1081 |
+
[convolutional]
|
1082 |
+
batch_normalize=1
|
1083 |
+
size=3
|
1084 |
+
stride=1
|
1085 |
+
pad=1
|
1086 |
+
filters=192
|
1087 |
+
activation=silu
|
1088 |
+
|
1089 |
+
[convolutional]
|
1090 |
+
batch_normalize=1
|
1091 |
+
filters=192
|
1092 |
+
size=1
|
1093 |
+
stride=1
|
1094 |
+
pad=1
|
1095 |
+
activation=silu
|
1096 |
+
|
1097 |
+
[convolutional]
|
1098 |
+
batch_normalize=1
|
1099 |
+
size=3
|
1100 |
+
stride=1
|
1101 |
+
pad=1
|
1102 |
+
filters=192
|
1103 |
+
activation=silu
|
1104 |
+
|
1105 |
+
[convolutional]
|
1106 |
+
batch_normalize=1
|
1107 |
+
filters=192
|
1108 |
+
size=1
|
1109 |
+
stride=1
|
1110 |
+
pad=1
|
1111 |
+
activation=silu
|
1112 |
+
|
1113 |
+
[convolutional]
|
1114 |
+
batch_normalize=1
|
1115 |
+
size=3
|
1116 |
+
stride=1
|
1117 |
+
pad=1
|
1118 |
+
filters=192
|
1119 |
+
activation=silu
|
1120 |
+
|
1121 |
+
# Merge [-1, -(2k+2)]
|
1122 |
+
|
1123 |
+
[route]
|
1124 |
+
layers = -1, -8
|
1125 |
+
|
1126 |
+
# Transition last
|
1127 |
+
|
1128 |
+
# 147 (previous+6+4+2k)
|
1129 |
+
[convolutional]
|
1130 |
+
batch_normalize=1
|
1131 |
+
filters=192
|
1132 |
+
size=1
|
1133 |
+
stride=1
|
1134 |
+
pad=1
|
1135 |
+
activation=silu
|
1136 |
+
|
1137 |
+
|
1138 |
+
# FPN-3
|
1139 |
+
|
1140 |
+
[convolutional]
|
1141 |
+
batch_normalize=1
|
1142 |
+
filters=128
|
1143 |
+
size=1
|
1144 |
+
stride=1
|
1145 |
+
pad=1
|
1146 |
+
activation=silu
|
1147 |
+
|
1148 |
+
[upsample]
|
1149 |
+
stride=2
|
1150 |
+
|
1151 |
+
[route]
|
1152 |
+
layers = 43
|
1153 |
+
|
1154 |
+
[convolutional]
|
1155 |
+
batch_normalize=1
|
1156 |
+
filters=128
|
1157 |
+
size=1
|
1158 |
+
stride=1
|
1159 |
+
pad=1
|
1160 |
+
activation=silu
|
1161 |
+
|
1162 |
+
[route]
|
1163 |
+
layers = -1, -3
|
1164 |
+
|
1165 |
+
[convolutional]
|
1166 |
+
batch_normalize=1
|
1167 |
+
filters=128
|
1168 |
+
size=1
|
1169 |
+
stride=1
|
1170 |
+
pad=1
|
1171 |
+
activation=silu
|
1172 |
+
|
1173 |
+
# Split
|
1174 |
+
|
1175 |
+
[convolutional]
|
1176 |
+
batch_normalize=1
|
1177 |
+
filters=128
|
1178 |
+
size=1
|
1179 |
+
stride=1
|
1180 |
+
pad=1
|
1181 |
+
activation=silu
|
1182 |
+
|
1183 |
+
[route]
|
1184 |
+
layers = -2
|
1185 |
+
|
1186 |
+
# Plain Block
|
1187 |
+
|
1188 |
+
[convolutional]
|
1189 |
+
batch_normalize=1
|
1190 |
+
filters=128
|
1191 |
+
size=1
|
1192 |
+
stride=1
|
1193 |
+
pad=1
|
1194 |
+
activation=silu
|
1195 |
+
|
1196 |
+
[convolutional]
|
1197 |
+
batch_normalize=1
|
1198 |
+
size=3
|
1199 |
+
stride=1
|
1200 |
+
pad=1
|
1201 |
+
filters=128
|
1202 |
+
activation=silu
|
1203 |
+
|
1204 |
+
[convolutional]
|
1205 |
+
batch_normalize=1
|
1206 |
+
filters=128
|
1207 |
+
size=1
|
1208 |
+
stride=1
|
1209 |
+
pad=1
|
1210 |
+
activation=silu
|
1211 |
+
|
1212 |
+
[convolutional]
|
1213 |
+
batch_normalize=1
|
1214 |
+
size=3
|
1215 |
+
stride=1
|
1216 |
+
pad=1
|
1217 |
+
filters=128
|
1218 |
+
activation=silu
|
1219 |
+
|
1220 |
+
[convolutional]
|
1221 |
+
batch_normalize=1
|
1222 |
+
filters=128
|
1223 |
+
size=1
|
1224 |
+
stride=1
|
1225 |
+
pad=1
|
1226 |
+
activation=silu
|
1227 |
+
|
1228 |
+
[convolutional]
|
1229 |
+
batch_normalize=1
|
1230 |
+
size=3
|
1231 |
+
stride=1
|
1232 |
+
pad=1
|
1233 |
+
filters=128
|
1234 |
+
activation=silu
|
1235 |
+
|
1236 |
+
# Merge [-1, -(2k+2)]
|
1237 |
+
|
1238 |
+
[route]
|
1239 |
+
layers = -1, -8
|
1240 |
+
|
1241 |
+
# Transition last
|
1242 |
+
|
1243 |
+
# 163 (previous+6+4+2k)
|
1244 |
+
[convolutional]
|
1245 |
+
batch_normalize=1
|
1246 |
+
filters=128
|
1247 |
+
size=1
|
1248 |
+
stride=1
|
1249 |
+
pad=1
|
1250 |
+
activation=silu
|
1251 |
+
|
1252 |
+
|
1253 |
+
# PAN-4
|
1254 |
+
|
1255 |
+
[convolutional]
|
1256 |
+
batch_normalize=1
|
1257 |
+
size=3
|
1258 |
+
stride=2
|
1259 |
+
pad=1
|
1260 |
+
filters=192
|
1261 |
+
activation=silu
|
1262 |
+
|
1263 |
+
[route]
|
1264 |
+
layers = -1, 147
|
1265 |
+
|
1266 |
+
[convolutional]
|
1267 |
+
batch_normalize=1
|
1268 |
+
filters=192
|
1269 |
+
size=1
|
1270 |
+
stride=1
|
1271 |
+
pad=1
|
1272 |
+
activation=silu
|
1273 |
+
|
1274 |
+
# Split
|
1275 |
+
|
1276 |
+
[convolutional]
|
1277 |
+
batch_normalize=1
|
1278 |
+
filters=192
|
1279 |
+
size=1
|
1280 |
+
stride=1
|
1281 |
+
pad=1
|
1282 |
+
activation=silu
|
1283 |
+
|
1284 |
+
[route]
|
1285 |
+
layers = -2
|
1286 |
+
|
1287 |
+
# Plain Block
|
1288 |
+
|
1289 |
+
[convolutional]
|
1290 |
+
batch_normalize=1
|
1291 |
+
filters=192
|
1292 |
+
size=1
|
1293 |
+
stride=1
|
1294 |
+
pad=1
|
1295 |
+
activation=silu
|
1296 |
+
|
1297 |
+
[convolutional]
|
1298 |
+
batch_normalize=1
|
1299 |
+
size=3
|
1300 |
+
stride=1
|
1301 |
+
pad=1
|
1302 |
+
filters=192
|
1303 |
+
activation=silu
|
1304 |
+
|
1305 |
+
[convolutional]
|
1306 |
+
batch_normalize=1
|
1307 |
+
filters=192
|
1308 |
+
size=1
|
1309 |
+
stride=1
|
1310 |
+
pad=1
|
1311 |
+
activation=silu
|
1312 |
+
|
1313 |
+
[convolutional]
|
1314 |
+
batch_normalize=1
|
1315 |
+
size=3
|
1316 |
+
stride=1
|
1317 |
+
pad=1
|
1318 |
+
filters=192
|
1319 |
+
activation=silu
|
1320 |
+
|
1321 |
+
[convolutional]
|
1322 |
+
batch_normalize=1
|
1323 |
+
filters=192
|
1324 |
+
size=1
|
1325 |
+
stride=1
|
1326 |
+
pad=1
|
1327 |
+
activation=silu
|
1328 |
+
|
1329 |
+
[convolutional]
|
1330 |
+
batch_normalize=1
|
1331 |
+
size=3
|
1332 |
+
stride=1
|
1333 |
+
pad=1
|
1334 |
+
filters=192
|
1335 |
+
activation=silu
|
1336 |
+
|
1337 |
+
[route]
|
1338 |
+
layers = -1,-8
|
1339 |
+
|
1340 |
+
# Transition last
|
1341 |
+
|
1342 |
+
# 176 (previous+3+4+2k)
|
1343 |
+
[convolutional]
|
1344 |
+
batch_normalize=1
|
1345 |
+
filters=192
|
1346 |
+
size=1
|
1347 |
+
stride=1
|
1348 |
+
pad=1
|
1349 |
+
activation=silu
|
1350 |
+
|
1351 |
+
|
1352 |
+
# PAN-5
|
1353 |
+
|
1354 |
+
[convolutional]
|
1355 |
+
batch_normalize=1
|
1356 |
+
size=3
|
1357 |
+
stride=2
|
1358 |
+
pad=1
|
1359 |
+
filters=256
|
1360 |
+
activation=silu
|
1361 |
+
|
1362 |
+
[route]
|
1363 |
+
layers = -1, 131
|
1364 |
+
|
1365 |
+
[convolutional]
|
1366 |
+
batch_normalize=1
|
1367 |
+
filters=256
|
1368 |
+
size=1
|
1369 |
+
stride=1
|
1370 |
+
pad=1
|
1371 |
+
activation=silu
|
1372 |
+
|
1373 |
+
# Split
|
1374 |
+
|
1375 |
+
[convolutional]
|
1376 |
+
batch_normalize=1
|
1377 |
+
filters=256
|
1378 |
+
size=1
|
1379 |
+
stride=1
|
1380 |
+
pad=1
|
1381 |
+
activation=silu
|
1382 |
+
|
1383 |
+
[route]
|
1384 |
+
layers = -2
|
1385 |
+
|
1386 |
+
# Plain Block
|
1387 |
+
|
1388 |
+
[convolutional]
|
1389 |
+
batch_normalize=1
|
1390 |
+
filters=256
|
1391 |
+
size=1
|
1392 |
+
stride=1
|
1393 |
+
pad=1
|
1394 |
+
activation=silu
|
1395 |
+
|
1396 |
+
[convolutional]
|
1397 |
+
batch_normalize=1
|
1398 |
+
size=3
|
1399 |
+
stride=1
|
1400 |
+
pad=1
|
1401 |
+
filters=256
|
1402 |
+
activation=silu
|
1403 |
+
|
1404 |
+
[convolutional]
|
1405 |
+
batch_normalize=1
|
1406 |
+
filters=256
|
1407 |
+
size=1
|
1408 |
+
stride=1
|
1409 |
+
pad=1
|
1410 |
+
activation=silu
|
1411 |
+
|
1412 |
+
[convolutional]
|
1413 |
+
batch_normalize=1
|
1414 |
+
size=3
|
1415 |
+
stride=1
|
1416 |
+
pad=1
|
1417 |
+
filters=256
|
1418 |
+
activation=silu
|
1419 |
+
|
1420 |
+
[convolutional]
|
1421 |
+
batch_normalize=1
|
1422 |
+
filters=256
|
1423 |
+
size=1
|
1424 |
+
stride=1
|
1425 |
+
pad=1
|
1426 |
+
activation=silu
|
1427 |
+
|
1428 |
+
[convolutional]
|
1429 |
+
batch_normalize=1
|
1430 |
+
size=3
|
1431 |
+
stride=1
|
1432 |
+
pad=1
|
1433 |
+
filters=256
|
1434 |
+
activation=silu
|
1435 |
+
|
1436 |
+
[route]
|
1437 |
+
layers = -1,-8
|
1438 |
+
|
1439 |
+
# Transition last
|
1440 |
+
|
1441 |
+
# 189 (previous+3+4+2k)
|
1442 |
+
[convolutional]
|
1443 |
+
batch_normalize=1
|
1444 |
+
filters=256
|
1445 |
+
size=1
|
1446 |
+
stride=1
|
1447 |
+
pad=1
|
1448 |
+
activation=silu
|
1449 |
+
|
1450 |
+
|
1451 |
+
# PAN-6
|
1452 |
+
|
1453 |
+
[convolutional]
|
1454 |
+
batch_normalize=1
|
1455 |
+
size=3
|
1456 |
+
stride=2
|
1457 |
+
pad=1
|
1458 |
+
filters=320
|
1459 |
+
activation=silu
|
1460 |
+
|
1461 |
+
[route]
|
1462 |
+
layers = -1, 115
|
1463 |
+
|
1464 |
+
[convolutional]
|
1465 |
+
batch_normalize=1
|
1466 |
+
filters=320
|
1467 |
+
size=1
|
1468 |
+
stride=1
|
1469 |
+
pad=1
|
1470 |
+
activation=silu
|
1471 |
+
|
1472 |
+
# Split
|
1473 |
+
|
1474 |
+
[convolutional]
|
1475 |
+
batch_normalize=1
|
1476 |
+
filters=320
|
1477 |
+
size=1
|
1478 |
+
stride=1
|
1479 |
+
pad=1
|
1480 |
+
activation=silu
|
1481 |
+
|
1482 |
+
[route]
|
1483 |
+
layers = -2
|
1484 |
+
|
1485 |
+
# Plain Block
|
1486 |
+
|
1487 |
+
[convolutional]
|
1488 |
+
batch_normalize=1
|
1489 |
+
filters=320
|
1490 |
+
size=1
|
1491 |
+
stride=1
|
1492 |
+
pad=1
|
1493 |
+
activation=silu
|
1494 |
+
|
1495 |
+
[convolutional]
|
1496 |
+
batch_normalize=1
|
1497 |
+
size=3
|
1498 |
+
stride=1
|
1499 |
+
pad=1
|
1500 |
+
filters=320
|
1501 |
+
activation=silu
|
1502 |
+
|
1503 |
+
[convolutional]
|
1504 |
+
batch_normalize=1
|
1505 |
+
filters=320
|
1506 |
+
size=1
|
1507 |
+
stride=1
|
1508 |
+
pad=1
|
1509 |
+
activation=silu
|
1510 |
+
|
1511 |
+
[convolutional]
|
1512 |
+
batch_normalize=1
|
1513 |
+
size=3
|
1514 |
+
stride=1
|
1515 |
+
pad=1
|
1516 |
+
filters=320
|
1517 |
+
activation=silu
|
1518 |
+
|
1519 |
+
[convolutional]
|
1520 |
+
batch_normalize=1
|
1521 |
+
filters=320
|
1522 |
+
size=1
|
1523 |
+
stride=1
|
1524 |
+
pad=1
|
1525 |
+
activation=silu
|
1526 |
+
|
1527 |
+
[convolutional]
|
1528 |
+
batch_normalize=1
|
1529 |
+
size=3
|
1530 |
+
stride=1
|
1531 |
+
pad=1
|
1532 |
+
filters=320
|
1533 |
+
activation=silu
|
1534 |
+
|
1535 |
+
[route]
|
1536 |
+
layers = -1,-8
|
1537 |
+
|
1538 |
+
# Transition last
|
1539 |
+
|
1540 |
+
# 202 (previous+3+4+2k)
|
1541 |
+
[convolutional]
|
1542 |
+
batch_normalize=1
|
1543 |
+
filters=320
|
1544 |
+
size=1
|
1545 |
+
stride=1
|
1546 |
+
pad=1
|
1547 |
+
activation=silu
|
1548 |
+
|
1549 |
+
# ============ End of Neck ============ #
|
1550 |
+
|
1551 |
+
# 203
|
1552 |
+
[implicit_add]
|
1553 |
+
filters=256
|
1554 |
+
|
1555 |
+
# 204
|
1556 |
+
[implicit_add]
|
1557 |
+
filters=384
|
1558 |
+
|
1559 |
+
# 205
|
1560 |
+
[implicit_add]
|
1561 |
+
filters=512
|
1562 |
+
|
1563 |
+
# 206
|
1564 |
+
[implicit_add]
|
1565 |
+
filters=640
|
1566 |
+
|
1567 |
+
# 207
|
1568 |
+
[implicit_mul]
|
1569 |
+
filters=255
|
1570 |
+
|
1571 |
+
# 208
|
1572 |
+
[implicit_mul]
|
1573 |
+
filters=255
|
1574 |
+
|
1575 |
+
# 209
|
1576 |
+
[implicit_mul]
|
1577 |
+
filters=255
|
1578 |
+
|
1579 |
+
# 210
|
1580 |
+
[implicit_mul]
|
1581 |
+
filters=255
|
1582 |
+
|
1583 |
+
# ============ Head ============ #
|
1584 |
+
|
1585 |
+
# YOLO-3
|
1586 |
+
|
1587 |
+
[route]
|
1588 |
+
layers = 163
|
1589 |
+
|
1590 |
+
[convolutional]
|
1591 |
+
batch_normalize=1
|
1592 |
+
size=3
|
1593 |
+
stride=1
|
1594 |
+
pad=1
|
1595 |
+
filters=256
|
1596 |
+
activation=silu
|
1597 |
+
|
1598 |
+
[shift_channels]
|
1599 |
+
from=203
|
1600 |
+
|
1601 |
+
[convolutional]
|
1602 |
+
size=1
|
1603 |
+
stride=1
|
1604 |
+
pad=1
|
1605 |
+
filters=255
|
1606 |
+
activation=linear
|
1607 |
+
|
1608 |
+
[control_channels]
|
1609 |
+
from=207
|
1610 |
+
|
1611 |
+
[yolo]
|
1612 |
+
mask = 0,1,2
|
1613 |
+
anchors = 19,27, 44,40, 38,94, 96,68, 86,152, 180,137, 140,301, 303,264, 238,542, 436,615, 739,380, 925,792
|
1614 |
+
classes=80
|
1615 |
+
num=12
|
1616 |
+
jitter=.3
|
1617 |
+
ignore_thresh = .7
|
1618 |
+
truth_thresh = 1
|
1619 |
+
random=1
|
1620 |
+
scale_x_y = 1.05
|
1621 |
+
iou_thresh=0.213
|
1622 |
+
cls_normalizer=1.0
|
1623 |
+
iou_normalizer=0.07
|
1624 |
+
iou_loss=ciou
|
1625 |
+
nms_kind=greedynms
|
1626 |
+
beta_nms=0.6
|
1627 |
+
|
1628 |
+
|
1629 |
+
# YOLO-4
|
1630 |
+
|
1631 |
+
[route]
|
1632 |
+
layers = 176
|
1633 |
+
|
1634 |
+
[convolutional]
|
1635 |
+
batch_normalize=1
|
1636 |
+
size=3
|
1637 |
+
stride=1
|
1638 |
+
pad=1
|
1639 |
+
filters=384
|
1640 |
+
activation=silu
|
1641 |
+
|
1642 |
+
[shift_channels]
|
1643 |
+
from=204
|
1644 |
+
|
1645 |
+
[convolutional]
|
1646 |
+
size=1
|
1647 |
+
stride=1
|
1648 |
+
pad=1
|
1649 |
+
filters=255
|
1650 |
+
activation=linear
|
1651 |
+
|
1652 |
+
[control_channels]
|
1653 |
+
from=208
|
1654 |
+
|
1655 |
+
[yolo]
|
1656 |
+
mask = 3,4,5
|
1657 |
+
anchors = 19,27, 44,40, 38,94, 96,68, 86,152, 180,137, 140,301, 303,264, 238,542, 436,615, 739,380, 925,792
|
1658 |
+
classes=80
|
1659 |
+
num=12
|
1660 |
+
jitter=.3
|
1661 |
+
ignore_thresh = .7
|
1662 |
+
truth_thresh = 1
|
1663 |
+
random=1
|
1664 |
+
scale_x_y = 1.05
|
1665 |
+
iou_thresh=0.213
|
1666 |
+
cls_normalizer=1.0
|
1667 |
+
iou_normalizer=0.07
|
1668 |
+
iou_loss=ciou
|
1669 |
+
nms_kind=greedynms
|
1670 |
+
beta_nms=0.6
|
1671 |
+
|
1672 |
+
|
1673 |
+
# YOLO-5
|
1674 |
+
|
1675 |
+
[route]
|
1676 |
+
layers = 189
|
1677 |
+
|
1678 |
+
[convolutional]
|
1679 |
+
batch_normalize=1
|
1680 |
+
size=3
|
1681 |
+
stride=1
|
1682 |
+
pad=1
|
1683 |
+
filters=512
|
1684 |
+
activation=silu
|
1685 |
+
|
1686 |
+
[shift_channels]
|
1687 |
+
from=205
|
1688 |
+
|
1689 |
+
[convolutional]
|
1690 |
+
size=1
|
1691 |
+
stride=1
|
1692 |
+
pad=1
|
1693 |
+
filters=255
|
1694 |
+
activation=linear
|
1695 |
+
|
1696 |
+
[control_channels]
|
1697 |
+
from=209
|
1698 |
+
|
1699 |
+
[yolo]
|
1700 |
+
mask = 6,7,8
|
1701 |
+
anchors = 19,27, 44,40, 38,94, 96,68, 86,152, 180,137, 140,301, 303,264, 238,542, 436,615, 739,380, 925,792
|
1702 |
+
classes=80
|
1703 |
+
num=12
|
1704 |
+
jitter=.3
|
1705 |
+
ignore_thresh = .7
|
1706 |
+
truth_thresh = 1
|
1707 |
+
random=1
|
1708 |
+
scale_x_y = 1.05
|
1709 |
+
iou_thresh=0.213
|
1710 |
+
cls_normalizer=1.0
|
1711 |
+
iou_normalizer=0.07
|
1712 |
+
iou_loss=ciou
|
1713 |
+
nms_kind=greedynms
|
1714 |
+
beta_nms=0.6
|
1715 |
+
|
1716 |
+
|
1717 |
+
# YOLO-6
|
1718 |
+
|
1719 |
+
[route]
|
1720 |
+
layers = 202
|
1721 |
+
|
1722 |
+
[convolutional]
|
1723 |
+
batch_normalize=1
|
1724 |
+
size=3
|
1725 |
+
stride=1
|
1726 |
+
pad=1
|
1727 |
+
filters=640
|
1728 |
+
activation=silu
|
1729 |
+
|
1730 |
+
[shift_channels]
|
1731 |
+
from=206
|
1732 |
+
|
1733 |
+
[convolutional]
|
1734 |
+
size=1
|
1735 |
+
stride=1
|
1736 |
+
pad=1
|
1737 |
+
filters=255
|
1738 |
+
activation=linear
|
1739 |
+
|
1740 |
+
[control_channels]
|
1741 |
+
from=210
|
1742 |
+
|
1743 |
+
[yolo]
|
1744 |
+
mask = 9,10,11
|
1745 |
+
anchors = 19,27, 44,40, 38,94, 96,68, 86,152, 180,137, 140,301, 303,264, 238,542, 436,615, 739,380, 925,792
|
1746 |
+
classes=80
|
1747 |
+
num=12
|
1748 |
+
jitter=.3
|
1749 |
+
ignore_thresh = .7
|
1750 |
+
truth_thresh = 1
|
1751 |
+
random=1
|
1752 |
+
scale_x_y = 1.05
|
1753 |
+
iou_thresh=0.213
|
1754 |
+
cls_normalizer=1.0
|
1755 |
+
iou_normalizer=0.07
|
1756 |
+
iou_loss=ciou
|
1757 |
+
nms_kind=greedynms
|
1758 |
+
beta_nms=0.6
|
1759 |
+
|
1760 |
+
# ============ End of Head ============ #
|
cfg/yolor_w6.cfg
ADDED
@@ -0,0 +1,1760 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[net]
|
2 |
+
batch=64
|
3 |
+
subdivisions=8
|
4 |
+
width=1280
|
5 |
+
height=1280
|
6 |
+
channels=3
|
7 |
+
momentum=0.949
|
8 |
+
decay=0.0005
|
9 |
+
angle=0
|
10 |
+
saturation = 1.5
|
11 |
+
exposure = 1.5
|
12 |
+
hue=.1
|
13 |
+
|
14 |
+
learning_rate=0.00261
|
15 |
+
burn_in=1000
|
16 |
+
max_batches = 500500
|
17 |
+
policy=steps
|
18 |
+
steps=400000,450000
|
19 |
+
scales=.1,.1
|
20 |
+
|
21 |
+
mosaic=1
|
22 |
+
|
23 |
+
|
24 |
+
# ============ Backbone ============ #
|
25 |
+
|
26 |
+
# Stem
|
27 |
+
|
28 |
+
# P1
|
29 |
+
|
30 |
+
# Downsample
|
31 |
+
|
32 |
+
# 0
|
33 |
+
[reorg]
|
34 |
+
|
35 |
+
[convolutional]
|
36 |
+
batch_normalize=1
|
37 |
+
filters=64
|
38 |
+
size=3
|
39 |
+
stride=1
|
40 |
+
pad=1
|
41 |
+
activation=silu
|
42 |
+
|
43 |
+
|
44 |
+
# P2
|
45 |
+
|
46 |
+
# Downsample
|
47 |
+
|
48 |
+
[convolutional]
|
49 |
+
batch_normalize=1
|
50 |
+
filters=128
|
51 |
+
size=3
|
52 |
+
stride=2
|
53 |
+
pad=1
|
54 |
+
activation=silu
|
55 |
+
|
56 |
+
# Split
|
57 |
+
|
58 |
+
[convolutional]
|
59 |
+
batch_normalize=1
|
60 |
+
filters=64
|
61 |
+
size=1
|
62 |
+
stride=1
|
63 |
+
pad=1
|
64 |
+
activation=silu
|
65 |
+
|
66 |
+
[route]
|
67 |
+
layers = -2
|
68 |
+
|
69 |
+
[convolutional]
|
70 |
+
batch_normalize=1
|
71 |
+
filters=64
|
72 |
+
size=1
|
73 |
+
stride=1
|
74 |
+
pad=1
|
75 |
+
activation=silu
|
76 |
+
|
77 |
+
# Residual Block
|
78 |
+
|
79 |
+
[convolutional]
|
80 |
+
batch_normalize=1
|
81 |
+
filters=64
|
82 |
+
size=1
|
83 |
+
stride=1
|
84 |
+
pad=1
|
85 |
+
activation=silu
|
86 |
+
|
87 |
+
[convolutional]
|
88 |
+
batch_normalize=1
|
89 |
+
filters=64
|
90 |
+
size=3
|
91 |
+
stride=1
|
92 |
+
pad=1
|
93 |
+
activation=silu
|
94 |
+
|
95 |
+
[shortcut]
|
96 |
+
from=-3
|
97 |
+
activation=linear
|
98 |
+
|
99 |
+
[convolutional]
|
100 |
+
batch_normalize=1
|
101 |
+
filters=64
|
102 |
+
size=1
|
103 |
+
stride=1
|
104 |
+
pad=1
|
105 |
+
activation=silu
|
106 |
+
|
107 |
+
[convolutional]
|
108 |
+
batch_normalize=1
|
109 |
+
filters=64
|
110 |
+
size=3
|
111 |
+
stride=1
|
112 |
+
pad=1
|
113 |
+
activation=silu
|
114 |
+
|
115 |
+
[shortcut]
|
116 |
+
from=-3
|
117 |
+
activation=linear
|
118 |
+
|
119 |
+
[convolutional]
|
120 |
+
batch_normalize=1
|
121 |
+
filters=64
|
122 |
+
size=1
|
123 |
+
stride=1
|
124 |
+
pad=1
|
125 |
+
activation=silu
|
126 |
+
|
127 |
+
[convolutional]
|
128 |
+
batch_normalize=1
|
129 |
+
filters=64
|
130 |
+
size=3
|
131 |
+
stride=1
|
132 |
+
pad=1
|
133 |
+
activation=silu
|
134 |
+
|
135 |
+
[shortcut]
|
136 |
+
from=-3
|
137 |
+
activation=linear
|
138 |
+
|
139 |
+
# Transition first
|
140 |
+
#
|
141 |
+
#[convolutional]
|
142 |
+
#batch_normalize=1
|
143 |
+
#filters=64
|
144 |
+
#size=1
|
145 |
+
#stride=1
|
146 |
+
#pad=1
|
147 |
+
#activation=silu
|
148 |
+
|
149 |
+
# Merge [-1, -(3k+3)]
|
150 |
+
|
151 |
+
[route]
|
152 |
+
layers = -1,-12
|
153 |
+
|
154 |
+
# Transition last
|
155 |
+
|
156 |
+
# 16 (previous+6+3k)
|
157 |
+
[convolutional]
|
158 |
+
batch_normalize=1
|
159 |
+
filters=128
|
160 |
+
size=1
|
161 |
+
stride=1
|
162 |
+
pad=1
|
163 |
+
activation=silu
|
164 |
+
|
165 |
+
|
166 |
+
# P3
|
167 |
+
|
168 |
+
# Downsample
|
169 |
+
|
170 |
+
[convolutional]
|
171 |
+
batch_normalize=1
|
172 |
+
filters=256
|
173 |
+
size=3
|
174 |
+
stride=2
|
175 |
+
pad=1
|
176 |
+
activation=silu
|
177 |
+
|
178 |
+
# Split
|
179 |
+
|
180 |
+
[convolutional]
|
181 |
+
batch_normalize=1
|
182 |
+
filters=128
|
183 |
+
size=1
|
184 |
+
stride=1
|
185 |
+
pad=1
|
186 |
+
activation=silu
|
187 |
+
|
188 |
+
[route]
|
189 |
+
layers = -2
|
190 |
+
|
191 |
+
[convolutional]
|
192 |
+
batch_normalize=1
|
193 |
+
filters=128
|
194 |
+
size=1
|
195 |
+
stride=1
|
196 |
+
pad=1
|
197 |
+
activation=silu
|
198 |
+
|
199 |
+
# Residual Block
|
200 |
+
|
201 |
+
[convolutional]
|
202 |
+
batch_normalize=1
|
203 |
+
filters=128
|
204 |
+
size=1
|
205 |
+
stride=1
|
206 |
+
pad=1
|
207 |
+
activation=silu
|
208 |
+
|
209 |
+
[convolutional]
|
210 |
+
batch_normalize=1
|
211 |
+
filters=128
|
212 |
+
size=3
|
213 |
+
stride=1
|
214 |
+
pad=1
|
215 |
+
activation=silu
|
216 |
+
|
217 |
+
[shortcut]
|
218 |
+
from=-3
|
219 |
+
activation=linear
|
220 |
+
|
221 |
+
[convolutional]
|
222 |
+
batch_normalize=1
|
223 |
+
filters=128
|
224 |
+
size=1
|
225 |
+
stride=1
|
226 |
+
pad=1
|
227 |
+
activation=silu
|
228 |
+
|
229 |
+
[convolutional]
|
230 |
+
batch_normalize=1
|
231 |
+
filters=128
|
232 |
+
size=3
|
233 |
+
stride=1
|
234 |
+
pad=1
|
235 |
+
activation=silu
|
236 |
+
|
237 |
+
[shortcut]
|
238 |
+
from=-3
|
239 |
+
activation=linear
|
240 |
+
|
241 |
+
[convolutional]
|
242 |
+
batch_normalize=1
|
243 |
+
filters=128
|
244 |
+
size=1
|
245 |
+
stride=1
|
246 |
+
pad=1
|
247 |
+
activation=silu
|
248 |
+
|
249 |
+
[convolutional]
|
250 |
+
batch_normalize=1
|
251 |
+
filters=128
|
252 |
+
size=3
|
253 |
+
stride=1
|
254 |
+
pad=1
|
255 |
+
activation=silu
|
256 |
+
|
257 |
+
[shortcut]
|
258 |
+
from=-3
|
259 |
+
activation=linear
|
260 |
+
|
261 |
+
[convolutional]
|
262 |
+
batch_normalize=1
|
263 |
+
filters=128
|
264 |
+
size=1
|
265 |
+
stride=1
|
266 |
+
pad=1
|
267 |
+
activation=silu
|
268 |
+
|
269 |
+
[convolutional]
|
270 |
+
batch_normalize=1
|
271 |
+
filters=128
|
272 |
+
size=3
|
273 |
+
stride=1
|
274 |
+
pad=1
|
275 |
+
activation=silu
|
276 |
+
|
277 |
+
[shortcut]
|
278 |
+
from=-3
|
279 |
+
activation=linear
|
280 |
+
|
281 |
+
[convolutional]
|
282 |
+
batch_normalize=1
|
283 |
+
filters=128
|
284 |
+
size=1
|
285 |
+
stride=1
|
286 |
+
pad=1
|
287 |
+
activation=silu
|
288 |
+
|
289 |
+
[convolutional]
|
290 |
+
batch_normalize=1
|
291 |
+
filters=128
|
292 |
+
size=3
|
293 |
+
stride=1
|
294 |
+
pad=1
|
295 |
+
activation=silu
|
296 |
+
|
297 |
+
[shortcut]
|
298 |
+
from=-3
|
299 |
+
activation=linear
|
300 |
+
|
301 |
+
[convolutional]
|
302 |
+
batch_normalize=1
|
303 |
+
filters=128
|
304 |
+
size=1
|
305 |
+
stride=1
|
306 |
+
pad=1
|
307 |
+
activation=silu
|
308 |
+
|
309 |
+
[convolutional]
|
310 |
+
batch_normalize=1
|
311 |
+
filters=128
|
312 |
+
size=3
|
313 |
+
stride=1
|
314 |
+
pad=1
|
315 |
+
activation=silu
|
316 |
+
|
317 |
+
[shortcut]
|
318 |
+
from=-3
|
319 |
+
activation=linear
|
320 |
+
|
321 |
+
[convolutional]
|
322 |
+
batch_normalize=1
|
323 |
+
filters=128
|
324 |
+
size=1
|
325 |
+
stride=1
|
326 |
+
pad=1
|
327 |
+
activation=silu
|
328 |
+
|
329 |
+
[convolutional]
|
330 |
+
batch_normalize=1
|
331 |
+
filters=128
|
332 |
+
size=3
|
333 |
+
stride=1
|
334 |
+
pad=1
|
335 |
+
activation=silu
|
336 |
+
|
337 |
+
[shortcut]
|
338 |
+
from=-3
|
339 |
+
activation=linear
|
340 |
+
|
341 |
+
# Transition first
|
342 |
+
#
|
343 |
+
#[convolutional]
|
344 |
+
#batch_normalize=1
|
345 |
+
#filters=128
|
346 |
+
#size=1
|
347 |
+
#stride=1
|
348 |
+
#pad=1
|
349 |
+
#activation=silu
|
350 |
+
|
351 |
+
# Merge [-1, -(3k+3)]
|
352 |
+
|
353 |
+
[route]
|
354 |
+
layers = -1,-24
|
355 |
+
|
356 |
+
# Transition last
|
357 |
+
|
358 |
+
# 43 (previous+6+3k)
|
359 |
+
[convolutional]
|
360 |
+
batch_normalize=1
|
361 |
+
filters=256
|
362 |
+
size=1
|
363 |
+
stride=1
|
364 |
+
pad=1
|
365 |
+
activation=silu
|
366 |
+
|
367 |
+
|
368 |
+
# P4
|
369 |
+
|
370 |
+
# Downsample
|
371 |
+
|
372 |
+
[convolutional]
|
373 |
+
batch_normalize=1
|
374 |
+
filters=512
|
375 |
+
size=3
|
376 |
+
stride=2
|
377 |
+
pad=1
|
378 |
+
activation=silu
|
379 |
+
|
380 |
+
# Split
|
381 |
+
|
382 |
+
[convolutional]
|
383 |
+
batch_normalize=1
|
384 |
+
filters=256
|
385 |
+
size=1
|
386 |
+
stride=1
|
387 |
+
pad=1
|
388 |
+
activation=silu
|
389 |
+
|
390 |
+
[route]
|
391 |
+
layers = -2
|
392 |
+
|
393 |
+
[convolutional]
|
394 |
+
batch_normalize=1
|
395 |
+
filters=256
|
396 |
+
size=1
|
397 |
+
stride=1
|
398 |
+
pad=1
|
399 |
+
activation=silu
|
400 |
+
|
401 |
+
# Residual Block
|
402 |
+
|
403 |
+
[convolutional]
|
404 |
+
batch_normalize=1
|
405 |
+
filters=256
|
406 |
+
size=1
|
407 |
+
stride=1
|
408 |
+
pad=1
|
409 |
+
activation=silu
|
410 |
+
|
411 |
+
[convolutional]
|
412 |
+
batch_normalize=1
|
413 |
+
filters=256
|
414 |
+
size=3
|
415 |
+
stride=1
|
416 |
+
pad=1
|
417 |
+
activation=silu
|
418 |
+
|
419 |
+
[shortcut]
|
420 |
+
from=-3
|
421 |
+
activation=linear
|
422 |
+
|
423 |
+
[convolutional]
|
424 |
+
batch_normalize=1
|
425 |
+
filters=256
|
426 |
+
size=1
|
427 |
+
stride=1
|
428 |
+
pad=1
|
429 |
+
activation=silu
|
430 |
+
|
431 |
+
[convolutional]
|
432 |
+
batch_normalize=1
|
433 |
+
filters=256
|
434 |
+
size=3
|
435 |
+
stride=1
|
436 |
+
pad=1
|
437 |
+
activation=silu
|
438 |
+
|
439 |
+
[shortcut]
|
440 |
+
from=-3
|
441 |
+
activation=linear
|
442 |
+
|
443 |
+
[convolutional]
|
444 |
+
batch_normalize=1
|
445 |
+
filters=256
|
446 |
+
size=1
|
447 |
+
stride=1
|
448 |
+
pad=1
|
449 |
+
activation=silu
|
450 |
+
|
451 |
+
[convolutional]
|
452 |
+
batch_normalize=1
|
453 |
+
filters=256
|
454 |
+
size=3
|
455 |
+
stride=1
|
456 |
+
pad=1
|
457 |
+
activation=silu
|
458 |
+
|
459 |
+
[shortcut]
|
460 |
+
from=-3
|
461 |
+
activation=linear
|
462 |
+
|
463 |
+
[convolutional]
|
464 |
+
batch_normalize=1
|
465 |
+
filters=256
|
466 |
+
size=1
|
467 |
+
stride=1
|
468 |
+
pad=1
|
469 |
+
activation=silu
|
470 |
+
|
471 |
+
[convolutional]
|
472 |
+
batch_normalize=1
|
473 |
+
filters=256
|
474 |
+
size=3
|
475 |
+
stride=1
|
476 |
+
pad=1
|
477 |
+
activation=silu
|
478 |
+
|
479 |
+
[shortcut]
|
480 |
+
from=-3
|
481 |
+
activation=linear
|
482 |
+
|
483 |
+
[convolutional]
|
484 |
+
batch_normalize=1
|
485 |
+
filters=256
|
486 |
+
size=1
|
487 |
+
stride=1
|
488 |
+
pad=1
|
489 |
+
activation=silu
|
490 |
+
|
491 |
+
[convolutional]
|
492 |
+
batch_normalize=1
|
493 |
+
filters=256
|
494 |
+
size=3
|
495 |
+
stride=1
|
496 |
+
pad=1
|
497 |
+
activation=silu
|
498 |
+
|
499 |
+
[shortcut]
|
500 |
+
from=-3
|
501 |
+
activation=linear
|
502 |
+
|
503 |
+
[convolutional]
|
504 |
+
batch_normalize=1
|
505 |
+
filters=256
|
506 |
+
size=1
|
507 |
+
stride=1
|
508 |
+
pad=1
|
509 |
+
activation=silu
|
510 |
+
|
511 |
+
[convolutional]
|
512 |
+
batch_normalize=1
|
513 |
+
filters=256
|
514 |
+
size=3
|
515 |
+
stride=1
|
516 |
+
pad=1
|
517 |
+
activation=silu
|
518 |
+
|
519 |
+
[shortcut]
|
520 |
+
from=-3
|
521 |
+
activation=linear
|
522 |
+
|
523 |
+
[convolutional]
|
524 |
+
batch_normalize=1
|
525 |
+
filters=256
|
526 |
+
size=1
|
527 |
+
stride=1
|
528 |
+
pad=1
|
529 |
+
activation=silu
|
530 |
+
|
531 |
+
[convolutional]
|
532 |
+
batch_normalize=1
|
533 |
+
filters=256
|
534 |
+
size=3
|
535 |
+
stride=1
|
536 |
+
pad=1
|
537 |
+
activation=silu
|
538 |
+
|
539 |
+
[shortcut]
|
540 |
+
from=-3
|
541 |
+
activation=linear
|
542 |
+
|
543 |
+
# Transition first
|
544 |
+
#
|
545 |
+
#[convolutional]
|
546 |
+
#batch_normalize=1
|
547 |
+
#filters=256
|
548 |
+
#size=1
|
549 |
+
#stride=1
|
550 |
+
#pad=1
|
551 |
+
#activation=silu
|
552 |
+
|
553 |
+
# Merge [-1, -(3k+3)]
|
554 |
+
|
555 |
+
[route]
|
556 |
+
layers = -1,-24
|
557 |
+
|
558 |
+
# Transition last
|
559 |
+
|
560 |
+
# 70 (previous+6+3k)
|
561 |
+
[convolutional]
|
562 |
+
batch_normalize=1
|
563 |
+
filters=512
|
564 |
+
size=1
|
565 |
+
stride=1
|
566 |
+
pad=1
|
567 |
+
activation=silu
|
568 |
+
|
569 |
+
|
570 |
+
# P5
|
571 |
+
|
572 |
+
# Downsample
|
573 |
+
|
574 |
+
[convolutional]
|
575 |
+
batch_normalize=1
|
576 |
+
filters=768
|
577 |
+
size=3
|
578 |
+
stride=2
|
579 |
+
pad=1
|
580 |
+
activation=silu
|
581 |
+
|
582 |
+
# Split
|
583 |
+
|
584 |
+
[convolutional]
|
585 |
+
batch_normalize=1
|
586 |
+
filters=384
|
587 |
+
size=1
|
588 |
+
stride=1
|
589 |
+
pad=1
|
590 |
+
activation=silu
|
591 |
+
|
592 |
+
[route]
|
593 |
+
layers = -2
|
594 |
+
|
595 |
+
[convolutional]
|
596 |
+
batch_normalize=1
|
597 |
+
filters=384
|
598 |
+
size=1
|
599 |
+
stride=1
|
600 |
+
pad=1
|
601 |
+
activation=silu
|
602 |
+
|
603 |
+
# Residual Block
|
604 |
+
|
605 |
+
[convolutional]
|
606 |
+
batch_normalize=1
|
607 |
+
filters=384
|
608 |
+
size=1
|
609 |
+
stride=1
|
610 |
+
pad=1
|
611 |
+
activation=silu
|
612 |
+
|
613 |
+
[convolutional]
|
614 |
+
batch_normalize=1
|
615 |
+
filters=384
|
616 |
+
size=3
|
617 |
+
stride=1
|
618 |
+
pad=1
|
619 |
+
activation=silu
|
620 |
+
|
621 |
+
[shortcut]
|
622 |
+
from=-3
|
623 |
+
activation=linear
|
624 |
+
|
625 |
+
[convolutional]
|
626 |
+
batch_normalize=1
|
627 |
+
filters=384
|
628 |
+
size=1
|
629 |
+
stride=1
|
630 |
+
pad=1
|
631 |
+
activation=silu
|
632 |
+
|
633 |
+
[convolutional]
|
634 |
+
batch_normalize=1
|
635 |
+
filters=384
|
636 |
+
size=3
|
637 |
+
stride=1
|
638 |
+
pad=1
|
639 |
+
activation=silu
|
640 |
+
|
641 |
+
[shortcut]
|
642 |
+
from=-3
|
643 |
+
activation=linear
|
644 |
+
|
645 |
+
[convolutional]
|
646 |
+
batch_normalize=1
|
647 |
+
filters=384
|
648 |
+
size=1
|
649 |
+
stride=1
|
650 |
+
pad=1
|
651 |
+
activation=silu
|
652 |
+
|
653 |
+
[convolutional]
|
654 |
+
batch_normalize=1
|
655 |
+
filters=384
|
656 |
+
size=3
|
657 |
+
stride=1
|
658 |
+
pad=1
|
659 |
+
activation=silu
|
660 |
+
|
661 |
+
[shortcut]
|
662 |
+
from=-3
|
663 |
+
activation=linear
|
664 |
+
|
665 |
+
# Transition first
|
666 |
+
#
|
667 |
+
#[convolutional]
|
668 |
+
#batch_normalize=1
|
669 |
+
#filters=384
|
670 |
+
#size=1
|
671 |
+
#stride=1
|
672 |
+
#pad=1
|
673 |
+
#activation=silu
|
674 |
+
|
675 |
+
# Merge [-1, -(3k+3)]
|
676 |
+
|
677 |
+
[route]
|
678 |
+
layers = -1,-12
|
679 |
+
|
680 |
+
# Transition last
|
681 |
+
|
682 |
+
# 85 (previous+6+3k)
|
683 |
+
[convolutional]
|
684 |
+
batch_normalize=1
|
685 |
+
filters=768
|
686 |
+
size=1
|
687 |
+
stride=1
|
688 |
+
pad=1
|
689 |
+
activation=silu
|
690 |
+
|
691 |
+
|
692 |
+
# P6
|
693 |
+
|
694 |
+
# Downsample
|
695 |
+
|
696 |
+
[convolutional]
|
697 |
+
batch_normalize=1
|
698 |
+
filters=1024
|
699 |
+
size=3
|
700 |
+
stride=2
|
701 |
+
pad=1
|
702 |
+
activation=silu
|
703 |
+
|
704 |
+
# Split
|
705 |
+
|
706 |
+
[convolutional]
|
707 |
+
batch_normalize=1
|
708 |
+
filters=512
|
709 |
+
size=1
|
710 |
+
stride=1
|
711 |
+
pad=1
|
712 |
+
activation=silu
|
713 |
+
|
714 |
+
[route]
|
715 |
+
layers = -2
|
716 |
+
|
717 |
+
[convolutional]
|
718 |
+
batch_normalize=1
|
719 |
+
filters=512
|
720 |
+
size=1
|
721 |
+
stride=1
|
722 |
+
pad=1
|
723 |
+
activation=silu
|
724 |
+
|
725 |
+
# Residual Block
|
726 |
+
|
727 |
+
[convolutional]
|
728 |
+
batch_normalize=1
|
729 |
+
filters=512
|
730 |
+
size=1
|
731 |
+
stride=1
|
732 |
+
pad=1
|
733 |
+
activation=silu
|
734 |
+
|
735 |
+
[convolutional]
|
736 |
+
batch_normalize=1
|
737 |
+
filters=512
|
738 |
+
size=3
|
739 |
+
stride=1
|
740 |
+
pad=1
|
741 |
+
activation=silu
|
742 |
+
|
743 |
+
[shortcut]
|
744 |
+
from=-3
|
745 |
+
activation=linear
|
746 |
+
|
747 |
+
[convolutional]
|
748 |
+
batch_normalize=1
|
749 |
+
filters=512
|
750 |
+
size=1
|
751 |
+
stride=1
|
752 |
+
pad=1
|
753 |
+
activation=silu
|
754 |
+
|
755 |
+
[convolutional]
|
756 |
+
batch_normalize=1
|
757 |
+
filters=512
|
758 |
+
size=3
|
759 |
+
stride=1
|
760 |
+
pad=1
|
761 |
+
activation=silu
|
762 |
+
|
763 |
+
[shortcut]
|
764 |
+
from=-3
|
765 |
+
activation=linear
|
766 |
+
|
767 |
+
[convolutional]
|
768 |
+
batch_normalize=1
|
769 |
+
filters=512
|
770 |
+
size=1
|
771 |
+
stride=1
|
772 |
+
pad=1
|
773 |
+
activation=silu
|
774 |
+
|
775 |
+
[convolutional]
|
776 |
+
batch_normalize=1
|
777 |
+
filters=512
|
778 |
+
size=3
|
779 |
+
stride=1
|
780 |
+
pad=1
|
781 |
+
activation=silu
|
782 |
+
|
783 |
+
[shortcut]
|
784 |
+
from=-3
|
785 |
+
activation=linear
|
786 |
+
|
787 |
+
# Transition first
|
788 |
+
#
|
789 |
+
#[convolutional]
|
790 |
+
#batch_normalize=1
|
791 |
+
#filters=512
|
792 |
+
#size=1
|
793 |
+
#stride=1
|
794 |
+
#pad=1
|
795 |
+
#activation=silu
|
796 |
+
|
797 |
+
# Merge [-1, -(3k+3)]
|
798 |
+
|
799 |
+
[route]
|
800 |
+
layers = -1,-12
|
801 |
+
|
802 |
+
# Transition last
|
803 |
+
|
804 |
+
# 100 (previous+6+3k)
|
805 |
+
[convolutional]
|
806 |
+
batch_normalize=1
|
807 |
+
filters=1024
|
808 |
+
size=1
|
809 |
+
stride=1
|
810 |
+
pad=1
|
811 |
+
activation=silu
|
812 |
+
|
813 |
+
# ============ End of Backbone ============ #
|
814 |
+
|
815 |
+
# ============ Neck ============ #
|
816 |
+
|
817 |
+
# CSPSPP
|
818 |
+
|
819 |
+
[convolutional]
|
820 |
+
batch_normalize=1
|
821 |
+
filters=512
|
822 |
+
size=1
|
823 |
+
stride=1
|
824 |
+
pad=1
|
825 |
+
activation=silu
|
826 |
+
|
827 |
+
[route]
|
828 |
+
layers = -2
|
829 |
+
|
830 |
+
[convolutional]
|
831 |
+
batch_normalize=1
|
832 |
+
filters=512
|
833 |
+
size=1
|
834 |
+
stride=1
|
835 |
+
pad=1
|
836 |
+
activation=silu
|
837 |
+
|
838 |
+
[convolutional]
|
839 |
+
batch_normalize=1
|
840 |
+
size=3
|
841 |
+
stride=1
|
842 |
+
pad=1
|
843 |
+
filters=512
|
844 |
+
activation=silu
|
845 |
+
|
846 |
+
[convolutional]
|
847 |
+
batch_normalize=1
|
848 |
+
filters=512
|
849 |
+
size=1
|
850 |
+
stride=1
|
851 |
+
pad=1
|
852 |
+
activation=silu
|
853 |
+
|
854 |
+
### SPP ###
|
855 |
+
[maxpool]
|
856 |
+
stride=1
|
857 |
+
size=5
|
858 |
+
|
859 |
+
[route]
|
860 |
+
layers=-2
|
861 |
+
|
862 |
+
[maxpool]
|
863 |
+
stride=1
|
864 |
+
size=9
|
865 |
+
|
866 |
+
[route]
|
867 |
+
layers=-4
|
868 |
+
|
869 |
+
[maxpool]
|
870 |
+
stride=1
|
871 |
+
size=13
|
872 |
+
|
873 |
+
[route]
|
874 |
+
layers=-1,-3,-5,-6
|
875 |
+
### End SPP ###
|
876 |
+
|
877 |
+
[convolutional]
|
878 |
+
batch_normalize=1
|
879 |
+
filters=512
|
880 |
+
size=1
|
881 |
+
stride=1
|
882 |
+
pad=1
|
883 |
+
activation=silu
|
884 |
+
|
885 |
+
[convolutional]
|
886 |
+
batch_normalize=1
|
887 |
+
size=3
|
888 |
+
stride=1
|
889 |
+
pad=1
|
890 |
+
filters=512
|
891 |
+
activation=silu
|
892 |
+
|
893 |
+
[route]
|
894 |
+
layers = -1, -13
|
895 |
+
|
896 |
+
# 115 (previous+6+5+2k)
|
897 |
+
[convolutional]
|
898 |
+
batch_normalize=1
|
899 |
+
filters=512
|
900 |
+
size=1
|
901 |
+
stride=1
|
902 |
+
pad=1
|
903 |
+
activation=silu
|
904 |
+
|
905 |
+
# End of CSPSPP
|
906 |
+
|
907 |
+
|
908 |
+
# FPN-5
|
909 |
+
|
910 |
+
[convolutional]
|
911 |
+
batch_normalize=1
|
912 |
+
filters=384
|
913 |
+
size=1
|
914 |
+
stride=1
|
915 |
+
pad=1
|
916 |
+
activation=silu
|
917 |
+
|
918 |
+
[upsample]
|
919 |
+
stride=2
|
920 |
+
|
921 |
+
[route]
|
922 |
+
layers = 85
|
923 |
+
|
924 |
+
[convolutional]
|
925 |
+
batch_normalize=1
|
926 |
+
filters=384
|
927 |
+
size=1
|
928 |
+
stride=1
|
929 |
+
pad=1
|
930 |
+
activation=silu
|
931 |
+
|
932 |
+
[route]
|
933 |
+
layers = -1, -3
|
934 |
+
|
935 |
+
[convolutional]
|
936 |
+
batch_normalize=1
|
937 |
+
filters=384
|
938 |
+
size=1
|
939 |
+
stride=1
|
940 |
+
pad=1
|
941 |
+
activation=silu
|
942 |
+
|
943 |
+
# Split
|
944 |
+
|
945 |
+
[convolutional]
|
946 |
+
batch_normalize=1
|
947 |
+
filters=384
|
948 |
+
size=1
|
949 |
+
stride=1
|
950 |
+
pad=1
|
951 |
+
activation=silu
|
952 |
+
|
953 |
+
[route]
|
954 |
+
layers = -2
|
955 |
+
|
956 |
+
# Plain Block
|
957 |
+
|
958 |
+
[convolutional]
|
959 |
+
batch_normalize=1
|
960 |
+
filters=384
|
961 |
+
size=1
|
962 |
+
stride=1
|
963 |
+
pad=1
|
964 |
+
activation=silu
|
965 |
+
|
966 |
+
[convolutional]
|
967 |
+
batch_normalize=1
|
968 |
+
size=3
|
969 |
+
stride=1
|
970 |
+
pad=1
|
971 |
+
filters=384
|
972 |
+
activation=silu
|
973 |
+
|
974 |
+
[convolutional]
|
975 |
+
batch_normalize=1
|
976 |
+
filters=384
|
977 |
+
size=1
|
978 |
+
stride=1
|
979 |
+
pad=1
|
980 |
+
activation=silu
|
981 |
+
|
982 |
+
[convolutional]
|
983 |
+
batch_normalize=1
|
984 |
+
size=3
|
985 |
+
stride=1
|
986 |
+
pad=1
|
987 |
+
filters=384
|
988 |
+
activation=silu
|
989 |
+
|
990 |
+
[convolutional]
|
991 |
+
batch_normalize=1
|
992 |
+
filters=384
|
993 |
+
size=1
|
994 |
+
stride=1
|
995 |
+
pad=1
|
996 |
+
activation=silu
|
997 |
+
|
998 |
+
[convolutional]
|
999 |
+
batch_normalize=1
|
1000 |
+
size=3
|
1001 |
+
stride=1
|
1002 |
+
pad=1
|
1003 |
+
filters=384
|
1004 |
+
activation=silu
|
1005 |
+
|
1006 |
+
# Merge [-1, -(2k+2)]
|
1007 |
+
|
1008 |
+
[route]
|
1009 |
+
layers = -1, -8
|
1010 |
+
|
1011 |
+
# Transition last
|
1012 |
+
|
1013 |
+
# 131 (previous+6+4+2k)
|
1014 |
+
[convolutional]
|
1015 |
+
batch_normalize=1
|
1016 |
+
filters=384
|
1017 |
+
size=1
|
1018 |
+
stride=1
|
1019 |
+
pad=1
|
1020 |
+
activation=silu
|
1021 |
+
|
1022 |
+
|
1023 |
+
# FPN-4
|
1024 |
+
|
1025 |
+
[convolutional]
|
1026 |
+
batch_normalize=1
|
1027 |
+
filters=256
|
1028 |
+
size=1
|
1029 |
+
stride=1
|
1030 |
+
pad=1
|
1031 |
+
activation=silu
|
1032 |
+
|
1033 |
+
[upsample]
|
1034 |
+
stride=2
|
1035 |
+
|
1036 |
+
[route]
|
1037 |
+
layers = 70
|
1038 |
+
|
1039 |
+
[convolutional]
|
1040 |
+
batch_normalize=1
|
1041 |
+
filters=256
|
1042 |
+
size=1
|
1043 |
+
stride=1
|
1044 |
+
pad=1
|
1045 |
+
activation=silu
|
1046 |
+
|
1047 |
+
[route]
|
1048 |
+
layers = -1, -3
|
1049 |
+
|
1050 |
+
[convolutional]
|
1051 |
+
batch_normalize=1
|
1052 |
+
filters=256
|
1053 |
+
size=1
|
1054 |
+
stride=1
|
1055 |
+
pad=1
|
1056 |
+
activation=silu
|
1057 |
+
|
1058 |
+
# Split
|
1059 |
+
|
1060 |
+
[convolutional]
|
1061 |
+
batch_normalize=1
|
1062 |
+
filters=256
|
1063 |
+
size=1
|
1064 |
+
stride=1
|
1065 |
+
pad=1
|
1066 |
+
activation=silu
|
1067 |
+
|
1068 |
+
[route]
|
1069 |
+
layers = -2
|
1070 |
+
|
1071 |
+
# Plain Block
|
1072 |
+
|
1073 |
+
[convolutional]
|
1074 |
+
batch_normalize=1
|
1075 |
+
filters=256
|
1076 |
+
size=1
|
1077 |
+
stride=1
|
1078 |
+
pad=1
|
1079 |
+
activation=silu
|
1080 |
+
|
1081 |
+
[convolutional]
|
1082 |
+
batch_normalize=1
|
1083 |
+
size=3
|
1084 |
+
stride=1
|
1085 |
+
pad=1
|
1086 |
+
filters=256
|
1087 |
+
activation=silu
|
1088 |
+
|
1089 |
+
[convolutional]
|
1090 |
+
batch_normalize=1
|
1091 |
+
filters=256
|
1092 |
+
size=1
|
1093 |
+
stride=1
|
1094 |
+
pad=1
|
1095 |
+
activation=silu
|
1096 |
+
|
1097 |
+
[convolutional]
|
1098 |
+
batch_normalize=1
|
1099 |
+
size=3
|
1100 |
+
stride=1
|
1101 |
+
pad=1
|
1102 |
+
filters=256
|
1103 |
+
activation=silu
|
1104 |
+
|
1105 |
+
[convolutional]
|
1106 |
+
batch_normalize=1
|
1107 |
+
filters=256
|
1108 |
+
size=1
|
1109 |
+
stride=1
|
1110 |
+
pad=1
|
1111 |
+
activation=silu
|
1112 |
+
|
1113 |
+
[convolutional]
|
1114 |
+
batch_normalize=1
|
1115 |
+
size=3
|
1116 |
+
stride=1
|
1117 |
+
pad=1
|
1118 |
+
filters=256
|
1119 |
+
activation=silu
|
1120 |
+
|
1121 |
+
# Merge [-1, -(2k+2)]
|
1122 |
+
|
1123 |
+
[route]
|
1124 |
+
layers = -1, -8
|
1125 |
+
|
1126 |
+
# Transition last
|
1127 |
+
|
1128 |
+
# 147 (previous+6+4+2k)
|
1129 |
+
[convolutional]
|
1130 |
+
batch_normalize=1
|
1131 |
+
filters=256
|
1132 |
+
size=1
|
1133 |
+
stride=1
|
1134 |
+
pad=1
|
1135 |
+
activation=silu
|
1136 |
+
|
1137 |
+
|
1138 |
+
# FPN-3
|
1139 |
+
|
1140 |
+
[convolutional]
|
1141 |
+
batch_normalize=1
|
1142 |
+
filters=128
|
1143 |
+
size=1
|
1144 |
+
stride=1
|
1145 |
+
pad=1
|
1146 |
+
activation=silu
|
1147 |
+
|
1148 |
+
[upsample]
|
1149 |
+
stride=2
|
1150 |
+
|
1151 |
+
[route]
|
1152 |
+
layers = 43
|
1153 |
+
|
1154 |
+
[convolutional]
|
1155 |
+
batch_normalize=1
|
1156 |
+
filters=128
|
1157 |
+
size=1
|
1158 |
+
stride=1
|
1159 |
+
pad=1
|
1160 |
+
activation=silu
|
1161 |
+
|
1162 |
+
[route]
|
1163 |
+
layers = -1, -3
|
1164 |
+
|
1165 |
+
[convolutional]
|
1166 |
+
batch_normalize=1
|
1167 |
+
filters=128
|
1168 |
+
size=1
|
1169 |
+
stride=1
|
1170 |
+
pad=1
|
1171 |
+
activation=silu
|
1172 |
+
|
1173 |
+
# Split
|
1174 |
+
|
1175 |
+
[convolutional]
|
1176 |
+
batch_normalize=1
|
1177 |
+
filters=128
|
1178 |
+
size=1
|
1179 |
+
stride=1
|
1180 |
+
pad=1
|
1181 |
+
activation=silu
|
1182 |
+
|
1183 |
+
[route]
|
1184 |
+
layers = -2
|
1185 |
+
|
1186 |
+
# Plain Block
|
1187 |
+
|
1188 |
+
[convolutional]
|
1189 |
+
batch_normalize=1
|
1190 |
+
filters=128
|
1191 |
+
size=1
|
1192 |
+
stride=1
|
1193 |
+
pad=1
|
1194 |
+
activation=silu
|
1195 |
+
|
1196 |
+
[convolutional]
|
1197 |
+
batch_normalize=1
|
1198 |
+
size=3
|
1199 |
+
stride=1
|
1200 |
+
pad=1
|
1201 |
+
filters=128
|
1202 |
+
activation=silu
|
1203 |
+
|
1204 |
+
[convolutional]
|
1205 |
+
batch_normalize=1
|
1206 |
+
filters=128
|
1207 |
+
size=1
|
1208 |
+
stride=1
|
1209 |
+
pad=1
|
1210 |
+
activation=silu
|
1211 |
+
|
1212 |
+
[convolutional]
|
1213 |
+
batch_normalize=1
|
1214 |
+
size=3
|
1215 |
+
stride=1
|
1216 |
+
pad=1
|
1217 |
+
filters=128
|
1218 |
+
activation=silu
|
1219 |
+
|
1220 |
+
[convolutional]
|
1221 |
+
batch_normalize=1
|
1222 |
+
filters=128
|
1223 |
+
size=1
|
1224 |
+
stride=1
|
1225 |
+
pad=1
|
1226 |
+
activation=silu
|
1227 |
+
|
1228 |
+
[convolutional]
|
1229 |
+
batch_normalize=1
|
1230 |
+
size=3
|
1231 |
+
stride=1
|
1232 |
+
pad=1
|
1233 |
+
filters=128
|
1234 |
+
activation=silu
|
1235 |
+
|
1236 |
+
# Merge [-1, -(2k+2)]
|
1237 |
+
|
1238 |
+
[route]
|
1239 |
+
layers = -1, -8
|
1240 |
+
|
1241 |
+
# Transition last
|
1242 |
+
|
1243 |
+
# 163 (previous+6+4+2k)
|
1244 |
+
[convolutional]
|
1245 |
+
batch_normalize=1
|
1246 |
+
filters=128
|
1247 |
+
size=1
|
1248 |
+
stride=1
|
1249 |
+
pad=1
|
1250 |
+
activation=silu
|
1251 |
+
|
1252 |
+
|
1253 |
+
# PAN-4
|
1254 |
+
|
1255 |
+
[convolutional]
|
1256 |
+
batch_normalize=1
|
1257 |
+
size=3
|
1258 |
+
stride=2
|
1259 |
+
pad=1
|
1260 |
+
filters=256
|
1261 |
+
activation=silu
|
1262 |
+
|
1263 |
+
[route]
|
1264 |
+
layers = -1, 147
|
1265 |
+
|
1266 |
+
[convolutional]
|
1267 |
+
batch_normalize=1
|
1268 |
+
filters=256
|
1269 |
+
size=1
|
1270 |
+
stride=1
|
1271 |
+
pad=1
|
1272 |
+
activation=silu
|
1273 |
+
|
1274 |
+
# Split
|
1275 |
+
|
1276 |
+
[convolutional]
|
1277 |
+
batch_normalize=1
|
1278 |
+
filters=256
|
1279 |
+
size=1
|
1280 |
+
stride=1
|
1281 |
+
pad=1
|
1282 |
+
activation=silu
|
1283 |
+
|
1284 |
+
[route]
|
1285 |
+
layers = -2
|
1286 |
+
|
1287 |
+
# Plain Block
|
1288 |
+
|
1289 |
+
[convolutional]
|
1290 |
+
batch_normalize=1
|
1291 |
+
filters=256
|
1292 |
+
size=1
|
1293 |
+
stride=1
|
1294 |
+
pad=1
|
1295 |
+
activation=silu
|
1296 |
+
|
1297 |
+
[convolutional]
|
1298 |
+
batch_normalize=1
|
1299 |
+
size=3
|
1300 |
+
stride=1
|
1301 |
+
pad=1
|
1302 |
+
filters=256
|
1303 |
+
activation=silu
|
1304 |
+
|
1305 |
+
[convolutional]
|
1306 |
+
batch_normalize=1
|
1307 |
+
filters=256
|
1308 |
+
size=1
|
1309 |
+
stride=1
|
1310 |
+
pad=1
|
1311 |
+
activation=silu
|
1312 |
+
|
1313 |
+
[convolutional]
|
1314 |
+
batch_normalize=1
|
1315 |
+
size=3
|
1316 |
+
stride=1
|
1317 |
+
pad=1
|
1318 |
+
filters=256
|
1319 |
+
activation=silu
|
1320 |
+
|
1321 |
+
[convolutional]
|
1322 |
+
batch_normalize=1
|
1323 |
+
filters=256
|
1324 |
+
size=1
|
1325 |
+
stride=1
|
1326 |
+
pad=1
|
1327 |
+
activation=silu
|
1328 |
+
|
1329 |
+
[convolutional]
|
1330 |
+
batch_normalize=1
|
1331 |
+
size=3
|
1332 |
+
stride=1
|
1333 |
+
pad=1
|
1334 |
+
filters=256
|
1335 |
+
activation=silu
|
1336 |
+
|
1337 |
+
[route]
|
1338 |
+
layers = -1,-8
|
1339 |
+
|
1340 |
+
# Transition last
|
1341 |
+
|
1342 |
+
# 176 (previous+3+4+2k)
|
1343 |
+
[convolutional]
|
1344 |
+
batch_normalize=1
|
1345 |
+
filters=256
|
1346 |
+
size=1
|
1347 |
+
stride=1
|
1348 |
+
pad=1
|
1349 |
+
activation=silu
|
1350 |
+
|
1351 |
+
|
1352 |
+
# PAN-5
|
1353 |
+
|
1354 |
+
[convolutional]
|
1355 |
+
batch_normalize=1
|
1356 |
+
size=3
|
1357 |
+
stride=2
|
1358 |
+
pad=1
|
1359 |
+
filters=384
|
1360 |
+
activation=silu
|
1361 |
+
|
1362 |
+
[route]
|
1363 |
+
layers = -1, 131
|
1364 |
+
|
1365 |
+
[convolutional]
|
1366 |
+
batch_normalize=1
|
1367 |
+
filters=384
|
1368 |
+
size=1
|
1369 |
+
stride=1
|
1370 |
+
pad=1
|
1371 |
+
activation=silu
|
1372 |
+
|
1373 |
+
# Split
|
1374 |
+
|
1375 |
+
[convolutional]
|
1376 |
+
batch_normalize=1
|
1377 |
+
filters=384
|
1378 |
+
size=1
|
1379 |
+
stride=1
|
1380 |
+
pad=1
|
1381 |
+
activation=silu
|
1382 |
+
|
1383 |
+
[route]
|
1384 |
+
layers = -2
|
1385 |
+
|
1386 |
+
# Plain Block
|
1387 |
+
|
1388 |
+
[convolutional]
|
1389 |
+
batch_normalize=1
|
1390 |
+
filters=384
|
1391 |
+
size=1
|
1392 |
+
stride=1
|
1393 |
+
pad=1
|
1394 |
+
activation=silu
|
1395 |
+
|
1396 |
+
[convolutional]
|
1397 |
+
batch_normalize=1
|
1398 |
+
size=3
|
1399 |
+
stride=1
|
1400 |
+
pad=1
|
1401 |
+
filters=384
|
1402 |
+
activation=silu
|
1403 |
+
|
1404 |
+
[convolutional]
|
1405 |
+
batch_normalize=1
|
1406 |
+
filters=384
|
1407 |
+
size=1
|
1408 |
+
stride=1
|
1409 |
+
pad=1
|
1410 |
+
activation=silu
|
1411 |
+
|
1412 |
+
[convolutional]
|
1413 |
+
batch_normalize=1
|
1414 |
+
size=3
|
1415 |
+
stride=1
|
1416 |
+
pad=1
|
1417 |
+
filters=384
|
1418 |
+
activation=silu
|
1419 |
+
|
1420 |
+
[convolutional]
|
1421 |
+
batch_normalize=1
|
1422 |
+
filters=384
|
1423 |
+
size=1
|
1424 |
+
stride=1
|
1425 |
+
pad=1
|
1426 |
+
activation=silu
|
1427 |
+
|
1428 |
+
[convolutional]
|
1429 |
+
batch_normalize=1
|
1430 |
+
size=3
|
1431 |
+
stride=1
|
1432 |
+
pad=1
|
1433 |
+
filters=384
|
1434 |
+
activation=silu
|
1435 |
+
|
1436 |
+
[route]
|
1437 |
+
layers = -1,-8
|
1438 |
+
|
1439 |
+
# Transition last
|
1440 |
+
|
1441 |
+
# 189 (previous+3+4+2k)
|
1442 |
+
[convolutional]
|
1443 |
+
batch_normalize=1
|
1444 |
+
filters=384
|
1445 |
+
size=1
|
1446 |
+
stride=1
|
1447 |
+
pad=1
|
1448 |
+
activation=silu
|
1449 |
+
|
1450 |
+
|
1451 |
+
# PAN-6
|
1452 |
+
|
1453 |
+
[convolutional]
|
1454 |
+
batch_normalize=1
|
1455 |
+
size=3
|
1456 |
+
stride=2
|
1457 |
+
pad=1
|
1458 |
+
filters=512
|
1459 |
+
activation=silu
|
1460 |
+
|
1461 |
+
[route]
|
1462 |
+
layers = -1, 115
|
1463 |
+
|
1464 |
+
[convolutional]
|
1465 |
+
batch_normalize=1
|
1466 |
+
filters=512
|
1467 |
+
size=1
|
1468 |
+
stride=1
|
1469 |
+
pad=1
|
1470 |
+
activation=silu
|
1471 |
+
|
1472 |
+
# Split
|
1473 |
+
|
1474 |
+
[convolutional]
|
1475 |
+
batch_normalize=1
|
1476 |
+
filters=512
|
1477 |
+
size=1
|
1478 |
+
stride=1
|
1479 |
+
pad=1
|
1480 |
+
activation=silu
|
1481 |
+
|
1482 |
+
[route]
|
1483 |
+
layers = -2
|
1484 |
+
|
1485 |
+
# Plain Block
|
1486 |
+
|
1487 |
+
[convolutional]
|
1488 |
+
batch_normalize=1
|
1489 |
+
filters=512
|
1490 |
+
size=1
|
1491 |
+
stride=1
|
1492 |
+
pad=1
|
1493 |
+
activation=silu
|
1494 |
+
|
1495 |
+
[convolutional]
|
1496 |
+
batch_normalize=1
|
1497 |
+
size=3
|
1498 |
+
stride=1
|
1499 |
+
pad=1
|
1500 |
+
filters=512
|
1501 |
+
activation=silu
|
1502 |
+
|
1503 |
+
[convolutional]
|
1504 |
+
batch_normalize=1
|
1505 |
+
filters=512
|
1506 |
+
size=1
|
1507 |
+
stride=1
|
1508 |
+
pad=1
|
1509 |
+
activation=silu
|
1510 |
+
|
1511 |
+
[convolutional]
|
1512 |
+
batch_normalize=1
|
1513 |
+
size=3
|
1514 |
+
stride=1
|
1515 |
+
pad=1
|
1516 |
+
filters=512
|
1517 |
+
activation=silu
|
1518 |
+
|
1519 |
+
[convolutional]
|
1520 |
+
batch_normalize=1
|
1521 |
+
filters=512
|
1522 |
+
size=1
|
1523 |
+
stride=1
|
1524 |
+
pad=1
|
1525 |
+
activation=silu
|
1526 |
+
|
1527 |
+
[convolutional]
|
1528 |
+
batch_normalize=1
|
1529 |
+
size=3
|
1530 |
+
stride=1
|
1531 |
+
pad=1
|
1532 |
+
filters=512
|
1533 |
+
activation=silu
|
1534 |
+
|
1535 |
+
[route]
|
1536 |
+
layers = -1,-8
|
1537 |
+
|
1538 |
+
# Transition last
|
1539 |
+
|
1540 |
+
# 202 (previous+3+4+2k)
|
1541 |
+
[convolutional]
|
1542 |
+
batch_normalize=1
|
1543 |
+
filters=512
|
1544 |
+
size=1
|
1545 |
+
stride=1
|
1546 |
+
pad=1
|
1547 |
+
activation=silu
|
1548 |
+
|
1549 |
+
# ============ End of Neck ============ #
|
1550 |
+
|
1551 |
+
# 203
|
1552 |
+
[implicit_add]
|
1553 |
+
filters=256
|
1554 |
+
|
1555 |
+
# 204
|
1556 |
+
[implicit_add]
|
1557 |
+
filters=512
|
1558 |
+
|
1559 |
+
# 205
|
1560 |
+
[implicit_add]
|
1561 |
+
filters=768
|
1562 |
+
|
1563 |
+
# 206
|
1564 |
+
[implicit_add]
|
1565 |
+
filters=1024
|
1566 |
+
|
1567 |
+
# 207
|
1568 |
+
[implicit_mul]
|
1569 |
+
filters=255
|
1570 |
+
|
1571 |
+
# 208
|
1572 |
+
[implicit_mul]
|
1573 |
+
filters=255
|
1574 |
+
|
1575 |
+
# 209
|
1576 |
+
[implicit_mul]
|
1577 |
+
filters=255
|
1578 |
+
|
1579 |
+
# 210
|
1580 |
+
[implicit_mul]
|
1581 |
+
filters=255
|
1582 |
+
|
1583 |
+
# ============ Head ============ #
|
1584 |
+
|
1585 |
+
# YOLO-3
|
1586 |
+
|
1587 |
+
[route]
|
1588 |
+
layers = 163
|
1589 |
+
|
1590 |
+
[convolutional]
|
1591 |
+
batch_normalize=1
|
1592 |
+
size=3
|
1593 |
+
stride=1
|
1594 |
+
pad=1
|
1595 |
+
filters=256
|
1596 |
+
activation=silu
|
1597 |
+
|
1598 |
+
[shift_channels]
|
1599 |
+
from=203
|
1600 |
+
|
1601 |
+
[convolutional]
|
1602 |
+
size=1
|
1603 |
+
stride=1
|
1604 |
+
pad=1
|
1605 |
+
filters=255
|
1606 |
+
activation=linear
|
1607 |
+
|
1608 |
+
[control_channels]
|
1609 |
+
from=207
|
1610 |
+
|
1611 |
+
[yolo]
|
1612 |
+
mask = 0,1,2
|
1613 |
+
anchors = 19,27, 44,40, 38,94, 96,68, 86,152, 180,137, 140,301, 303,264, 238,542, 436,615, 739,380, 925,792
|
1614 |
+
classes=80
|
1615 |
+
num=12
|
1616 |
+
jitter=.3
|
1617 |
+
ignore_thresh = .7
|
1618 |
+
truth_thresh = 1
|
1619 |
+
random=1
|
1620 |
+
scale_x_y = 1.05
|
1621 |
+
iou_thresh=0.213
|
1622 |
+
cls_normalizer=1.0
|
1623 |
+
iou_normalizer=0.07
|
1624 |
+
iou_loss=ciou
|
1625 |
+
nms_kind=greedynms
|
1626 |
+
beta_nms=0.6
|
1627 |
+
|
1628 |
+
|
1629 |
+
# YOLO-4
|
1630 |
+
|
1631 |
+
[route]
|
1632 |
+
layers = 176
|
1633 |
+
|
1634 |
+
[convolutional]
|
1635 |
+
batch_normalize=1
|
1636 |
+
size=3
|
1637 |
+
stride=1
|
1638 |
+
pad=1
|
1639 |
+
filters=512
|
1640 |
+
activation=silu
|
1641 |
+
|
1642 |
+
[shift_channels]
|
1643 |
+
from=204
|
1644 |
+
|
1645 |
+
[convolutional]
|
1646 |
+
size=1
|
1647 |
+
stride=1
|
1648 |
+
pad=1
|
1649 |
+
filters=255
|
1650 |
+
activation=linear
|
1651 |
+
|
1652 |
+
[control_channels]
|
1653 |
+
from=208
|
1654 |
+
|
1655 |
+
[yolo]
|
1656 |
+
mask = 3,4,5
|
1657 |
+
anchors = 19,27, 44,40, 38,94, 96,68, 86,152, 180,137, 140,301, 303,264, 238,542, 436,615, 739,380, 925,792
|
1658 |
+
classes=80
|
1659 |
+
num=12
|
1660 |
+
jitter=.3
|
1661 |
+
ignore_thresh = .7
|
1662 |
+
truth_thresh = 1
|
1663 |
+
random=1
|
1664 |
+
scale_x_y = 1.05
|
1665 |
+
iou_thresh=0.213
|
1666 |
+
cls_normalizer=1.0
|
1667 |
+
iou_normalizer=0.07
|
1668 |
+
iou_loss=ciou
|
1669 |
+
nms_kind=greedynms
|
1670 |
+
beta_nms=0.6
|
1671 |
+
|
1672 |
+
|
1673 |
+
# YOLO-5
|
1674 |
+
|
1675 |
+
[route]
|
1676 |
+
layers = 189
|
1677 |
+
|
1678 |
+
[convolutional]
|
1679 |
+
batch_normalize=1
|
1680 |
+
size=3
|
1681 |
+
stride=1
|
1682 |
+
pad=1
|
1683 |
+
filters=768
|
1684 |
+
activation=silu
|
1685 |
+
|
1686 |
+
[shift_channels]
|
1687 |
+
from=205
|
1688 |
+
|
1689 |
+
[convolutional]
|
1690 |
+
size=1
|
1691 |
+
stride=1
|
1692 |
+
pad=1
|
1693 |
+
filters=255
|
1694 |
+
activation=linear
|
1695 |
+
|
1696 |
+
[control_channels]
|
1697 |
+
from=209
|
1698 |
+
|
1699 |
+
[yolo]
|
1700 |
+
mask = 6,7,8
|
1701 |
+
anchors = 19,27, 44,40, 38,94, 96,68, 86,152, 180,137, 140,301, 303,264, 238,542, 436,615, 739,380, 925,792
|
1702 |
+
classes=80
|
1703 |
+
num=12
|
1704 |
+
jitter=.3
|
1705 |
+
ignore_thresh = .7
|
1706 |
+
truth_thresh = 1
|
1707 |
+
random=1
|
1708 |
+
scale_x_y = 1.05
|
1709 |
+
iou_thresh=0.213
|
1710 |
+
cls_normalizer=1.0
|
1711 |
+
iou_normalizer=0.07
|
1712 |
+
iou_loss=ciou
|
1713 |
+
nms_kind=greedynms
|
1714 |
+
beta_nms=0.6
|
1715 |
+
|
1716 |
+
|
1717 |
+
# YOLO-6
|
1718 |
+
|
1719 |
+
[route]
|
1720 |
+
layers = 202
|
1721 |
+
|
1722 |
+
[convolutional]
|
1723 |
+
batch_normalize=1
|
1724 |
+
size=3
|
1725 |
+
stride=1
|
1726 |
+
pad=1
|
1727 |
+
filters=1024
|
1728 |
+
activation=silu
|
1729 |
+
|
1730 |
+
[shift_channels]
|
1731 |
+
from=206
|
1732 |
+
|
1733 |
+
[convolutional]
|
1734 |
+
size=1
|
1735 |
+
stride=1
|
1736 |
+
pad=1
|
1737 |
+
filters=255
|
1738 |
+
activation=linear
|
1739 |
+
|
1740 |
+
[control_channels]
|
1741 |
+
from=210
|
1742 |
+
|
1743 |
+
[yolo]
|
1744 |
+
mask = 9,10,11
|
1745 |
+
anchors = 19,27, 44,40, 38,94, 96,68, 86,152, 180,137, 140,301, 303,264, 238,542, 436,615, 739,380, 925,792
|
1746 |
+
classes=80
|
1747 |
+
num=12
|
1748 |
+
jitter=.3
|
1749 |
+
ignore_thresh = .7
|
1750 |
+
truth_thresh = 1
|
1751 |
+
random=1
|
1752 |
+
scale_x_y = 1.05
|
1753 |
+
iou_thresh=0.213
|
1754 |
+
cls_normalizer=1.0
|
1755 |
+
iou_normalizer=0.07
|
1756 |
+
iou_loss=ciou
|
1757 |
+
nms_kind=greedynms
|
1758 |
+
beta_nms=0.6
|
1759 |
+
|
1760 |
+
# ============ End of Head ============ #
|
cfg/yolov4_csp.cfg
ADDED
@@ -0,0 +1,1334 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[net]
|
2 |
+
# Testing
|
3 |
+
#batch=1
|
4 |
+
#subdivisions=1
|
5 |
+
# Training
|
6 |
+
batch=64
|
7 |
+
subdivisions=8
|
8 |
+
width=512
|
9 |
+
height=512
|
10 |
+
channels=3
|
11 |
+
momentum=0.949
|
12 |
+
decay=0.0005
|
13 |
+
angle=0
|
14 |
+
saturation = 1.5
|
15 |
+
exposure = 1.5
|
16 |
+
hue=.1
|
17 |
+
|
18 |
+
learning_rate=0.00261
|
19 |
+
burn_in=1000
|
20 |
+
max_batches = 500500
|
21 |
+
policy=steps
|
22 |
+
steps=400000,450000
|
23 |
+
scales=.1,.1
|
24 |
+
|
25 |
+
#cutmix=1
|
26 |
+
mosaic=1
|
27 |
+
|
28 |
+
|
29 |
+
# ============ Backbone ============ #
|
30 |
+
|
31 |
+
# Stem
|
32 |
+
|
33 |
+
# 0
|
34 |
+
[convolutional]
|
35 |
+
batch_normalize=1
|
36 |
+
filters=32
|
37 |
+
size=3
|
38 |
+
stride=1
|
39 |
+
pad=1
|
40 |
+
activation=silu
|
41 |
+
|
42 |
+
# P1
|
43 |
+
|
44 |
+
# Downsample
|
45 |
+
|
46 |
+
[convolutional]
|
47 |
+
batch_normalize=1
|
48 |
+
filters=64
|
49 |
+
size=3
|
50 |
+
stride=2
|
51 |
+
pad=1
|
52 |
+
activation=silu
|
53 |
+
|
54 |
+
# Residual Block
|
55 |
+
|
56 |
+
[convolutional]
|
57 |
+
batch_normalize=1
|
58 |
+
filters=32
|
59 |
+
size=1
|
60 |
+
stride=1
|
61 |
+
pad=1
|
62 |
+
activation=silu
|
63 |
+
|
64 |
+
[convolutional]
|
65 |
+
batch_normalize=1
|
66 |
+
filters=64
|
67 |
+
size=3
|
68 |
+
stride=1
|
69 |
+
pad=1
|
70 |
+
activation=silu
|
71 |
+
|
72 |
+
# 4 (previous+1+3k)
|
73 |
+
[shortcut]
|
74 |
+
from=-3
|
75 |
+
activation=linear
|
76 |
+
|
77 |
+
# P2
|
78 |
+
|
79 |
+
# Downsample
|
80 |
+
|
81 |
+
[convolutional]
|
82 |
+
batch_normalize=1
|
83 |
+
filters=128
|
84 |
+
size=3
|
85 |
+
stride=2
|
86 |
+
pad=1
|
87 |
+
activation=silu
|
88 |
+
|
89 |
+
# Split
|
90 |
+
|
91 |
+
[convolutional]
|
92 |
+
batch_normalize=1
|
93 |
+
filters=64
|
94 |
+
size=1
|
95 |
+
stride=1
|
96 |
+
pad=1
|
97 |
+
activation=silu
|
98 |
+
|
99 |
+
[route]
|
100 |
+
layers = -2
|
101 |
+
|
102 |
+
[convolutional]
|
103 |
+
batch_normalize=1
|
104 |
+
filters=64
|
105 |
+
size=1
|
106 |
+
stride=1
|
107 |
+
pad=1
|
108 |
+
activation=silu
|
109 |
+
|
110 |
+
# Residual Block
|
111 |
+
|
112 |
+
[convolutional]
|
113 |
+
batch_normalize=1
|
114 |
+
filters=64
|
115 |
+
size=1
|
116 |
+
stride=1
|
117 |
+
pad=1
|
118 |
+
activation=silu
|
119 |
+
|
120 |
+
[convolutional]
|
121 |
+
batch_normalize=1
|
122 |
+
filters=64
|
123 |
+
size=3
|
124 |
+
stride=1
|
125 |
+
pad=1
|
126 |
+
activation=silu
|
127 |
+
|
128 |
+
[shortcut]
|
129 |
+
from=-3
|
130 |
+
activation=linear
|
131 |
+
|
132 |
+
[convolutional]
|
133 |
+
batch_normalize=1
|
134 |
+
filters=64
|
135 |
+
size=1
|
136 |
+
stride=1
|
137 |
+
pad=1
|
138 |
+
activation=silu
|
139 |
+
|
140 |
+
[convolutional]
|
141 |
+
batch_normalize=1
|
142 |
+
filters=64
|
143 |
+
size=3
|
144 |
+
stride=1
|
145 |
+
pad=1
|
146 |
+
activation=silu
|
147 |
+
|
148 |
+
[shortcut]
|
149 |
+
from=-3
|
150 |
+
activation=linear
|
151 |
+
|
152 |
+
# Transition first
|
153 |
+
|
154 |
+
[convolutional]
|
155 |
+
batch_normalize=1
|
156 |
+
filters=64
|
157 |
+
size=1
|
158 |
+
stride=1
|
159 |
+
pad=1
|
160 |
+
activation=silu
|
161 |
+
|
162 |
+
# Merge [-1, -(3k+4)]
|
163 |
+
|
164 |
+
[route]
|
165 |
+
layers = -1,-10
|
166 |
+
|
167 |
+
# Transition last
|
168 |
+
|
169 |
+
# 17 (previous+7+3k)
|
170 |
+
[convolutional]
|
171 |
+
batch_normalize=1
|
172 |
+
filters=128
|
173 |
+
size=1
|
174 |
+
stride=1
|
175 |
+
pad=1
|
176 |
+
activation=silu
|
177 |
+
|
178 |
+
# P3
|
179 |
+
|
180 |
+
# Downsample
|
181 |
+
|
182 |
+
[convolutional]
|
183 |
+
batch_normalize=1
|
184 |
+
filters=256
|
185 |
+
size=3
|
186 |
+
stride=2
|
187 |
+
pad=1
|
188 |
+
activation=silu
|
189 |
+
|
190 |
+
# Split
|
191 |
+
|
192 |
+
[convolutional]
|
193 |
+
batch_normalize=1
|
194 |
+
filters=128
|
195 |
+
size=1
|
196 |
+
stride=1
|
197 |
+
pad=1
|
198 |
+
activation=silu
|
199 |
+
|
200 |
+
[route]
|
201 |
+
layers = -2
|
202 |
+
|
203 |
+
[convolutional]
|
204 |
+
batch_normalize=1
|
205 |
+
filters=128
|
206 |
+
size=1
|
207 |
+
stride=1
|
208 |
+
pad=1
|
209 |
+
activation=silu
|
210 |
+
|
211 |
+
# Residual Block
|
212 |
+
|
213 |
+
[convolutional]
|
214 |
+
batch_normalize=1
|
215 |
+
filters=128
|
216 |
+
size=1
|
217 |
+
stride=1
|
218 |
+
pad=1
|
219 |
+
activation=silu
|
220 |
+
|
221 |
+
[convolutional]
|
222 |
+
batch_normalize=1
|
223 |
+
filters=128
|
224 |
+
size=3
|
225 |
+
stride=1
|
226 |
+
pad=1
|
227 |
+
activation=silu
|
228 |
+
|
229 |
+
[shortcut]
|
230 |
+
from=-3
|
231 |
+
activation=linear
|
232 |
+
|
233 |
+
[convolutional]
|
234 |
+
batch_normalize=1
|
235 |
+
filters=128
|
236 |
+
size=1
|
237 |
+
stride=1
|
238 |
+
pad=1
|
239 |
+
activation=silu
|
240 |
+
|
241 |
+
[convolutional]
|
242 |
+
batch_normalize=1
|
243 |
+
filters=128
|
244 |
+
size=3
|
245 |
+
stride=1
|
246 |
+
pad=1
|
247 |
+
activation=silu
|
248 |
+
|
249 |
+
[shortcut]
|
250 |
+
from=-3
|
251 |
+
activation=linear
|
252 |
+
|
253 |
+
[convolutional]
|
254 |
+
batch_normalize=1
|
255 |
+
filters=128
|
256 |
+
size=1
|
257 |
+
stride=1
|
258 |
+
pad=1
|
259 |
+
activation=silu
|
260 |
+
|
261 |
+
[convolutional]
|
262 |
+
batch_normalize=1
|
263 |
+
filters=128
|
264 |
+
size=3
|
265 |
+
stride=1
|
266 |
+
pad=1
|
267 |
+
activation=silu
|
268 |
+
|
269 |
+
[shortcut]
|
270 |
+
from=-3
|
271 |
+
activation=linear
|
272 |
+
|
273 |
+
[convolutional]
|
274 |
+
batch_normalize=1
|
275 |
+
filters=128
|
276 |
+
size=1
|
277 |
+
stride=1
|
278 |
+
pad=1
|
279 |
+
activation=silu
|
280 |
+
|
281 |
+
[convolutional]
|
282 |
+
batch_normalize=1
|
283 |
+
filters=128
|
284 |
+
size=3
|
285 |
+
stride=1
|
286 |
+
pad=1
|
287 |
+
activation=silu
|
288 |
+
|
289 |
+
[shortcut]
|
290 |
+
from=-3
|
291 |
+
activation=linear
|
292 |
+
|
293 |
+
[convolutional]
|
294 |
+
batch_normalize=1
|
295 |
+
filters=128
|
296 |
+
size=1
|
297 |
+
stride=1
|
298 |
+
pad=1
|
299 |
+
activation=silu
|
300 |
+
|
301 |
+
[convolutional]
|
302 |
+
batch_normalize=1
|
303 |
+
filters=128
|
304 |
+
size=3
|
305 |
+
stride=1
|
306 |
+
pad=1
|
307 |
+
activation=silu
|
308 |
+
|
309 |
+
[shortcut]
|
310 |
+
from=-3
|
311 |
+
activation=linear
|
312 |
+
|
313 |
+
[convolutional]
|
314 |
+
batch_normalize=1
|
315 |
+
filters=128
|
316 |
+
size=1
|
317 |
+
stride=1
|
318 |
+
pad=1
|
319 |
+
activation=silu
|
320 |
+
|
321 |
+
[convolutional]
|
322 |
+
batch_normalize=1
|
323 |
+
filters=128
|
324 |
+
size=3
|
325 |
+
stride=1
|
326 |
+
pad=1
|
327 |
+
activation=silu
|
328 |
+
|
329 |
+
[shortcut]
|
330 |
+
from=-3
|
331 |
+
activation=linear
|
332 |
+
|
333 |
+
[convolutional]
|
334 |
+
batch_normalize=1
|
335 |
+
filters=128
|
336 |
+
size=1
|
337 |
+
stride=1
|
338 |
+
pad=1
|
339 |
+
activation=silu
|
340 |
+
|
341 |
+
[convolutional]
|
342 |
+
batch_normalize=1
|
343 |
+
filters=128
|
344 |
+
size=3
|
345 |
+
stride=1
|
346 |
+
pad=1
|
347 |
+
activation=silu
|
348 |
+
|
349 |
+
[shortcut]
|
350 |
+
from=-3
|
351 |
+
activation=linear
|
352 |
+
|
353 |
+
[convolutional]
|
354 |
+
batch_normalize=1
|
355 |
+
filters=128
|
356 |
+
size=1
|
357 |
+
stride=1
|
358 |
+
pad=1
|
359 |
+
activation=silu
|
360 |
+
|
361 |
+
[convolutional]
|
362 |
+
batch_normalize=1
|
363 |
+
filters=128
|
364 |
+
size=3
|
365 |
+
stride=1
|
366 |
+
pad=1
|
367 |
+
activation=silu
|
368 |
+
|
369 |
+
[shortcut]
|
370 |
+
from=-3
|
371 |
+
activation=linear
|
372 |
+
|
373 |
+
# Transition first
|
374 |
+
|
375 |
+
[convolutional]
|
376 |
+
batch_normalize=1
|
377 |
+
filters=128
|
378 |
+
size=1
|
379 |
+
stride=1
|
380 |
+
pad=1
|
381 |
+
activation=silu
|
382 |
+
|
383 |
+
# Merge [-1 -(4+3k)]
|
384 |
+
|
385 |
+
[route]
|
386 |
+
layers = -1,-28
|
387 |
+
|
388 |
+
# Transition last
|
389 |
+
|
390 |
+
# 48 (previous+7+3k)
|
391 |
+
[convolutional]
|
392 |
+
batch_normalize=1
|
393 |
+
filters=256
|
394 |
+
size=1
|
395 |
+
stride=1
|
396 |
+
pad=1
|
397 |
+
activation=silu
|
398 |
+
|
399 |
+
# P4
|
400 |
+
|
401 |
+
# Downsample
|
402 |
+
|
403 |
+
[convolutional]
|
404 |
+
batch_normalize=1
|
405 |
+
filters=512
|
406 |
+
size=3
|
407 |
+
stride=2
|
408 |
+
pad=1
|
409 |
+
activation=silu
|
410 |
+
|
411 |
+
# Split
|
412 |
+
|
413 |
+
[convolutional]
|
414 |
+
batch_normalize=1
|
415 |
+
filters=256
|
416 |
+
size=1
|
417 |
+
stride=1
|
418 |
+
pad=1
|
419 |
+
activation=silu
|
420 |
+
|
421 |
+
[route]
|
422 |
+
layers = -2
|
423 |
+
|
424 |
+
[convolutional]
|
425 |
+
batch_normalize=1
|
426 |
+
filters=256
|
427 |
+
size=1
|
428 |
+
stride=1
|
429 |
+
pad=1
|
430 |
+
activation=silu
|
431 |
+
|
432 |
+
# Residual Block
|
433 |
+
|
434 |
+
[convolutional]
|
435 |
+
batch_normalize=1
|
436 |
+
filters=256
|
437 |
+
size=1
|
438 |
+
stride=1
|
439 |
+
pad=1
|
440 |
+
activation=silu
|
441 |
+
|
442 |
+
[convolutional]
|
443 |
+
batch_normalize=1
|
444 |
+
filters=256
|
445 |
+
size=3
|
446 |
+
stride=1
|
447 |
+
pad=1
|
448 |
+
activation=silu
|
449 |
+
|
450 |
+
[shortcut]
|
451 |
+
from=-3
|
452 |
+
activation=linear
|
453 |
+
|
454 |
+
[convolutional]
|
455 |
+
batch_normalize=1
|
456 |
+
filters=256
|
457 |
+
size=1
|
458 |
+
stride=1
|
459 |
+
pad=1
|
460 |
+
activation=silu
|
461 |
+
|
462 |
+
[convolutional]
|
463 |
+
batch_normalize=1
|
464 |
+
filters=256
|
465 |
+
size=3
|
466 |
+
stride=1
|
467 |
+
pad=1
|
468 |
+
activation=silu
|
469 |
+
|
470 |
+
[shortcut]
|
471 |
+
from=-3
|
472 |
+
activation=linear
|
473 |
+
|
474 |
+
[convolutional]
|
475 |
+
batch_normalize=1
|
476 |
+
filters=256
|
477 |
+
size=1
|
478 |
+
stride=1
|
479 |
+
pad=1
|
480 |
+
activation=silu
|
481 |
+
|
482 |
+
[convolutional]
|
483 |
+
batch_normalize=1
|
484 |
+
filters=256
|
485 |
+
size=3
|
486 |
+
stride=1
|
487 |
+
pad=1
|
488 |
+
activation=silu
|
489 |
+
|
490 |
+
[shortcut]
|
491 |
+
from=-3
|
492 |
+
activation=linear
|
493 |
+
|
494 |
+
[convolutional]
|
495 |
+
batch_normalize=1
|
496 |
+
filters=256
|
497 |
+
size=1
|
498 |
+
stride=1
|
499 |
+
pad=1
|
500 |
+
activation=silu
|
501 |
+
|
502 |
+
[convolutional]
|
503 |
+
batch_normalize=1
|
504 |
+
filters=256
|
505 |
+
size=3
|
506 |
+
stride=1
|
507 |
+
pad=1
|
508 |
+
activation=silu
|
509 |
+
|
510 |
+
[shortcut]
|
511 |
+
from=-3
|
512 |
+
activation=linear
|
513 |
+
|
514 |
+
[convolutional]
|
515 |
+
batch_normalize=1
|
516 |
+
filters=256
|
517 |
+
size=1
|
518 |
+
stride=1
|
519 |
+
pad=1
|
520 |
+
activation=silu
|
521 |
+
|
522 |
+
[convolutional]
|
523 |
+
batch_normalize=1
|
524 |
+
filters=256
|
525 |
+
size=3
|
526 |
+
stride=1
|
527 |
+
pad=1
|
528 |
+
activation=silu
|
529 |
+
|
530 |
+
[shortcut]
|
531 |
+
from=-3
|
532 |
+
activation=linear
|
533 |
+
|
534 |
+
[convolutional]
|
535 |
+
batch_normalize=1
|
536 |
+
filters=256
|
537 |
+
size=1
|
538 |
+
stride=1
|
539 |
+
pad=1
|
540 |
+
activation=silu
|
541 |
+
|
542 |
+
[convolutional]
|
543 |
+
batch_normalize=1
|
544 |
+
filters=256
|
545 |
+
size=3
|
546 |
+
stride=1
|
547 |
+
pad=1
|
548 |
+
activation=silu
|
549 |
+
|
550 |
+
[shortcut]
|
551 |
+
from=-3
|
552 |
+
activation=linear
|
553 |
+
|
554 |
+
[convolutional]
|
555 |
+
batch_normalize=1
|
556 |
+
filters=256
|
557 |
+
size=1
|
558 |
+
stride=1
|
559 |
+
pad=1
|
560 |
+
activation=silu
|
561 |
+
|
562 |
+
[convolutional]
|
563 |
+
batch_normalize=1
|
564 |
+
filters=256
|
565 |
+
size=3
|
566 |
+
stride=1
|
567 |
+
pad=1
|
568 |
+
activation=silu
|
569 |
+
|
570 |
+
[shortcut]
|
571 |
+
from=-3
|
572 |
+
activation=linear
|
573 |
+
|
574 |
+
[convolutional]
|
575 |
+
batch_normalize=1
|
576 |
+
filters=256
|
577 |
+
size=1
|
578 |
+
stride=1
|
579 |
+
pad=1
|
580 |
+
activation=silu
|
581 |
+
|
582 |
+
[convolutional]
|
583 |
+
batch_normalize=1
|
584 |
+
filters=256
|
585 |
+
size=3
|
586 |
+
stride=1
|
587 |
+
pad=1
|
588 |
+
activation=silu
|
589 |
+
|
590 |
+
[shortcut]
|
591 |
+
from=-3
|
592 |
+
activation=linear
|
593 |
+
|
594 |
+
# Transition first
|
595 |
+
|
596 |
+
[convolutional]
|
597 |
+
batch_normalize=1
|
598 |
+
filters=256
|
599 |
+
size=1
|
600 |
+
stride=1
|
601 |
+
pad=1
|
602 |
+
activation=silu
|
603 |
+
|
604 |
+
# Merge [-1 -(3k+4)]
|
605 |
+
|
606 |
+
[route]
|
607 |
+
layers = -1,-28
|
608 |
+
|
609 |
+
# Transition last
|
610 |
+
|
611 |
+
# 79 (previous+7+3k)
|
612 |
+
[convolutional]
|
613 |
+
batch_normalize=1
|
614 |
+
filters=512
|
615 |
+
size=1
|
616 |
+
stride=1
|
617 |
+
pad=1
|
618 |
+
activation=silu
|
619 |
+
|
620 |
+
# P5
|
621 |
+
|
622 |
+
# Downsample
|
623 |
+
|
624 |
+
[convolutional]
|
625 |
+
batch_normalize=1
|
626 |
+
filters=1024
|
627 |
+
size=3
|
628 |
+
stride=2
|
629 |
+
pad=1
|
630 |
+
activation=silu
|
631 |
+
|
632 |
+
# Split
|
633 |
+
|
634 |
+
[convolutional]
|
635 |
+
batch_normalize=1
|
636 |
+
filters=512
|
637 |
+
size=1
|
638 |
+
stride=1
|
639 |
+
pad=1
|
640 |
+
activation=silu
|
641 |
+
|
642 |
+
[route]
|
643 |
+
layers = -2
|
644 |
+
|
645 |
+
[convolutional]
|
646 |
+
batch_normalize=1
|
647 |
+
filters=512
|
648 |
+
size=1
|
649 |
+
stride=1
|
650 |
+
pad=1
|
651 |
+
activation=silu
|
652 |
+
|
653 |
+
# Residual Block
|
654 |
+
|
655 |
+
[convolutional]
|
656 |
+
batch_normalize=1
|
657 |
+
filters=512
|
658 |
+
size=1
|
659 |
+
stride=1
|
660 |
+
pad=1
|
661 |
+
activation=silu
|
662 |
+
|
663 |
+
[convolutional]
|
664 |
+
batch_normalize=1
|
665 |
+
filters=512
|
666 |
+
size=3
|
667 |
+
stride=1
|
668 |
+
pad=1
|
669 |
+
activation=silu
|
670 |
+
|
671 |
+
[shortcut]
|
672 |
+
from=-3
|
673 |
+
activation=linear
|
674 |
+
|
675 |
+
[convolutional]
|
676 |
+
batch_normalize=1
|
677 |
+
filters=512
|
678 |
+
size=1
|
679 |
+
stride=1
|
680 |
+
pad=1
|
681 |
+
activation=silu
|
682 |
+
|
683 |
+
[convolutional]
|
684 |
+
batch_normalize=1
|
685 |
+
filters=512
|
686 |
+
size=3
|
687 |
+
stride=1
|
688 |
+
pad=1
|
689 |
+
activation=silu
|
690 |
+
|
691 |
+
[shortcut]
|
692 |
+
from=-3
|
693 |
+
activation=linear
|
694 |
+
|
695 |
+
[convolutional]
|
696 |
+
batch_normalize=1
|
697 |
+
filters=512
|
698 |
+
size=1
|
699 |
+
stride=1
|
700 |
+
pad=1
|
701 |
+
activation=silu
|
702 |
+
|
703 |
+
[convolutional]
|
704 |
+
batch_normalize=1
|
705 |
+
filters=512
|
706 |
+
size=3
|
707 |
+
stride=1
|
708 |
+
pad=1
|
709 |
+
activation=silu
|
710 |
+
|
711 |
+
[shortcut]
|
712 |
+
from=-3
|
713 |
+
activation=linear
|
714 |
+
|
715 |
+
[convolutional]
|
716 |
+
batch_normalize=1
|
717 |
+
filters=512
|
718 |
+
size=1
|
719 |
+
stride=1
|
720 |
+
pad=1
|
721 |
+
activation=silu
|
722 |
+
|
723 |
+
[convolutional]
|
724 |
+
batch_normalize=1
|
725 |
+
filters=512
|
726 |
+
size=3
|
727 |
+
stride=1
|
728 |
+
pad=1
|
729 |
+
activation=silu
|
730 |
+
|
731 |
+
[shortcut]
|
732 |
+
from=-3
|
733 |
+
activation=linear
|
734 |
+
|
735 |
+
# Transition first
|
736 |
+
|
737 |
+
[convolutional]
|
738 |
+
batch_normalize=1
|
739 |
+
filters=512
|
740 |
+
size=1
|
741 |
+
stride=1
|
742 |
+
pad=1
|
743 |
+
activation=silu
|
744 |
+
|
745 |
+
# Merge [-1 -(3k+4)]
|
746 |
+
|
747 |
+
[route]
|
748 |
+
layers = -1,-16
|
749 |
+
|
750 |
+
# Transition last
|
751 |
+
|
752 |
+
# 98 (previous+7+3k)
|
753 |
+
[convolutional]
|
754 |
+
batch_normalize=1
|
755 |
+
filters=1024
|
756 |
+
size=1
|
757 |
+
stride=1
|
758 |
+
pad=1
|
759 |
+
activation=silu
|
760 |
+
|
761 |
+
# ============ End of Backbone ============ #
|
762 |
+
|
763 |
+
# ============ Neck ============ #
|
764 |
+
|
765 |
+
# CSPSPP
|
766 |
+
|
767 |
+
[convolutional]
|
768 |
+
batch_normalize=1
|
769 |
+
filters=512
|
770 |
+
size=1
|
771 |
+
stride=1
|
772 |
+
pad=1
|
773 |
+
activation=silu
|
774 |
+
|
775 |
+
[route]
|
776 |
+
layers = -2
|
777 |
+
|
778 |
+
[convolutional]
|
779 |
+
batch_normalize=1
|
780 |
+
filters=512
|
781 |
+
size=1
|
782 |
+
stride=1
|
783 |
+
pad=1
|
784 |
+
activation=silu
|
785 |
+
|
786 |
+
[convolutional]
|
787 |
+
batch_normalize=1
|
788 |
+
size=3
|
789 |
+
stride=1
|
790 |
+
pad=1
|
791 |
+
filters=512
|
792 |
+
activation=silu
|
793 |
+
|
794 |
+
[convolutional]
|
795 |
+
batch_normalize=1
|
796 |
+
filters=512
|
797 |
+
size=1
|
798 |
+
stride=1
|
799 |
+
pad=1
|
800 |
+
activation=silu
|
801 |
+
|
802 |
+
### SPP ###
|
803 |
+
[maxpool]
|
804 |
+
stride=1
|
805 |
+
size=5
|
806 |
+
|
807 |
+
[route]
|
808 |
+
layers=-2
|
809 |
+
|
810 |
+
[maxpool]
|
811 |
+
stride=1
|
812 |
+
size=9
|
813 |
+
|
814 |
+
[route]
|
815 |
+
layers=-4
|
816 |
+
|
817 |
+
[maxpool]
|
818 |
+
stride=1
|
819 |
+
size=13
|
820 |
+
|
821 |
+
[route]
|
822 |
+
layers=-1,-3,-5,-6
|
823 |
+
### End SPP ###
|
824 |
+
|
825 |
+
[convolutional]
|
826 |
+
batch_normalize=1
|
827 |
+
filters=512
|
828 |
+
size=1
|
829 |
+
stride=1
|
830 |
+
pad=1
|
831 |
+
activation=silu
|
832 |
+
|
833 |
+
[convolutional]
|
834 |
+
batch_normalize=1
|
835 |
+
size=3
|
836 |
+
stride=1
|
837 |
+
pad=1
|
838 |
+
filters=512
|
839 |
+
activation=silu
|
840 |
+
|
841 |
+
[route]
|
842 |
+
layers = -1, -13
|
843 |
+
|
844 |
+
# 113 (previous+6+5+2k)
|
845 |
+
[convolutional]
|
846 |
+
batch_normalize=1
|
847 |
+
filters=512
|
848 |
+
size=1
|
849 |
+
stride=1
|
850 |
+
pad=1
|
851 |
+
activation=silu
|
852 |
+
|
853 |
+
# End of CSPSPP
|
854 |
+
|
855 |
+
|
856 |
+
# FPN-4
|
857 |
+
|
858 |
+
[convolutional]
|
859 |
+
batch_normalize=1
|
860 |
+
filters=256
|
861 |
+
size=1
|
862 |
+
stride=1
|
863 |
+
pad=1
|
864 |
+
activation=silu
|
865 |
+
|
866 |
+
[upsample]
|
867 |
+
stride=2
|
868 |
+
|
869 |
+
[route]
|
870 |
+
layers = 79
|
871 |
+
|
872 |
+
[convolutional]
|
873 |
+
batch_normalize=1
|
874 |
+
filters=256
|
875 |
+
size=1
|
876 |
+
stride=1
|
877 |
+
pad=1
|
878 |
+
activation=silu
|
879 |
+
|
880 |
+
[route]
|
881 |
+
layers = -1, -3
|
882 |
+
|
883 |
+
[convolutional]
|
884 |
+
batch_normalize=1
|
885 |
+
filters=256
|
886 |
+
size=1
|
887 |
+
stride=1
|
888 |
+
pad=1
|
889 |
+
activation=silu
|
890 |
+
|
891 |
+
# Split
|
892 |
+
|
893 |
+
[convolutional]
|
894 |
+
batch_normalize=1
|
895 |
+
filters=256
|
896 |
+
size=1
|
897 |
+
stride=1
|
898 |
+
pad=1
|
899 |
+
activation=silu
|
900 |
+
|
901 |
+
[route]
|
902 |
+
layers = -2
|
903 |
+
|
904 |
+
# Plain Block
|
905 |
+
|
906 |
+
[convolutional]
|
907 |
+
batch_normalize=1
|
908 |
+
filters=256
|
909 |
+
size=1
|
910 |
+
stride=1
|
911 |
+
pad=1
|
912 |
+
activation=silu
|
913 |
+
|
914 |
+
[convolutional]
|
915 |
+
batch_normalize=1
|
916 |
+
size=3
|
917 |
+
stride=1
|
918 |
+
pad=1
|
919 |
+
filters=256
|
920 |
+
activation=silu
|
921 |
+
|
922 |
+
[convolutional]
|
923 |
+
batch_normalize=1
|
924 |
+
filters=256
|
925 |
+
size=1
|
926 |
+
stride=1
|
927 |
+
pad=1
|
928 |
+
activation=silu
|
929 |
+
|
930 |
+
[convolutional]
|
931 |
+
batch_normalize=1
|
932 |
+
size=3
|
933 |
+
stride=1
|
934 |
+
pad=1
|
935 |
+
filters=256
|
936 |
+
activation=silu
|
937 |
+
|
938 |
+
# Merge [-1, -(2k+2)]
|
939 |
+
|
940 |
+
[route]
|
941 |
+
layers = -1, -6
|
942 |
+
|
943 |
+
# Transition last
|
944 |
+
|
945 |
+
# 127 (previous+6+4+2k)
|
946 |
+
[convolutional]
|
947 |
+
batch_normalize=1
|
948 |
+
filters=256
|
949 |
+
size=1
|
950 |
+
stride=1
|
951 |
+
pad=1
|
952 |
+
activation=silu
|
953 |
+
|
954 |
+
|
955 |
+
# FPN-3
|
956 |
+
|
957 |
+
[convolutional]
|
958 |
+
batch_normalize=1
|
959 |
+
filters=128
|
960 |
+
size=1
|
961 |
+
stride=1
|
962 |
+
pad=1
|
963 |
+
activation=silu
|
964 |
+
|
965 |
+
[upsample]
|
966 |
+
stride=2
|
967 |
+
|
968 |
+
[route]
|
969 |
+
layers = 48
|
970 |
+
|
971 |
+
[convolutional]
|
972 |
+
batch_normalize=1
|
973 |
+
filters=128
|
974 |
+
size=1
|
975 |
+
stride=1
|
976 |
+
pad=1
|
977 |
+
activation=silu
|
978 |
+
|
979 |
+
[route]
|
980 |
+
layers = -1, -3
|
981 |
+
|
982 |
+
[convolutional]
|
983 |
+
batch_normalize=1
|
984 |
+
filters=128
|
985 |
+
size=1
|
986 |
+
stride=1
|
987 |
+
pad=1
|
988 |
+
activation=silu
|
989 |
+
|
990 |
+
# Split
|
991 |
+
|
992 |
+
[convolutional]
|
993 |
+
batch_normalize=1
|
994 |
+
filters=128
|
995 |
+
size=1
|
996 |
+
stride=1
|
997 |
+
pad=1
|
998 |
+
activation=silu
|
999 |
+
|
1000 |
+
[route]
|
1001 |
+
layers = -2
|
1002 |
+
|
1003 |
+
# Plain Block
|
1004 |
+
|
1005 |
+
[convolutional]
|
1006 |
+
batch_normalize=1
|
1007 |
+
filters=128
|
1008 |
+
size=1
|
1009 |
+
stride=1
|
1010 |
+
pad=1
|
1011 |
+
activation=silu
|
1012 |
+
|
1013 |
+
[convolutional]
|
1014 |
+
batch_normalize=1
|
1015 |
+
size=3
|
1016 |
+
stride=1
|
1017 |
+
pad=1
|
1018 |
+
filters=128
|
1019 |
+
activation=silu
|
1020 |
+
|
1021 |
+
[convolutional]
|
1022 |
+
batch_normalize=1
|
1023 |
+
filters=128
|
1024 |
+
size=1
|
1025 |
+
stride=1
|
1026 |
+
pad=1
|
1027 |
+
activation=silu
|
1028 |
+
|
1029 |
+
[convolutional]
|
1030 |
+
batch_normalize=1
|
1031 |
+
size=3
|
1032 |
+
stride=1
|
1033 |
+
pad=1
|
1034 |
+
filters=128
|
1035 |
+
activation=silu
|
1036 |
+
|
1037 |
+
# Merge [-1, -(2k+2)]
|
1038 |
+
|
1039 |
+
[route]
|
1040 |
+
layers = -1, -6
|
1041 |
+
|
1042 |
+
# Transition last
|
1043 |
+
|
1044 |
+
# 141 (previous+6+4+2k)
|
1045 |
+
[convolutional]
|
1046 |
+
batch_normalize=1
|
1047 |
+
filters=128
|
1048 |
+
size=1
|
1049 |
+
stride=1
|
1050 |
+
pad=1
|
1051 |
+
activation=silu
|
1052 |
+
|
1053 |
+
|
1054 |
+
# PAN-4
|
1055 |
+
|
1056 |
+
[convolutional]
|
1057 |
+
batch_normalize=1
|
1058 |
+
size=3
|
1059 |
+
stride=2
|
1060 |
+
pad=1
|
1061 |
+
filters=256
|
1062 |
+
activation=silu
|
1063 |
+
|
1064 |
+
[route]
|
1065 |
+
layers = -1, 127
|
1066 |
+
|
1067 |
+
[convolutional]
|
1068 |
+
batch_normalize=1
|
1069 |
+
filters=256
|
1070 |
+
size=1
|
1071 |
+
stride=1
|
1072 |
+
pad=1
|
1073 |
+
activation=silu
|
1074 |
+
|
1075 |
+
# Split
|
1076 |
+
|
1077 |
+
[convolutional]
|
1078 |
+
batch_normalize=1
|
1079 |
+
filters=256
|
1080 |
+
size=1
|
1081 |
+
stride=1
|
1082 |
+
pad=1
|
1083 |
+
activation=silu
|
1084 |
+
|
1085 |
+
[route]
|
1086 |
+
layers = -2
|
1087 |
+
|
1088 |
+
# Plain Block
|
1089 |
+
|
1090 |
+
[convolutional]
|
1091 |
+
batch_normalize=1
|
1092 |
+
filters=256
|
1093 |
+
size=1
|
1094 |
+
stride=1
|
1095 |
+
pad=1
|
1096 |
+
activation=silu
|
1097 |
+
|
1098 |
+
[convolutional]
|
1099 |
+
batch_normalize=1
|
1100 |
+
size=3
|
1101 |
+
stride=1
|
1102 |
+
pad=1
|
1103 |
+
filters=256
|
1104 |
+
activation=silu
|
1105 |
+
|
1106 |
+
[convolutional]
|
1107 |
+
batch_normalize=1
|
1108 |
+
filters=256
|
1109 |
+
size=1
|
1110 |
+
stride=1
|
1111 |
+
pad=1
|
1112 |
+
activation=silu
|
1113 |
+
|
1114 |
+
[convolutional]
|
1115 |
+
batch_normalize=1
|
1116 |
+
size=3
|
1117 |
+
stride=1
|
1118 |
+
pad=1
|
1119 |
+
filters=256
|
1120 |
+
activation=silu
|
1121 |
+
|
1122 |
+
[route]
|
1123 |
+
layers = -1,-6
|
1124 |
+
|
1125 |
+
# Transition last
|
1126 |
+
|
1127 |
+
# 152 (previous+3+4+2k)
|
1128 |
+
[convolutional]
|
1129 |
+
batch_normalize=1
|
1130 |
+
filters=256
|
1131 |
+
size=1
|
1132 |
+
stride=1
|
1133 |
+
pad=1
|
1134 |
+
activation=silu
|
1135 |
+
|
1136 |
+
|
1137 |
+
# PAN-5
|
1138 |
+
|
1139 |
+
[convolutional]
|
1140 |
+
batch_normalize=1
|
1141 |
+
size=3
|
1142 |
+
stride=2
|
1143 |
+
pad=1
|
1144 |
+
filters=512
|
1145 |
+
activation=silu
|
1146 |
+
|
1147 |
+
[route]
|
1148 |
+
layers = -1, 113
|
1149 |
+
|
1150 |
+
[convolutional]
|
1151 |
+
batch_normalize=1
|
1152 |
+
filters=512
|
1153 |
+
size=1
|
1154 |
+
stride=1
|
1155 |
+
pad=1
|
1156 |
+
activation=silu
|
1157 |
+
|
1158 |
+
# Split
|
1159 |
+
|
1160 |
+
[convolutional]
|
1161 |
+
batch_normalize=1
|
1162 |
+
filters=512
|
1163 |
+
size=1
|
1164 |
+
stride=1
|
1165 |
+
pad=1
|
1166 |
+
activation=silu
|
1167 |
+
|
1168 |
+
[route]
|
1169 |
+
layers = -2
|
1170 |
+
|
1171 |
+
# Plain Block
|
1172 |
+
|
1173 |
+
[convolutional]
|
1174 |
+
batch_normalize=1
|
1175 |
+
filters=512
|
1176 |
+
size=1
|
1177 |
+
stride=1
|
1178 |
+
pad=1
|
1179 |
+
activation=silu
|
1180 |
+
|
1181 |
+
[convolutional]
|
1182 |
+
batch_normalize=1
|
1183 |
+
size=3
|
1184 |
+
stride=1
|
1185 |
+
pad=1
|
1186 |
+
filters=512
|
1187 |
+
activation=silu
|
1188 |
+
|
1189 |
+
[convolutional]
|
1190 |
+
batch_normalize=1
|
1191 |
+
filters=512
|
1192 |
+
size=1
|
1193 |
+
stride=1
|
1194 |
+
pad=1
|
1195 |
+
activation=silu
|
1196 |
+
|
1197 |
+
[convolutional]
|
1198 |
+
batch_normalize=1
|
1199 |
+
size=3
|
1200 |
+
stride=1
|
1201 |
+
pad=1
|
1202 |
+
filters=512
|
1203 |
+
activation=silu
|
1204 |
+
|
1205 |
+
[route]
|
1206 |
+
layers = -1,-6
|
1207 |
+
|
1208 |
+
# Transition last
|
1209 |
+
|
1210 |
+
# 163 (previous+3+4+2k)
|
1211 |
+
[convolutional]
|
1212 |
+
batch_normalize=1
|
1213 |
+
filters=512
|
1214 |
+
size=1
|
1215 |
+
stride=1
|
1216 |
+
pad=1
|
1217 |
+
activation=silu
|
1218 |
+
|
1219 |
+
# ============ End of Neck ============ #
|
1220 |
+
|
1221 |
+
# ============ Head ============ #
|
1222 |
+
|
1223 |
+
# YOLO-3
|
1224 |
+
|
1225 |
+
[route]
|
1226 |
+
layers = 141
|
1227 |
+
|
1228 |
+
[convolutional]
|
1229 |
+
batch_normalize=1
|
1230 |
+
size=3
|
1231 |
+
stride=1
|
1232 |
+
pad=1
|
1233 |
+
filters=256
|
1234 |
+
activation=silu
|
1235 |
+
|
1236 |
+
[convolutional]
|
1237 |
+
size=1
|
1238 |
+
stride=1
|
1239 |
+
pad=1
|
1240 |
+
filters=255
|
1241 |
+
activation=linear
|
1242 |
+
|
1243 |
+
[yolo]
|
1244 |
+
mask = 0,1,2
|
1245 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1246 |
+
classes=80
|
1247 |
+
num=9
|
1248 |
+
jitter=.3
|
1249 |
+
ignore_thresh = .7
|
1250 |
+
truth_thresh = 1
|
1251 |
+
random=1
|
1252 |
+
scale_x_y = 1.05
|
1253 |
+
iou_thresh=0.213
|
1254 |
+
cls_normalizer=1.0
|
1255 |
+
iou_normalizer=0.07
|
1256 |
+
iou_loss=ciou
|
1257 |
+
nms_kind=greedynms
|
1258 |
+
beta_nms=0.6
|
1259 |
+
|
1260 |
+
|
1261 |
+
# YOLO-4
|
1262 |
+
|
1263 |
+
[route]
|
1264 |
+
layers = 152
|
1265 |
+
|
1266 |
+
[convolutional]
|
1267 |
+
batch_normalize=1
|
1268 |
+
size=3
|
1269 |
+
stride=1
|
1270 |
+
pad=1
|
1271 |
+
filters=512
|
1272 |
+
activation=silu
|
1273 |
+
|
1274 |
+
[convolutional]
|
1275 |
+
size=1
|
1276 |
+
stride=1
|
1277 |
+
pad=1
|
1278 |
+
filters=255
|
1279 |
+
activation=linear
|
1280 |
+
|
1281 |
+
[yolo]
|
1282 |
+
mask = 3,4,5
|
1283 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1284 |
+
classes=80
|
1285 |
+
num=9
|
1286 |
+
jitter=.3
|
1287 |
+
ignore_thresh = .7
|
1288 |
+
truth_thresh = 1
|
1289 |
+
random=1
|
1290 |
+
scale_x_y = 1.05
|
1291 |
+
iou_thresh=0.213
|
1292 |
+
cls_normalizer=1.0
|
1293 |
+
iou_normalizer=0.07
|
1294 |
+
iou_loss=ciou
|
1295 |
+
nms_kind=greedynms
|
1296 |
+
beta_nms=0.6
|
1297 |
+
|
1298 |
+
|
1299 |
+
# YOLO-5
|
1300 |
+
|
1301 |
+
[route]
|
1302 |
+
layers = 163
|
1303 |
+
|
1304 |
+
[convolutional]
|
1305 |
+
batch_normalize=1
|
1306 |
+
size=3
|
1307 |
+
stride=1
|
1308 |
+
pad=1
|
1309 |
+
filters=1024
|
1310 |
+
activation=silu
|
1311 |
+
|
1312 |
+
[convolutional]
|
1313 |
+
size=1
|
1314 |
+
stride=1
|
1315 |
+
pad=1
|
1316 |
+
filters=255
|
1317 |
+
activation=linear
|
1318 |
+
|
1319 |
+
[yolo]
|
1320 |
+
mask = 6,7,8
|
1321 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1322 |
+
classes=80
|
1323 |
+
num=9
|
1324 |
+
jitter=.3
|
1325 |
+
ignore_thresh = .7
|
1326 |
+
truth_thresh = 1
|
1327 |
+
random=1
|
1328 |
+
scale_x_y = 1.05
|
1329 |
+
iou_thresh=0.213
|
1330 |
+
cls_normalizer=1.0
|
1331 |
+
iou_normalizer=0.07
|
1332 |
+
iou_loss=ciou
|
1333 |
+
nms_kind=greedynms
|
1334 |
+
beta_nms=0.6
|
cfg/yolov4_csp_x.cfg
ADDED
@@ -0,0 +1,1534 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[net]
|
2 |
+
# Testing
|
3 |
+
#batch=1
|
4 |
+
#subdivisions=1
|
5 |
+
# Training
|
6 |
+
batch=64
|
7 |
+
subdivisions=8
|
8 |
+
width=512
|
9 |
+
height=512
|
10 |
+
channels=3
|
11 |
+
momentum=0.949
|
12 |
+
decay=0.0005
|
13 |
+
angle=0
|
14 |
+
saturation = 1.5
|
15 |
+
exposure = 1.5
|
16 |
+
hue=.1
|
17 |
+
|
18 |
+
learning_rate=0.00261
|
19 |
+
burn_in=1000
|
20 |
+
max_batches = 500500
|
21 |
+
policy=steps
|
22 |
+
steps=400000,450000
|
23 |
+
scales=.1,.1
|
24 |
+
|
25 |
+
#cutmix=1
|
26 |
+
mosaic=1
|
27 |
+
|
28 |
+
|
29 |
+
# ============ Backbone ============ #
|
30 |
+
|
31 |
+
# Stem
|
32 |
+
|
33 |
+
# 0
|
34 |
+
[convolutional]
|
35 |
+
batch_normalize=1
|
36 |
+
filters=32
|
37 |
+
size=3
|
38 |
+
stride=1
|
39 |
+
pad=1
|
40 |
+
activation=silu
|
41 |
+
|
42 |
+
# P1
|
43 |
+
|
44 |
+
# Downsample
|
45 |
+
|
46 |
+
[convolutional]
|
47 |
+
batch_normalize=1
|
48 |
+
filters=80
|
49 |
+
size=3
|
50 |
+
stride=2
|
51 |
+
pad=1
|
52 |
+
activation=silu
|
53 |
+
|
54 |
+
# Residual Block
|
55 |
+
|
56 |
+
[convolutional]
|
57 |
+
batch_normalize=1
|
58 |
+
filters=40
|
59 |
+
size=1
|
60 |
+
stride=1
|
61 |
+
pad=1
|
62 |
+
activation=silu
|
63 |
+
|
64 |
+
[convolutional]
|
65 |
+
batch_normalize=1
|
66 |
+
filters=80
|
67 |
+
size=3
|
68 |
+
stride=1
|
69 |
+
pad=1
|
70 |
+
activation=silu
|
71 |
+
|
72 |
+
# 4 (previous+1+3k)
|
73 |
+
[shortcut]
|
74 |
+
from=-3
|
75 |
+
activation=linear
|
76 |
+
|
77 |
+
# P2
|
78 |
+
|
79 |
+
# Downsample
|
80 |
+
|
81 |
+
[convolutional]
|
82 |
+
batch_normalize=1
|
83 |
+
filters=160
|
84 |
+
size=3
|
85 |
+
stride=2
|
86 |
+
pad=1
|
87 |
+
activation=silu
|
88 |
+
|
89 |
+
# Split
|
90 |
+
|
91 |
+
[convolutional]
|
92 |
+
batch_normalize=1
|
93 |
+
filters=80
|
94 |
+
size=1
|
95 |
+
stride=1
|
96 |
+
pad=1
|
97 |
+
activation=silu
|
98 |
+
|
99 |
+
[route]
|
100 |
+
layers = -2
|
101 |
+
|
102 |
+
[convolutional]
|
103 |
+
batch_normalize=1
|
104 |
+
filters=80
|
105 |
+
size=1
|
106 |
+
stride=1
|
107 |
+
pad=1
|
108 |
+
activation=silu
|
109 |
+
|
110 |
+
# Residual Block
|
111 |
+
|
112 |
+
[convolutional]
|
113 |
+
batch_normalize=1
|
114 |
+
filters=80
|
115 |
+
size=1
|
116 |
+
stride=1
|
117 |
+
pad=1
|
118 |
+
activation=silu
|
119 |
+
|
120 |
+
[convolutional]
|
121 |
+
batch_normalize=1
|
122 |
+
filters=80
|
123 |
+
size=3
|
124 |
+
stride=1
|
125 |
+
pad=1
|
126 |
+
activation=silu
|
127 |
+
|
128 |
+
[shortcut]
|
129 |
+
from=-3
|
130 |
+
activation=linear
|
131 |
+
|
132 |
+
[convolutional]
|
133 |
+
batch_normalize=1
|
134 |
+
filters=80
|
135 |
+
size=1
|
136 |
+
stride=1
|
137 |
+
pad=1
|
138 |
+
activation=silu
|
139 |
+
|
140 |
+
[convolutional]
|
141 |
+
batch_normalize=1
|
142 |
+
filters=80
|
143 |
+
size=3
|
144 |
+
stride=1
|
145 |
+
pad=1
|
146 |
+
activation=silu
|
147 |
+
|
148 |
+
[shortcut]
|
149 |
+
from=-3
|
150 |
+
activation=linear
|
151 |
+
|
152 |
+
[convolutional]
|
153 |
+
batch_normalize=1
|
154 |
+
filters=80
|
155 |
+
size=1
|
156 |
+
stride=1
|
157 |
+
pad=1
|
158 |
+
activation=silu
|
159 |
+
|
160 |
+
[convolutional]
|
161 |
+
batch_normalize=1
|
162 |
+
filters=80
|
163 |
+
size=3
|
164 |
+
stride=1
|
165 |
+
pad=1
|
166 |
+
activation=silu
|
167 |
+
|
168 |
+
[shortcut]
|
169 |
+
from=-3
|
170 |
+
activation=linear
|
171 |
+
|
172 |
+
# Transition first
|
173 |
+
|
174 |
+
[convolutional]
|
175 |
+
batch_normalize=1
|
176 |
+
filters=80
|
177 |
+
size=1
|
178 |
+
stride=1
|
179 |
+
pad=1
|
180 |
+
activation=silu
|
181 |
+
|
182 |
+
# Merge [-1, -(3k+4)]
|
183 |
+
|
184 |
+
[route]
|
185 |
+
layers = -1,-13
|
186 |
+
|
187 |
+
# Transition last
|
188 |
+
|
189 |
+
# 20 (previous+7+3k)
|
190 |
+
[convolutional]
|
191 |
+
batch_normalize=1
|
192 |
+
filters=160
|
193 |
+
size=1
|
194 |
+
stride=1
|
195 |
+
pad=1
|
196 |
+
activation=silu
|
197 |
+
|
198 |
+
# P3
|
199 |
+
|
200 |
+
# Downsample
|
201 |
+
|
202 |
+
[convolutional]
|
203 |
+
batch_normalize=1
|
204 |
+
filters=320
|
205 |
+
size=3
|
206 |
+
stride=2
|
207 |
+
pad=1
|
208 |
+
activation=silu
|
209 |
+
|
210 |
+
# Split
|
211 |
+
|
212 |
+
[convolutional]
|
213 |
+
batch_normalize=1
|
214 |
+
filters=160
|
215 |
+
size=1
|
216 |
+
stride=1
|
217 |
+
pad=1
|
218 |
+
activation=silu
|
219 |
+
|
220 |
+
[route]
|
221 |
+
layers = -2
|
222 |
+
|
223 |
+
[convolutional]
|
224 |
+
batch_normalize=1
|
225 |
+
filters=160
|
226 |
+
size=1
|
227 |
+
stride=1
|
228 |
+
pad=1
|
229 |
+
activation=silu
|
230 |
+
|
231 |
+
# Residual Block
|
232 |
+
|
233 |
+
[convolutional]
|
234 |
+
batch_normalize=1
|
235 |
+
filters=160
|
236 |
+
size=1
|
237 |
+
stride=1
|
238 |
+
pad=1
|
239 |
+
activation=silu
|
240 |
+
|
241 |
+
[convolutional]
|
242 |
+
batch_normalize=1
|
243 |
+
filters=160
|
244 |
+
size=3
|
245 |
+
stride=1
|
246 |
+
pad=1
|
247 |
+
activation=silu
|
248 |
+
|
249 |
+
[shortcut]
|
250 |
+
from=-3
|
251 |
+
activation=linear
|
252 |
+
|
253 |
+
[convolutional]
|
254 |
+
batch_normalize=1
|
255 |
+
filters=160
|
256 |
+
size=1
|
257 |
+
stride=1
|
258 |
+
pad=1
|
259 |
+
activation=silu
|
260 |
+
|
261 |
+
[convolutional]
|
262 |
+
batch_normalize=1
|
263 |
+
filters=160
|
264 |
+
size=3
|
265 |
+
stride=1
|
266 |
+
pad=1
|
267 |
+
activation=silu
|
268 |
+
|
269 |
+
[shortcut]
|
270 |
+
from=-3
|
271 |
+
activation=linear
|
272 |
+
|
273 |
+
[convolutional]
|
274 |
+
batch_normalize=1
|
275 |
+
filters=160
|
276 |
+
size=1
|
277 |
+
stride=1
|
278 |
+
pad=1
|
279 |
+
activation=silu
|
280 |
+
|
281 |
+
[convolutional]
|
282 |
+
batch_normalize=1
|
283 |
+
filters=160
|
284 |
+
size=3
|
285 |
+
stride=1
|
286 |
+
pad=1
|
287 |
+
activation=silu
|
288 |
+
|
289 |
+
[shortcut]
|
290 |
+
from=-3
|
291 |
+
activation=linear
|
292 |
+
|
293 |
+
[convolutional]
|
294 |
+
batch_normalize=1
|
295 |
+
filters=160
|
296 |
+
size=1
|
297 |
+
stride=1
|
298 |
+
pad=1
|
299 |
+
activation=silu
|
300 |
+
|
301 |
+
[convolutional]
|
302 |
+
batch_normalize=1
|
303 |
+
filters=160
|
304 |
+
size=3
|
305 |
+
stride=1
|
306 |
+
pad=1
|
307 |
+
activation=silu
|
308 |
+
|
309 |
+
[shortcut]
|
310 |
+
from=-3
|
311 |
+
activation=linear
|
312 |
+
|
313 |
+
[convolutional]
|
314 |
+
batch_normalize=1
|
315 |
+
filters=160
|
316 |
+
size=1
|
317 |
+
stride=1
|
318 |
+
pad=1
|
319 |
+
activation=silu
|
320 |
+
|
321 |
+
[convolutional]
|
322 |
+
batch_normalize=1
|
323 |
+
filters=160
|
324 |
+
size=3
|
325 |
+
stride=1
|
326 |
+
pad=1
|
327 |
+
activation=silu
|
328 |
+
|
329 |
+
[shortcut]
|
330 |
+
from=-3
|
331 |
+
activation=linear
|
332 |
+
|
333 |
+
[convolutional]
|
334 |
+
batch_normalize=1
|
335 |
+
filters=160
|
336 |
+
size=1
|
337 |
+
stride=1
|
338 |
+
pad=1
|
339 |
+
activation=silu
|
340 |
+
|
341 |
+
[convolutional]
|
342 |
+
batch_normalize=1
|
343 |
+
filters=160
|
344 |
+
size=3
|
345 |
+
stride=1
|
346 |
+
pad=1
|
347 |
+
activation=silu
|
348 |
+
|
349 |
+
[shortcut]
|
350 |
+
from=-3
|
351 |
+
activation=linear
|
352 |
+
|
353 |
+
[convolutional]
|
354 |
+
batch_normalize=1
|
355 |
+
filters=160
|
356 |
+
size=1
|
357 |
+
stride=1
|
358 |
+
pad=1
|
359 |
+
activation=silu
|
360 |
+
|
361 |
+
[convolutional]
|
362 |
+
batch_normalize=1
|
363 |
+
filters=160
|
364 |
+
size=3
|
365 |
+
stride=1
|
366 |
+
pad=1
|
367 |
+
activation=silu
|
368 |
+
|
369 |
+
[shortcut]
|
370 |
+
from=-3
|
371 |
+
activation=linear
|
372 |
+
|
373 |
+
[convolutional]
|
374 |
+
batch_normalize=1
|
375 |
+
filters=160
|
376 |
+
size=1
|
377 |
+
stride=1
|
378 |
+
pad=1
|
379 |
+
activation=silu
|
380 |
+
|
381 |
+
[convolutional]
|
382 |
+
batch_normalize=1
|
383 |
+
filters=160
|
384 |
+
size=3
|
385 |
+
stride=1
|
386 |
+
pad=1
|
387 |
+
activation=silu
|
388 |
+
|
389 |
+
[shortcut]
|
390 |
+
from=-3
|
391 |
+
activation=linear
|
392 |
+
|
393 |
+
[convolutional]
|
394 |
+
batch_normalize=1
|
395 |
+
filters=160
|
396 |
+
size=1
|
397 |
+
stride=1
|
398 |
+
pad=1
|
399 |
+
activation=silu
|
400 |
+
|
401 |
+
[convolutional]
|
402 |
+
batch_normalize=1
|
403 |
+
filters=160
|
404 |
+
size=3
|
405 |
+
stride=1
|
406 |
+
pad=1
|
407 |
+
activation=silu
|
408 |
+
|
409 |
+
[shortcut]
|
410 |
+
from=-3
|
411 |
+
activation=linear
|
412 |
+
|
413 |
+
[convolutional]
|
414 |
+
batch_normalize=1
|
415 |
+
filters=160
|
416 |
+
size=1
|
417 |
+
stride=1
|
418 |
+
pad=1
|
419 |
+
activation=silu
|
420 |
+
|
421 |
+
[convolutional]
|
422 |
+
batch_normalize=1
|
423 |
+
filters=160
|
424 |
+
size=3
|
425 |
+
stride=1
|
426 |
+
pad=1
|
427 |
+
activation=silu
|
428 |
+
|
429 |
+
[shortcut]
|
430 |
+
from=-3
|
431 |
+
activation=linear
|
432 |
+
|
433 |
+
# Transition first
|
434 |
+
|
435 |
+
[convolutional]
|
436 |
+
batch_normalize=1
|
437 |
+
filters=160
|
438 |
+
size=1
|
439 |
+
stride=1
|
440 |
+
pad=1
|
441 |
+
activation=silu
|
442 |
+
|
443 |
+
# Merge [-1 -(4+3k)]
|
444 |
+
|
445 |
+
[route]
|
446 |
+
layers = -1,-34
|
447 |
+
|
448 |
+
# Transition last
|
449 |
+
|
450 |
+
# 57 (previous+7+3k)
|
451 |
+
[convolutional]
|
452 |
+
batch_normalize=1
|
453 |
+
filters=320
|
454 |
+
size=1
|
455 |
+
stride=1
|
456 |
+
pad=1
|
457 |
+
activation=silu
|
458 |
+
|
459 |
+
# P4
|
460 |
+
|
461 |
+
# Downsample
|
462 |
+
|
463 |
+
[convolutional]
|
464 |
+
batch_normalize=1
|
465 |
+
filters=640
|
466 |
+
size=3
|
467 |
+
stride=2
|
468 |
+
pad=1
|
469 |
+
activation=silu
|
470 |
+
|
471 |
+
# Split
|
472 |
+
|
473 |
+
[convolutional]
|
474 |
+
batch_normalize=1
|
475 |
+
filters=320
|
476 |
+
size=1
|
477 |
+
stride=1
|
478 |
+
pad=1
|
479 |
+
activation=silu
|
480 |
+
|
481 |
+
[route]
|
482 |
+
layers = -2
|
483 |
+
|
484 |
+
[convolutional]
|
485 |
+
batch_normalize=1
|
486 |
+
filters=320
|
487 |
+
size=1
|
488 |
+
stride=1
|
489 |
+
pad=1
|
490 |
+
activation=silu
|
491 |
+
|
492 |
+
# Residual Block
|
493 |
+
|
494 |
+
[convolutional]
|
495 |
+
batch_normalize=1
|
496 |
+
filters=320
|
497 |
+
size=1
|
498 |
+
stride=1
|
499 |
+
pad=1
|
500 |
+
activation=silu
|
501 |
+
|
502 |
+
[convolutional]
|
503 |
+
batch_normalize=1
|
504 |
+
filters=320
|
505 |
+
size=3
|
506 |
+
stride=1
|
507 |
+
pad=1
|
508 |
+
activation=silu
|
509 |
+
|
510 |
+
[shortcut]
|
511 |
+
from=-3
|
512 |
+
activation=linear
|
513 |
+
|
514 |
+
[convolutional]
|
515 |
+
batch_normalize=1
|
516 |
+
filters=320
|
517 |
+
size=1
|
518 |
+
stride=1
|
519 |
+
pad=1
|
520 |
+
activation=silu
|
521 |
+
|
522 |
+
[convolutional]
|
523 |
+
batch_normalize=1
|
524 |
+
filters=320
|
525 |
+
size=3
|
526 |
+
stride=1
|
527 |
+
pad=1
|
528 |
+
activation=silu
|
529 |
+
|
530 |
+
[shortcut]
|
531 |
+
from=-3
|
532 |
+
activation=linear
|
533 |
+
|
534 |
+
[convolutional]
|
535 |
+
batch_normalize=1
|
536 |
+
filters=320
|
537 |
+
size=1
|
538 |
+
stride=1
|
539 |
+
pad=1
|
540 |
+
activation=silu
|
541 |
+
|
542 |
+
[convolutional]
|
543 |
+
batch_normalize=1
|
544 |
+
filters=320
|
545 |
+
size=3
|
546 |
+
stride=1
|
547 |
+
pad=1
|
548 |
+
activation=silu
|
549 |
+
|
550 |
+
[shortcut]
|
551 |
+
from=-3
|
552 |
+
activation=linear
|
553 |
+
|
554 |
+
[convolutional]
|
555 |
+
batch_normalize=1
|
556 |
+
filters=320
|
557 |
+
size=1
|
558 |
+
stride=1
|
559 |
+
pad=1
|
560 |
+
activation=silu
|
561 |
+
|
562 |
+
[convolutional]
|
563 |
+
batch_normalize=1
|
564 |
+
filters=320
|
565 |
+
size=3
|
566 |
+
stride=1
|
567 |
+
pad=1
|
568 |
+
activation=silu
|
569 |
+
|
570 |
+
[shortcut]
|
571 |
+
from=-3
|
572 |
+
activation=linear
|
573 |
+
|
574 |
+
[convolutional]
|
575 |
+
batch_normalize=1
|
576 |
+
filters=320
|
577 |
+
size=1
|
578 |
+
stride=1
|
579 |
+
pad=1
|
580 |
+
activation=silu
|
581 |
+
|
582 |
+
[convolutional]
|
583 |
+
batch_normalize=1
|
584 |
+
filters=320
|
585 |
+
size=3
|
586 |
+
stride=1
|
587 |
+
pad=1
|
588 |
+
activation=silu
|
589 |
+
|
590 |
+
[shortcut]
|
591 |
+
from=-3
|
592 |
+
activation=linear
|
593 |
+
|
594 |
+
[convolutional]
|
595 |
+
batch_normalize=1
|
596 |
+
filters=320
|
597 |
+
size=1
|
598 |
+
stride=1
|
599 |
+
pad=1
|
600 |
+
activation=silu
|
601 |
+
|
602 |
+
[convolutional]
|
603 |
+
batch_normalize=1
|
604 |
+
filters=320
|
605 |
+
size=3
|
606 |
+
stride=1
|
607 |
+
pad=1
|
608 |
+
activation=silu
|
609 |
+
|
610 |
+
[shortcut]
|
611 |
+
from=-3
|
612 |
+
activation=linear
|
613 |
+
|
614 |
+
[convolutional]
|
615 |
+
batch_normalize=1
|
616 |
+
filters=320
|
617 |
+
size=1
|
618 |
+
stride=1
|
619 |
+
pad=1
|
620 |
+
activation=silu
|
621 |
+
|
622 |
+
[convolutional]
|
623 |
+
batch_normalize=1
|
624 |
+
filters=320
|
625 |
+
size=3
|
626 |
+
stride=1
|
627 |
+
pad=1
|
628 |
+
activation=silu
|
629 |
+
|
630 |
+
[shortcut]
|
631 |
+
from=-3
|
632 |
+
activation=linear
|
633 |
+
|
634 |
+
[convolutional]
|
635 |
+
batch_normalize=1
|
636 |
+
filters=320
|
637 |
+
size=1
|
638 |
+
stride=1
|
639 |
+
pad=1
|
640 |
+
activation=silu
|
641 |
+
|
642 |
+
[convolutional]
|
643 |
+
batch_normalize=1
|
644 |
+
filters=320
|
645 |
+
size=3
|
646 |
+
stride=1
|
647 |
+
pad=1
|
648 |
+
activation=silu
|
649 |
+
|
650 |
+
[shortcut]
|
651 |
+
from=-3
|
652 |
+
activation=linear
|
653 |
+
|
654 |
+
[convolutional]
|
655 |
+
batch_normalize=1
|
656 |
+
filters=320
|
657 |
+
size=1
|
658 |
+
stride=1
|
659 |
+
pad=1
|
660 |
+
activation=silu
|
661 |
+
|
662 |
+
[convolutional]
|
663 |
+
batch_normalize=1
|
664 |
+
filters=320
|
665 |
+
size=3
|
666 |
+
stride=1
|
667 |
+
pad=1
|
668 |
+
activation=silu
|
669 |
+
|
670 |
+
[shortcut]
|
671 |
+
from=-3
|
672 |
+
activation=linear
|
673 |
+
|
674 |
+
[convolutional]
|
675 |
+
batch_normalize=1
|
676 |
+
filters=320
|
677 |
+
size=1
|
678 |
+
stride=1
|
679 |
+
pad=1
|
680 |
+
activation=silu
|
681 |
+
|
682 |
+
[convolutional]
|
683 |
+
batch_normalize=1
|
684 |
+
filters=320
|
685 |
+
size=3
|
686 |
+
stride=1
|
687 |
+
pad=1
|
688 |
+
activation=silu
|
689 |
+
|
690 |
+
[shortcut]
|
691 |
+
from=-3
|
692 |
+
activation=linear
|
693 |
+
|
694 |
+
# Transition first
|
695 |
+
|
696 |
+
[convolutional]
|
697 |
+
batch_normalize=1
|
698 |
+
filters=320
|
699 |
+
size=1
|
700 |
+
stride=1
|
701 |
+
pad=1
|
702 |
+
activation=silu
|
703 |
+
|
704 |
+
# Merge [-1 -(3k+4)]
|
705 |
+
|
706 |
+
[route]
|
707 |
+
layers = -1,-34
|
708 |
+
|
709 |
+
# Transition last
|
710 |
+
|
711 |
+
# 94 (previous+7+3k)
|
712 |
+
[convolutional]
|
713 |
+
batch_normalize=1
|
714 |
+
filters=640
|
715 |
+
size=1
|
716 |
+
stride=1
|
717 |
+
pad=1
|
718 |
+
activation=silu
|
719 |
+
|
720 |
+
# P5
|
721 |
+
|
722 |
+
# Downsample
|
723 |
+
|
724 |
+
[convolutional]
|
725 |
+
batch_normalize=1
|
726 |
+
filters=1280
|
727 |
+
size=3
|
728 |
+
stride=2
|
729 |
+
pad=1
|
730 |
+
activation=silu
|
731 |
+
|
732 |
+
# Split
|
733 |
+
|
734 |
+
[convolutional]
|
735 |
+
batch_normalize=1
|
736 |
+
filters=640
|
737 |
+
size=1
|
738 |
+
stride=1
|
739 |
+
pad=1
|
740 |
+
activation=silu
|
741 |
+
|
742 |
+
[route]
|
743 |
+
layers = -2
|
744 |
+
|
745 |
+
[convolutional]
|
746 |
+
batch_normalize=1
|
747 |
+
filters=640
|
748 |
+
size=1
|
749 |
+
stride=1
|
750 |
+
pad=1
|
751 |
+
activation=silu
|
752 |
+
|
753 |
+
# Residual Block
|
754 |
+
|
755 |
+
[convolutional]
|
756 |
+
batch_normalize=1
|
757 |
+
filters=640
|
758 |
+
size=1
|
759 |
+
stride=1
|
760 |
+
pad=1
|
761 |
+
activation=silu
|
762 |
+
|
763 |
+
[convolutional]
|
764 |
+
batch_normalize=1
|
765 |
+
filters=640
|
766 |
+
size=3
|
767 |
+
stride=1
|
768 |
+
pad=1
|
769 |
+
activation=silu
|
770 |
+
|
771 |
+
[shortcut]
|
772 |
+
from=-3
|
773 |
+
activation=linear
|
774 |
+
|
775 |
+
[convolutional]
|
776 |
+
batch_normalize=1
|
777 |
+
filters=640
|
778 |
+
size=1
|
779 |
+
stride=1
|
780 |
+
pad=1
|
781 |
+
activation=silu
|
782 |
+
|
783 |
+
[convolutional]
|
784 |
+
batch_normalize=1
|
785 |
+
filters=640
|
786 |
+
size=3
|
787 |
+
stride=1
|
788 |
+
pad=1
|
789 |
+
activation=silu
|
790 |
+
|
791 |
+
[shortcut]
|
792 |
+
from=-3
|
793 |
+
activation=linear
|
794 |
+
|
795 |
+
[convolutional]
|
796 |
+
batch_normalize=1
|
797 |
+
filters=640
|
798 |
+
size=1
|
799 |
+
stride=1
|
800 |
+
pad=1
|
801 |
+
activation=silu
|
802 |
+
|
803 |
+
[convolutional]
|
804 |
+
batch_normalize=1
|
805 |
+
filters=640
|
806 |
+
size=3
|
807 |
+
stride=1
|
808 |
+
pad=1
|
809 |
+
activation=silu
|
810 |
+
|
811 |
+
[shortcut]
|
812 |
+
from=-3
|
813 |
+
activation=linear
|
814 |
+
|
815 |
+
[convolutional]
|
816 |
+
batch_normalize=1
|
817 |
+
filters=640
|
818 |
+
size=1
|
819 |
+
stride=1
|
820 |
+
pad=1
|
821 |
+
activation=silu
|
822 |
+
|
823 |
+
[convolutional]
|
824 |
+
batch_normalize=1
|
825 |
+
filters=640
|
826 |
+
size=3
|
827 |
+
stride=1
|
828 |
+
pad=1
|
829 |
+
activation=silu
|
830 |
+
|
831 |
+
[shortcut]
|
832 |
+
from=-3
|
833 |
+
activation=linear
|
834 |
+
|
835 |
+
[convolutional]
|
836 |
+
batch_normalize=1
|
837 |
+
filters=640
|
838 |
+
size=1
|
839 |
+
stride=1
|
840 |
+
pad=1
|
841 |
+
activation=silu
|
842 |
+
|
843 |
+
[convolutional]
|
844 |
+
batch_normalize=1
|
845 |
+
filters=640
|
846 |
+
size=3
|
847 |
+
stride=1
|
848 |
+
pad=1
|
849 |
+
activation=silu
|
850 |
+
|
851 |
+
[shortcut]
|
852 |
+
from=-3
|
853 |
+
activation=linear
|
854 |
+
|
855 |
+
# Transition first
|
856 |
+
|
857 |
+
[convolutional]
|
858 |
+
batch_normalize=1
|
859 |
+
filters=640
|
860 |
+
size=1
|
861 |
+
stride=1
|
862 |
+
pad=1
|
863 |
+
activation=silu
|
864 |
+
|
865 |
+
# Merge [-1 -(3k+4)]
|
866 |
+
|
867 |
+
[route]
|
868 |
+
layers = -1,-19
|
869 |
+
|
870 |
+
# Transition last
|
871 |
+
|
872 |
+
# 116 (previous+7+3k)
|
873 |
+
[convolutional]
|
874 |
+
batch_normalize=1
|
875 |
+
filters=1280
|
876 |
+
size=1
|
877 |
+
stride=1
|
878 |
+
pad=1
|
879 |
+
activation=silu
|
880 |
+
|
881 |
+
# ============ End of Backbone ============ #
|
882 |
+
|
883 |
+
# ============ Neck ============ #
|
884 |
+
|
885 |
+
# CSPSPP
|
886 |
+
|
887 |
+
[convolutional]
|
888 |
+
batch_normalize=1
|
889 |
+
filters=640
|
890 |
+
size=1
|
891 |
+
stride=1
|
892 |
+
pad=1
|
893 |
+
activation=silu
|
894 |
+
|
895 |
+
[route]
|
896 |
+
layers = -2
|
897 |
+
|
898 |
+
[convolutional]
|
899 |
+
batch_normalize=1
|
900 |
+
filters=640
|
901 |
+
size=1
|
902 |
+
stride=1
|
903 |
+
pad=1
|
904 |
+
activation=silu
|
905 |
+
|
906 |
+
[convolutional]
|
907 |
+
batch_normalize=1
|
908 |
+
size=3
|
909 |
+
stride=1
|
910 |
+
pad=1
|
911 |
+
filters=640
|
912 |
+
activation=silu
|
913 |
+
|
914 |
+
[convolutional]
|
915 |
+
batch_normalize=1
|
916 |
+
filters=640
|
917 |
+
size=1
|
918 |
+
stride=1
|
919 |
+
pad=1
|
920 |
+
activation=silu
|
921 |
+
|
922 |
+
### SPP ###
|
923 |
+
[maxpool]
|
924 |
+
stride=1
|
925 |
+
size=5
|
926 |
+
|
927 |
+
[route]
|
928 |
+
layers=-2
|
929 |
+
|
930 |
+
[maxpool]
|
931 |
+
stride=1
|
932 |
+
size=9
|
933 |
+
|
934 |
+
[route]
|
935 |
+
layers=-4
|
936 |
+
|
937 |
+
[maxpool]
|
938 |
+
stride=1
|
939 |
+
size=13
|
940 |
+
|
941 |
+
[route]
|
942 |
+
layers=-1,-3,-5,-6
|
943 |
+
### End SPP ###
|
944 |
+
|
945 |
+
[convolutional]
|
946 |
+
batch_normalize=1
|
947 |
+
filters=640
|
948 |
+
size=1
|
949 |
+
stride=1
|
950 |
+
pad=1
|
951 |
+
activation=silu
|
952 |
+
|
953 |
+
[convolutional]
|
954 |
+
batch_normalize=1
|
955 |
+
size=3
|
956 |
+
stride=1
|
957 |
+
pad=1
|
958 |
+
filters=640
|
959 |
+
activation=silu
|
960 |
+
|
961 |
+
[convolutional]
|
962 |
+
batch_normalize=1
|
963 |
+
filters=640
|
964 |
+
size=1
|
965 |
+
stride=1
|
966 |
+
pad=1
|
967 |
+
activation=silu
|
968 |
+
|
969 |
+
[convolutional]
|
970 |
+
batch_normalize=1
|
971 |
+
size=3
|
972 |
+
stride=1
|
973 |
+
pad=1
|
974 |
+
filters=640
|
975 |
+
activation=silu
|
976 |
+
|
977 |
+
[route]
|
978 |
+
layers = -1, -15
|
979 |
+
|
980 |
+
# 133 (previous+6+5+2k)
|
981 |
+
[convolutional]
|
982 |
+
batch_normalize=1
|
983 |
+
filters=640
|
984 |
+
size=1
|
985 |
+
stride=1
|
986 |
+
pad=1
|
987 |
+
activation=silu
|
988 |
+
|
989 |
+
# End of CSPSPP
|
990 |
+
|
991 |
+
|
992 |
+
# FPN-4
|
993 |
+
|
994 |
+
[convolutional]
|
995 |
+
batch_normalize=1
|
996 |
+
filters=320
|
997 |
+
size=1
|
998 |
+
stride=1
|
999 |
+
pad=1
|
1000 |
+
activation=silu
|
1001 |
+
|
1002 |
+
[upsample]
|
1003 |
+
stride=2
|
1004 |
+
|
1005 |
+
[route]
|
1006 |
+
layers = 94
|
1007 |
+
|
1008 |
+
[convolutional]
|
1009 |
+
batch_normalize=1
|
1010 |
+
filters=320
|
1011 |
+
size=1
|
1012 |
+
stride=1
|
1013 |
+
pad=1
|
1014 |
+
activation=silu
|
1015 |
+
|
1016 |
+
[route]
|
1017 |
+
layers = -1, -3
|
1018 |
+
|
1019 |
+
[convolutional]
|
1020 |
+
batch_normalize=1
|
1021 |
+
filters=320
|
1022 |
+
size=1
|
1023 |
+
stride=1
|
1024 |
+
pad=1
|
1025 |
+
activation=silu
|
1026 |
+
|
1027 |
+
# Split
|
1028 |
+
|
1029 |
+
[convolutional]
|
1030 |
+
batch_normalize=1
|
1031 |
+
filters=320
|
1032 |
+
size=1
|
1033 |
+
stride=1
|
1034 |
+
pad=1
|
1035 |
+
activation=silu
|
1036 |
+
|
1037 |
+
[route]
|
1038 |
+
layers = -2
|
1039 |
+
|
1040 |
+
# Plain Block
|
1041 |
+
|
1042 |
+
[convolutional]
|
1043 |
+
batch_normalize=1
|
1044 |
+
filters=320
|
1045 |
+
size=1
|
1046 |
+
stride=1
|
1047 |
+
pad=1
|
1048 |
+
activation=silu
|
1049 |
+
|
1050 |
+
[convolutional]
|
1051 |
+
batch_normalize=1
|
1052 |
+
size=3
|
1053 |
+
stride=1
|
1054 |
+
pad=1
|
1055 |
+
filters=320
|
1056 |
+
activation=silu
|
1057 |
+
|
1058 |
+
[convolutional]
|
1059 |
+
batch_normalize=1
|
1060 |
+
filters=320
|
1061 |
+
size=1
|
1062 |
+
stride=1
|
1063 |
+
pad=1
|
1064 |
+
activation=silu
|
1065 |
+
|
1066 |
+
[convolutional]
|
1067 |
+
batch_normalize=1
|
1068 |
+
size=3
|
1069 |
+
stride=1
|
1070 |
+
pad=1
|
1071 |
+
filters=320
|
1072 |
+
activation=silu
|
1073 |
+
|
1074 |
+
[convolutional]
|
1075 |
+
batch_normalize=1
|
1076 |
+
filters=320
|
1077 |
+
size=1
|
1078 |
+
stride=1
|
1079 |
+
pad=1
|
1080 |
+
activation=silu
|
1081 |
+
|
1082 |
+
[convolutional]
|
1083 |
+
batch_normalize=1
|
1084 |
+
size=3
|
1085 |
+
stride=1
|
1086 |
+
pad=1
|
1087 |
+
filters=320
|
1088 |
+
activation=silu
|
1089 |
+
|
1090 |
+
# Merge [-1, -(2k+2)]
|
1091 |
+
|
1092 |
+
[route]
|
1093 |
+
layers = -1, -8
|
1094 |
+
|
1095 |
+
# Transition last
|
1096 |
+
|
1097 |
+
# 149 (previous+6+4+2k)
|
1098 |
+
[convolutional]
|
1099 |
+
batch_normalize=1
|
1100 |
+
filters=320
|
1101 |
+
size=1
|
1102 |
+
stride=1
|
1103 |
+
pad=1
|
1104 |
+
activation=silu
|
1105 |
+
|
1106 |
+
|
1107 |
+
# FPN-3
|
1108 |
+
|
1109 |
+
[convolutional]
|
1110 |
+
batch_normalize=1
|
1111 |
+
filters=160
|
1112 |
+
size=1
|
1113 |
+
stride=1
|
1114 |
+
pad=1
|
1115 |
+
activation=silu
|
1116 |
+
|
1117 |
+
[upsample]
|
1118 |
+
stride=2
|
1119 |
+
|
1120 |
+
[route]
|
1121 |
+
layers = 57
|
1122 |
+
|
1123 |
+
[convolutional]
|
1124 |
+
batch_normalize=1
|
1125 |
+
filters=160
|
1126 |
+
size=1
|
1127 |
+
stride=1
|
1128 |
+
pad=1
|
1129 |
+
activation=silu
|
1130 |
+
|
1131 |
+
[route]
|
1132 |
+
layers = -1, -3
|
1133 |
+
|
1134 |
+
[convolutional]
|
1135 |
+
batch_normalize=1
|
1136 |
+
filters=160
|
1137 |
+
size=1
|
1138 |
+
stride=1
|
1139 |
+
pad=1
|
1140 |
+
activation=silu
|
1141 |
+
|
1142 |
+
# Split
|
1143 |
+
|
1144 |
+
[convolutional]
|
1145 |
+
batch_normalize=1
|
1146 |
+
filters=160
|
1147 |
+
size=1
|
1148 |
+
stride=1
|
1149 |
+
pad=1
|
1150 |
+
activation=silu
|
1151 |
+
|
1152 |
+
[route]
|
1153 |
+
layers = -2
|
1154 |
+
|
1155 |
+
# Plain Block
|
1156 |
+
|
1157 |
+
[convolutional]
|
1158 |
+
batch_normalize=1
|
1159 |
+
filters=160
|
1160 |
+
size=1
|
1161 |
+
stride=1
|
1162 |
+
pad=1
|
1163 |
+
activation=silu
|
1164 |
+
|
1165 |
+
[convolutional]
|
1166 |
+
batch_normalize=1
|
1167 |
+
size=3
|
1168 |
+
stride=1
|
1169 |
+
pad=1
|
1170 |
+
filters=160
|
1171 |
+
activation=silu
|
1172 |
+
|
1173 |
+
[convolutional]
|
1174 |
+
batch_normalize=1
|
1175 |
+
filters=160
|
1176 |
+
size=1
|
1177 |
+
stride=1
|
1178 |
+
pad=1
|
1179 |
+
activation=silu
|
1180 |
+
|
1181 |
+
[convolutional]
|
1182 |
+
batch_normalize=1
|
1183 |
+
size=3
|
1184 |
+
stride=1
|
1185 |
+
pad=1
|
1186 |
+
filters=160
|
1187 |
+
activation=silu
|
1188 |
+
|
1189 |
+
[convolutional]
|
1190 |
+
batch_normalize=1
|
1191 |
+
filters=160
|
1192 |
+
size=1
|
1193 |
+
stride=1
|
1194 |
+
pad=1
|
1195 |
+
activation=silu
|
1196 |
+
|
1197 |
+
[convolutional]
|
1198 |
+
batch_normalize=1
|
1199 |
+
size=3
|
1200 |
+
stride=1
|
1201 |
+
pad=1
|
1202 |
+
filters=160
|
1203 |
+
activation=silu
|
1204 |
+
|
1205 |
+
# Merge [-1, -(2k+2)]
|
1206 |
+
|
1207 |
+
[route]
|
1208 |
+
layers = -1, -8
|
1209 |
+
|
1210 |
+
# Transition last
|
1211 |
+
|
1212 |
+
# 165 (previous+6+4+2k)
|
1213 |
+
[convolutional]
|
1214 |
+
batch_normalize=1
|
1215 |
+
filters=160
|
1216 |
+
size=1
|
1217 |
+
stride=1
|
1218 |
+
pad=1
|
1219 |
+
activation=silu
|
1220 |
+
|
1221 |
+
|
1222 |
+
# PAN-4
|
1223 |
+
|
1224 |
+
[convolutional]
|
1225 |
+
batch_normalize=1
|
1226 |
+
size=3
|
1227 |
+
stride=2
|
1228 |
+
pad=1
|
1229 |
+
filters=320
|
1230 |
+
activation=silu
|
1231 |
+
|
1232 |
+
[route]
|
1233 |
+
layers = -1, 149
|
1234 |
+
|
1235 |
+
[convolutional]
|
1236 |
+
batch_normalize=1
|
1237 |
+
filters=320
|
1238 |
+
size=1
|
1239 |
+
stride=1
|
1240 |
+
pad=1
|
1241 |
+
activation=silu
|
1242 |
+
|
1243 |
+
# Split
|
1244 |
+
|
1245 |
+
[convolutional]
|
1246 |
+
batch_normalize=1
|
1247 |
+
filters=320
|
1248 |
+
size=1
|
1249 |
+
stride=1
|
1250 |
+
pad=1
|
1251 |
+
activation=silu
|
1252 |
+
|
1253 |
+
[route]
|
1254 |
+
layers = -2
|
1255 |
+
|
1256 |
+
# Plain Block
|
1257 |
+
|
1258 |
+
[convolutional]
|
1259 |
+
batch_normalize=1
|
1260 |
+
filters=320
|
1261 |
+
size=1
|
1262 |
+
stride=1
|
1263 |
+
pad=1
|
1264 |
+
activation=silu
|
1265 |
+
|
1266 |
+
[convolutional]
|
1267 |
+
batch_normalize=1
|
1268 |
+
size=3
|
1269 |
+
stride=1
|
1270 |
+
pad=1
|
1271 |
+
filters=320
|
1272 |
+
activation=silu
|
1273 |
+
|
1274 |
+
[convolutional]
|
1275 |
+
batch_normalize=1
|
1276 |
+
filters=320
|
1277 |
+
size=1
|
1278 |
+
stride=1
|
1279 |
+
pad=1
|
1280 |
+
activation=silu
|
1281 |
+
|
1282 |
+
[convolutional]
|
1283 |
+
batch_normalize=1
|
1284 |
+
size=3
|
1285 |
+
stride=1
|
1286 |
+
pad=1
|
1287 |
+
filters=320
|
1288 |
+
activation=silu
|
1289 |
+
|
1290 |
+
[convolutional]
|
1291 |
+
batch_normalize=1
|
1292 |
+
filters=320
|
1293 |
+
size=1
|
1294 |
+
stride=1
|
1295 |
+
pad=1
|
1296 |
+
activation=silu
|
1297 |
+
|
1298 |
+
[convolutional]
|
1299 |
+
batch_normalize=1
|
1300 |
+
size=3
|
1301 |
+
stride=1
|
1302 |
+
pad=1
|
1303 |
+
filters=320
|
1304 |
+
activation=silu
|
1305 |
+
|
1306 |
+
[route]
|
1307 |
+
layers = -1,-8
|
1308 |
+
|
1309 |
+
# Transition last
|
1310 |
+
|
1311 |
+
# 178 (previous+3+4+2k)
|
1312 |
+
[convolutional]
|
1313 |
+
batch_normalize=1
|
1314 |
+
filters=320
|
1315 |
+
size=1
|
1316 |
+
stride=1
|
1317 |
+
pad=1
|
1318 |
+
activation=silu
|
1319 |
+
|
1320 |
+
|
1321 |
+
# PAN-5
|
1322 |
+
|
1323 |
+
[convolutional]
|
1324 |
+
batch_normalize=1
|
1325 |
+
size=3
|
1326 |
+
stride=2
|
1327 |
+
pad=1
|
1328 |
+
filters=640
|
1329 |
+
activation=silu
|
1330 |
+
|
1331 |
+
[route]
|
1332 |
+
layers = -1, 133
|
1333 |
+
|
1334 |
+
[convolutional]
|
1335 |
+
batch_normalize=1
|
1336 |
+
filters=640
|
1337 |
+
size=1
|
1338 |
+
stride=1
|
1339 |
+
pad=1
|
1340 |
+
activation=silu
|
1341 |
+
|
1342 |
+
# Split
|
1343 |
+
|
1344 |
+
[convolutional]
|
1345 |
+
batch_normalize=1
|
1346 |
+
filters=640
|
1347 |
+
size=1
|
1348 |
+
stride=1
|
1349 |
+
pad=1
|
1350 |
+
activation=silu
|
1351 |
+
|
1352 |
+
[route]
|
1353 |
+
layers = -2
|
1354 |
+
|
1355 |
+
# Plain Block
|
1356 |
+
|
1357 |
+
[convolutional]
|
1358 |
+
batch_normalize=1
|
1359 |
+
filters=640
|
1360 |
+
size=1
|
1361 |
+
stride=1
|
1362 |
+
pad=1
|
1363 |
+
activation=silu
|
1364 |
+
|
1365 |
+
[convolutional]
|
1366 |
+
batch_normalize=1
|
1367 |
+
size=3
|
1368 |
+
stride=1
|
1369 |
+
pad=1
|
1370 |
+
filters=640
|
1371 |
+
activation=silu
|
1372 |
+
|
1373 |
+
[convolutional]
|
1374 |
+
batch_normalize=1
|
1375 |
+
filters=640
|
1376 |
+
size=1
|
1377 |
+
stride=1
|
1378 |
+
pad=1
|
1379 |
+
activation=silu
|
1380 |
+
|
1381 |
+
[convolutional]
|
1382 |
+
batch_normalize=1
|
1383 |
+
size=3
|
1384 |
+
stride=1
|
1385 |
+
pad=1
|
1386 |
+
filters=640
|
1387 |
+
activation=silu
|
1388 |
+
|
1389 |
+
[convolutional]
|
1390 |
+
batch_normalize=1
|
1391 |
+
filters=640
|
1392 |
+
size=1
|
1393 |
+
stride=1
|
1394 |
+
pad=1
|
1395 |
+
activation=silu
|
1396 |
+
|
1397 |
+
[convolutional]
|
1398 |
+
batch_normalize=1
|
1399 |
+
size=3
|
1400 |
+
stride=1
|
1401 |
+
pad=1
|
1402 |
+
filters=640
|
1403 |
+
activation=silu
|
1404 |
+
|
1405 |
+
[route]
|
1406 |
+
layers = -1,-8
|
1407 |
+
|
1408 |
+
# Transition last
|
1409 |
+
|
1410 |
+
# 191 (previous+3+4+2k)
|
1411 |
+
[convolutional]
|
1412 |
+
batch_normalize=1
|
1413 |
+
filters=640
|
1414 |
+
size=1
|
1415 |
+
stride=1
|
1416 |
+
pad=1
|
1417 |
+
activation=silu
|
1418 |
+
|
1419 |
+
# ============ End of Neck ============ #
|
1420 |
+
|
1421 |
+
# ============ Head ============ #
|
1422 |
+
|
1423 |
+
# YOLO-3
|
1424 |
+
|
1425 |
+
[route]
|
1426 |
+
layers = 165
|
1427 |
+
|
1428 |
+
[convolutional]
|
1429 |
+
batch_normalize=1
|
1430 |
+
size=3
|
1431 |
+
stride=1
|
1432 |
+
pad=1
|
1433 |
+
filters=320
|
1434 |
+
activation=silu
|
1435 |
+
|
1436 |
+
[convolutional]
|
1437 |
+
size=1
|
1438 |
+
stride=1
|
1439 |
+
pad=1
|
1440 |
+
filters=255
|
1441 |
+
activation=linear
|
1442 |
+
|
1443 |
+
[yolo]
|
1444 |
+
mask = 0,1,2
|
1445 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1446 |
+
classes=80
|
1447 |
+
num=9
|
1448 |
+
jitter=.3
|
1449 |
+
ignore_thresh = .7
|
1450 |
+
truth_thresh = 1
|
1451 |
+
random=1
|
1452 |
+
scale_x_y = 1.05
|
1453 |
+
iou_thresh=0.213
|
1454 |
+
cls_normalizer=1.0
|
1455 |
+
iou_normalizer=0.07
|
1456 |
+
iou_loss=ciou
|
1457 |
+
nms_kind=greedynms
|
1458 |
+
beta_nms=0.6
|
1459 |
+
|
1460 |
+
|
1461 |
+
# YOLO-4
|
1462 |
+
|
1463 |
+
[route]
|
1464 |
+
layers = 178
|
1465 |
+
|
1466 |
+
[convolutional]
|
1467 |
+
batch_normalize=1
|
1468 |
+
size=3
|
1469 |
+
stride=1
|
1470 |
+
pad=1
|
1471 |
+
filters=640
|
1472 |
+
activation=silu
|
1473 |
+
|
1474 |
+
[convolutional]
|
1475 |
+
size=1
|
1476 |
+
stride=1
|
1477 |
+
pad=1
|
1478 |
+
filters=255
|
1479 |
+
activation=linear
|
1480 |
+
|
1481 |
+
[yolo]
|
1482 |
+
mask = 3,4,5
|
1483 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1484 |
+
classes=80
|
1485 |
+
num=9
|
1486 |
+
jitter=.3
|
1487 |
+
ignore_thresh = .7
|
1488 |
+
truth_thresh = 1
|
1489 |
+
random=1
|
1490 |
+
scale_x_y = 1.05
|
1491 |
+
iou_thresh=0.213
|
1492 |
+
cls_normalizer=1.0
|
1493 |
+
iou_normalizer=0.07
|
1494 |
+
iou_loss=ciou
|
1495 |
+
nms_kind=greedynms
|
1496 |
+
beta_nms=0.6
|
1497 |
+
|
1498 |
+
|
1499 |
+
# YOLO-5
|
1500 |
+
|
1501 |
+
[route]
|
1502 |
+
layers = 191
|
1503 |
+
|
1504 |
+
[convolutional]
|
1505 |
+
batch_normalize=1
|
1506 |
+
size=3
|
1507 |
+
stride=1
|
1508 |
+
pad=1
|
1509 |
+
filters=1280
|
1510 |
+
activation=silu
|
1511 |
+
|
1512 |
+
[convolutional]
|
1513 |
+
size=1
|
1514 |
+
stride=1
|
1515 |
+
pad=1
|
1516 |
+
filters=255
|
1517 |
+
activation=linear
|
1518 |
+
|
1519 |
+
[yolo]
|
1520 |
+
mask = 6,7,8
|
1521 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1522 |
+
classes=80
|
1523 |
+
num=9
|
1524 |
+
jitter=.3
|
1525 |
+
ignore_thresh = .7
|
1526 |
+
truth_thresh = 1
|
1527 |
+
random=1
|
1528 |
+
scale_x_y = 1.05
|
1529 |
+
iou_thresh=0.213
|
1530 |
+
cls_normalizer=1.0
|
1531 |
+
iou_normalizer=0.07
|
1532 |
+
iou_loss=ciou
|
1533 |
+
nms_kind=greedynms
|
1534 |
+
beta_nms=0.6
|
cfg/yolov4_p6.cfg
ADDED
@@ -0,0 +1,2260 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[net]
|
2 |
+
batch=64
|
3 |
+
subdivisions=8
|
4 |
+
width=1280
|
5 |
+
height=1280
|
6 |
+
channels=3
|
7 |
+
momentum=0.949
|
8 |
+
decay=0.0005
|
9 |
+
angle=0
|
10 |
+
saturation = 1.5
|
11 |
+
exposure = 1.5
|
12 |
+
hue=.1
|
13 |
+
|
14 |
+
learning_rate=0.00261
|
15 |
+
burn_in=1000
|
16 |
+
max_batches = 500500
|
17 |
+
policy=steps
|
18 |
+
steps=400000,450000
|
19 |
+
scales=.1,.1
|
20 |
+
|
21 |
+
mosaic=1
|
22 |
+
|
23 |
+
|
24 |
+
# ============ Backbone ============ #
|
25 |
+
|
26 |
+
# Stem
|
27 |
+
|
28 |
+
# 0
|
29 |
+
[convolutional]
|
30 |
+
batch_normalize=1
|
31 |
+
filters=32
|
32 |
+
size=3
|
33 |
+
stride=1
|
34 |
+
pad=1
|
35 |
+
activation=mish
|
36 |
+
|
37 |
+
|
38 |
+
# P1
|
39 |
+
|
40 |
+
# Downsample
|
41 |
+
|
42 |
+
[convolutional]
|
43 |
+
batch_normalize=1
|
44 |
+
filters=64
|
45 |
+
size=3
|
46 |
+
stride=2
|
47 |
+
pad=1
|
48 |
+
activation=mish
|
49 |
+
|
50 |
+
# Split
|
51 |
+
|
52 |
+
[convolutional]
|
53 |
+
batch_normalize=1
|
54 |
+
filters=32
|
55 |
+
size=1
|
56 |
+
stride=1
|
57 |
+
pad=1
|
58 |
+
activation=mish
|
59 |
+
|
60 |
+
[route]
|
61 |
+
layers = -2
|
62 |
+
|
63 |
+
[convolutional]
|
64 |
+
batch_normalize=1
|
65 |
+
filters=32
|
66 |
+
size=1
|
67 |
+
stride=1
|
68 |
+
pad=1
|
69 |
+
activation=mish
|
70 |
+
|
71 |
+
# Residual Block
|
72 |
+
|
73 |
+
[convolutional]
|
74 |
+
batch_normalize=1
|
75 |
+
filters=32
|
76 |
+
size=1
|
77 |
+
stride=1
|
78 |
+
pad=1
|
79 |
+
activation=mish
|
80 |
+
|
81 |
+
[convolutional]
|
82 |
+
batch_normalize=1
|
83 |
+
filters=32
|
84 |
+
size=3
|
85 |
+
stride=1
|
86 |
+
pad=1
|
87 |
+
activation=mish
|
88 |
+
|
89 |
+
[shortcut]
|
90 |
+
from=-3
|
91 |
+
activation=linear
|
92 |
+
|
93 |
+
# Transition first
|
94 |
+
|
95 |
+
[convolutional]
|
96 |
+
batch_normalize=1
|
97 |
+
filters=32
|
98 |
+
size=1
|
99 |
+
stride=1
|
100 |
+
pad=1
|
101 |
+
activation=mish
|
102 |
+
|
103 |
+
# Merge [-1, -(3k+4)]
|
104 |
+
|
105 |
+
[route]
|
106 |
+
layers = -1,-7
|
107 |
+
|
108 |
+
# Transition last
|
109 |
+
|
110 |
+
# 10 (previous+7+3k)
|
111 |
+
[convolutional]
|
112 |
+
batch_normalize=1
|
113 |
+
filters=64
|
114 |
+
size=1
|
115 |
+
stride=1
|
116 |
+
pad=1
|
117 |
+
activation=mish
|
118 |
+
|
119 |
+
|
120 |
+
# P2
|
121 |
+
|
122 |
+
# Downsample
|
123 |
+
|
124 |
+
[convolutional]
|
125 |
+
batch_normalize=1
|
126 |
+
filters=128
|
127 |
+
size=3
|
128 |
+
stride=2
|
129 |
+
pad=1
|
130 |
+
activation=mish
|
131 |
+
|
132 |
+
# Split
|
133 |
+
|
134 |
+
[convolutional]
|
135 |
+
batch_normalize=1
|
136 |
+
filters=64
|
137 |
+
size=1
|
138 |
+
stride=1
|
139 |
+
pad=1
|
140 |
+
activation=mish
|
141 |
+
|
142 |
+
[route]
|
143 |
+
layers = -2
|
144 |
+
|
145 |
+
[convolutional]
|
146 |
+
batch_normalize=1
|
147 |
+
filters=64
|
148 |
+
size=1
|
149 |
+
stride=1
|
150 |
+
pad=1
|
151 |
+
activation=mish
|
152 |
+
|
153 |
+
# Residual Block
|
154 |
+
|
155 |
+
[convolutional]
|
156 |
+
batch_normalize=1
|
157 |
+
filters=64
|
158 |
+
size=1
|
159 |
+
stride=1
|
160 |
+
pad=1
|
161 |
+
activation=mish
|
162 |
+
|
163 |
+
[convolutional]
|
164 |
+
batch_normalize=1
|
165 |
+
filters=64
|
166 |
+
size=3
|
167 |
+
stride=1
|
168 |
+
pad=1
|
169 |
+
activation=mish
|
170 |
+
|
171 |
+
[shortcut]
|
172 |
+
from=-3
|
173 |
+
activation=linear
|
174 |
+
|
175 |
+
[convolutional]
|
176 |
+
batch_normalize=1
|
177 |
+
filters=64
|
178 |
+
size=1
|
179 |
+
stride=1
|
180 |
+
pad=1
|
181 |
+
activation=mish
|
182 |
+
|
183 |
+
[convolutional]
|
184 |
+
batch_normalize=1
|
185 |
+
filters=64
|
186 |
+
size=3
|
187 |
+
stride=1
|
188 |
+
pad=1
|
189 |
+
activation=mish
|
190 |
+
|
191 |
+
[shortcut]
|
192 |
+
from=-3
|
193 |
+
activation=linear
|
194 |
+
|
195 |
+
[convolutional]
|
196 |
+
batch_normalize=1
|
197 |
+
filters=64
|
198 |
+
size=1
|
199 |
+
stride=1
|
200 |
+
pad=1
|
201 |
+
activation=mish
|
202 |
+
|
203 |
+
[convolutional]
|
204 |
+
batch_normalize=1
|
205 |
+
filters=64
|
206 |
+
size=3
|
207 |
+
stride=1
|
208 |
+
pad=1
|
209 |
+
activation=mish
|
210 |
+
|
211 |
+
[shortcut]
|
212 |
+
from=-3
|
213 |
+
activation=linear
|
214 |
+
|
215 |
+
# Transition first
|
216 |
+
|
217 |
+
[convolutional]
|
218 |
+
batch_normalize=1
|
219 |
+
filters=64
|
220 |
+
size=1
|
221 |
+
stride=1
|
222 |
+
pad=1
|
223 |
+
activation=mish
|
224 |
+
|
225 |
+
# Merge [-1, -(3k+4)]
|
226 |
+
|
227 |
+
[route]
|
228 |
+
layers = -1,-13
|
229 |
+
|
230 |
+
# Transition last
|
231 |
+
|
232 |
+
# 26 (previous+7+3k)
|
233 |
+
[convolutional]
|
234 |
+
batch_normalize=1
|
235 |
+
filters=128
|
236 |
+
size=1
|
237 |
+
stride=1
|
238 |
+
pad=1
|
239 |
+
activation=mish
|
240 |
+
|
241 |
+
|
242 |
+
# P3
|
243 |
+
|
244 |
+
# Downsample
|
245 |
+
|
246 |
+
[convolutional]
|
247 |
+
batch_normalize=1
|
248 |
+
filters=256
|
249 |
+
size=3
|
250 |
+
stride=2
|
251 |
+
pad=1
|
252 |
+
activation=mish
|
253 |
+
|
254 |
+
# Split
|
255 |
+
|
256 |
+
[convolutional]
|
257 |
+
batch_normalize=1
|
258 |
+
filters=128
|
259 |
+
size=1
|
260 |
+
stride=1
|
261 |
+
pad=1
|
262 |
+
activation=mish
|
263 |
+
|
264 |
+
[route]
|
265 |
+
layers = -2
|
266 |
+
|
267 |
+
[convolutional]
|
268 |
+
batch_normalize=1
|
269 |
+
filters=128
|
270 |
+
size=1
|
271 |
+
stride=1
|
272 |
+
pad=1
|
273 |
+
activation=mish
|
274 |
+
|
275 |
+
# Residual Block
|
276 |
+
|
277 |
+
[convolutional]
|
278 |
+
batch_normalize=1
|
279 |
+
filters=128
|
280 |
+
size=1
|
281 |
+
stride=1
|
282 |
+
pad=1
|
283 |
+
activation=mish
|
284 |
+
|
285 |
+
[convolutional]
|
286 |
+
batch_normalize=1
|
287 |
+
filters=128
|
288 |
+
size=3
|
289 |
+
stride=1
|
290 |
+
pad=1
|
291 |
+
activation=mish
|
292 |
+
|
293 |
+
[shortcut]
|
294 |
+
from=-3
|
295 |
+
activation=linear
|
296 |
+
|
297 |
+
[convolutional]
|
298 |
+
batch_normalize=1
|
299 |
+
filters=128
|
300 |
+
size=1
|
301 |
+
stride=1
|
302 |
+
pad=1
|
303 |
+
activation=mish
|
304 |
+
|
305 |
+
[convolutional]
|
306 |
+
batch_normalize=1
|
307 |
+
filters=128
|
308 |
+
size=3
|
309 |
+
stride=1
|
310 |
+
pad=1
|
311 |
+
activation=mish
|
312 |
+
|
313 |
+
[shortcut]
|
314 |
+
from=-3
|
315 |
+
activation=linear
|
316 |
+
|
317 |
+
[convolutional]
|
318 |
+
batch_normalize=1
|
319 |
+
filters=128
|
320 |
+
size=1
|
321 |
+
stride=1
|
322 |
+
pad=1
|
323 |
+
activation=mish
|
324 |
+
|
325 |
+
[convolutional]
|
326 |
+
batch_normalize=1
|
327 |
+
filters=128
|
328 |
+
size=3
|
329 |
+
stride=1
|
330 |
+
pad=1
|
331 |
+
activation=mish
|
332 |
+
|
333 |
+
[shortcut]
|
334 |
+
from=-3
|
335 |
+
activation=linear
|
336 |
+
|
337 |
+
[convolutional]
|
338 |
+
batch_normalize=1
|
339 |
+
filters=128
|
340 |
+
size=1
|
341 |
+
stride=1
|
342 |
+
pad=1
|
343 |
+
activation=mish
|
344 |
+
|
345 |
+
[convolutional]
|
346 |
+
batch_normalize=1
|
347 |
+
filters=128
|
348 |
+
size=3
|
349 |
+
stride=1
|
350 |
+
pad=1
|
351 |
+
activation=mish
|
352 |
+
|
353 |
+
[shortcut]
|
354 |
+
from=-3
|
355 |
+
activation=linear
|
356 |
+
|
357 |
+
[convolutional]
|
358 |
+
batch_normalize=1
|
359 |
+
filters=128
|
360 |
+
size=1
|
361 |
+
stride=1
|
362 |
+
pad=1
|
363 |
+
activation=mish
|
364 |
+
|
365 |
+
[convolutional]
|
366 |
+
batch_normalize=1
|
367 |
+
filters=128
|
368 |
+
size=3
|
369 |
+
stride=1
|
370 |
+
pad=1
|
371 |
+
activation=mish
|
372 |
+
|
373 |
+
[shortcut]
|
374 |
+
from=-3
|
375 |
+
activation=linear
|
376 |
+
|
377 |
+
[convolutional]
|
378 |
+
batch_normalize=1
|
379 |
+
filters=128
|
380 |
+
size=1
|
381 |
+
stride=1
|
382 |
+
pad=1
|
383 |
+
activation=mish
|
384 |
+
|
385 |
+
[convolutional]
|
386 |
+
batch_normalize=1
|
387 |
+
filters=128
|
388 |
+
size=3
|
389 |
+
stride=1
|
390 |
+
pad=1
|
391 |
+
activation=mish
|
392 |
+
|
393 |
+
[shortcut]
|
394 |
+
from=-3
|
395 |
+
activation=linear
|
396 |
+
|
397 |
+
[convolutional]
|
398 |
+
batch_normalize=1
|
399 |
+
filters=128
|
400 |
+
size=1
|
401 |
+
stride=1
|
402 |
+
pad=1
|
403 |
+
activation=mish
|
404 |
+
|
405 |
+
[convolutional]
|
406 |
+
batch_normalize=1
|
407 |
+
filters=128
|
408 |
+
size=3
|
409 |
+
stride=1
|
410 |
+
pad=1
|
411 |
+
activation=mish
|
412 |
+
|
413 |
+
[shortcut]
|
414 |
+
from=-3
|
415 |
+
activation=linear
|
416 |
+
|
417 |
+
[convolutional]
|
418 |
+
batch_normalize=1
|
419 |
+
filters=128
|
420 |
+
size=1
|
421 |
+
stride=1
|
422 |
+
pad=1
|
423 |
+
activation=mish
|
424 |
+
|
425 |
+
[convolutional]
|
426 |
+
batch_normalize=1
|
427 |
+
filters=128
|
428 |
+
size=3
|
429 |
+
stride=1
|
430 |
+
pad=1
|
431 |
+
activation=mish
|
432 |
+
|
433 |
+
[shortcut]
|
434 |
+
from=-3
|
435 |
+
activation=linear
|
436 |
+
|
437 |
+
[convolutional]
|
438 |
+
batch_normalize=1
|
439 |
+
filters=128
|
440 |
+
size=1
|
441 |
+
stride=1
|
442 |
+
pad=1
|
443 |
+
activation=mish
|
444 |
+
|
445 |
+
[convolutional]
|
446 |
+
batch_normalize=1
|
447 |
+
filters=128
|
448 |
+
size=3
|
449 |
+
stride=1
|
450 |
+
pad=1
|
451 |
+
activation=mish
|
452 |
+
|
453 |
+
[shortcut]
|
454 |
+
from=-3
|
455 |
+
activation=linear
|
456 |
+
|
457 |
+
[convolutional]
|
458 |
+
batch_normalize=1
|
459 |
+
filters=128
|
460 |
+
size=1
|
461 |
+
stride=1
|
462 |
+
pad=1
|
463 |
+
activation=mish
|
464 |
+
|
465 |
+
[convolutional]
|
466 |
+
batch_normalize=1
|
467 |
+
filters=128
|
468 |
+
size=3
|
469 |
+
stride=1
|
470 |
+
pad=1
|
471 |
+
activation=mish
|
472 |
+
|
473 |
+
[shortcut]
|
474 |
+
from=-3
|
475 |
+
activation=linear
|
476 |
+
|
477 |
+
[convolutional]
|
478 |
+
batch_normalize=1
|
479 |
+
filters=128
|
480 |
+
size=1
|
481 |
+
stride=1
|
482 |
+
pad=1
|
483 |
+
activation=mish
|
484 |
+
|
485 |
+
[convolutional]
|
486 |
+
batch_normalize=1
|
487 |
+
filters=128
|
488 |
+
size=3
|
489 |
+
stride=1
|
490 |
+
pad=1
|
491 |
+
activation=mish
|
492 |
+
|
493 |
+
[shortcut]
|
494 |
+
from=-3
|
495 |
+
activation=linear
|
496 |
+
|
497 |
+
[convolutional]
|
498 |
+
batch_normalize=1
|
499 |
+
filters=128
|
500 |
+
size=1
|
501 |
+
stride=1
|
502 |
+
pad=1
|
503 |
+
activation=mish
|
504 |
+
|
505 |
+
[convolutional]
|
506 |
+
batch_normalize=1
|
507 |
+
filters=128
|
508 |
+
size=3
|
509 |
+
stride=1
|
510 |
+
pad=1
|
511 |
+
activation=mish
|
512 |
+
|
513 |
+
[shortcut]
|
514 |
+
from=-3
|
515 |
+
activation=linear
|
516 |
+
|
517 |
+
[convolutional]
|
518 |
+
batch_normalize=1
|
519 |
+
filters=128
|
520 |
+
size=1
|
521 |
+
stride=1
|
522 |
+
pad=1
|
523 |
+
activation=mish
|
524 |
+
|
525 |
+
[convolutional]
|
526 |
+
batch_normalize=1
|
527 |
+
filters=128
|
528 |
+
size=3
|
529 |
+
stride=1
|
530 |
+
pad=1
|
531 |
+
activation=mish
|
532 |
+
|
533 |
+
[shortcut]
|
534 |
+
from=-3
|
535 |
+
activation=linear
|
536 |
+
|
537 |
+
[convolutional]
|
538 |
+
batch_normalize=1
|
539 |
+
filters=128
|
540 |
+
size=1
|
541 |
+
stride=1
|
542 |
+
pad=1
|
543 |
+
activation=mish
|
544 |
+
|
545 |
+
[convolutional]
|
546 |
+
batch_normalize=1
|
547 |
+
filters=128
|
548 |
+
size=3
|
549 |
+
stride=1
|
550 |
+
pad=1
|
551 |
+
activation=mish
|
552 |
+
|
553 |
+
[shortcut]
|
554 |
+
from=-3
|
555 |
+
activation=linear
|
556 |
+
|
557 |
+
[convolutional]
|
558 |
+
batch_normalize=1
|
559 |
+
filters=128
|
560 |
+
size=1
|
561 |
+
stride=1
|
562 |
+
pad=1
|
563 |
+
activation=mish
|
564 |
+
|
565 |
+
[convolutional]
|
566 |
+
batch_normalize=1
|
567 |
+
filters=128
|
568 |
+
size=3
|
569 |
+
stride=1
|
570 |
+
pad=1
|
571 |
+
activation=mish
|
572 |
+
|
573 |
+
[shortcut]
|
574 |
+
from=-3
|
575 |
+
activation=linear
|
576 |
+
|
577 |
+
# Transition first
|
578 |
+
|
579 |
+
[convolutional]
|
580 |
+
batch_normalize=1
|
581 |
+
filters=128
|
582 |
+
size=1
|
583 |
+
stride=1
|
584 |
+
pad=1
|
585 |
+
activation=mish
|
586 |
+
|
587 |
+
# Merge [-1, -(3k+4)]
|
588 |
+
|
589 |
+
[route]
|
590 |
+
layers = -1,-49
|
591 |
+
|
592 |
+
# Transition last
|
593 |
+
|
594 |
+
# 78 (previous+7+3k)
|
595 |
+
[convolutional]
|
596 |
+
batch_normalize=1
|
597 |
+
filters=256
|
598 |
+
size=1
|
599 |
+
stride=1
|
600 |
+
pad=1
|
601 |
+
activation=mish
|
602 |
+
|
603 |
+
|
604 |
+
# P4
|
605 |
+
|
606 |
+
# Downsample
|
607 |
+
|
608 |
+
[convolutional]
|
609 |
+
batch_normalize=1
|
610 |
+
filters=512
|
611 |
+
size=3
|
612 |
+
stride=2
|
613 |
+
pad=1
|
614 |
+
activation=mish
|
615 |
+
|
616 |
+
# Split
|
617 |
+
|
618 |
+
[convolutional]
|
619 |
+
batch_normalize=1
|
620 |
+
filters=256
|
621 |
+
size=1
|
622 |
+
stride=1
|
623 |
+
pad=1
|
624 |
+
activation=mish
|
625 |
+
|
626 |
+
[route]
|
627 |
+
layers = -2
|
628 |
+
|
629 |
+
[convolutional]
|
630 |
+
batch_normalize=1
|
631 |
+
filters=256
|
632 |
+
size=1
|
633 |
+
stride=1
|
634 |
+
pad=1
|
635 |
+
activation=mish
|
636 |
+
|
637 |
+
# Residual Block
|
638 |
+
|
639 |
+
[convolutional]
|
640 |
+
batch_normalize=1
|
641 |
+
filters=256
|
642 |
+
size=1
|
643 |
+
stride=1
|
644 |
+
pad=1
|
645 |
+
activation=mish
|
646 |
+
|
647 |
+
[convolutional]
|
648 |
+
batch_normalize=1
|
649 |
+
filters=256
|
650 |
+
size=3
|
651 |
+
stride=1
|
652 |
+
pad=1
|
653 |
+
activation=mish
|
654 |
+
|
655 |
+
[shortcut]
|
656 |
+
from=-3
|
657 |
+
activation=linear
|
658 |
+
|
659 |
+
[convolutional]
|
660 |
+
batch_normalize=1
|
661 |
+
filters=256
|
662 |
+
size=1
|
663 |
+
stride=1
|
664 |
+
pad=1
|
665 |
+
activation=mish
|
666 |
+
|
667 |
+
[convolutional]
|
668 |
+
batch_normalize=1
|
669 |
+
filters=256
|
670 |
+
size=3
|
671 |
+
stride=1
|
672 |
+
pad=1
|
673 |
+
activation=mish
|
674 |
+
|
675 |
+
[shortcut]
|
676 |
+
from=-3
|
677 |
+
activation=linear
|
678 |
+
|
679 |
+
[convolutional]
|
680 |
+
batch_normalize=1
|
681 |
+
filters=256
|
682 |
+
size=1
|
683 |
+
stride=1
|
684 |
+
pad=1
|
685 |
+
activation=mish
|
686 |
+
|
687 |
+
[convolutional]
|
688 |
+
batch_normalize=1
|
689 |
+
filters=256
|
690 |
+
size=3
|
691 |
+
stride=1
|
692 |
+
pad=1
|
693 |
+
activation=mish
|
694 |
+
|
695 |
+
[shortcut]
|
696 |
+
from=-3
|
697 |
+
activation=linear
|
698 |
+
|
699 |
+
[convolutional]
|
700 |
+
batch_normalize=1
|
701 |
+
filters=256
|
702 |
+
size=1
|
703 |
+
stride=1
|
704 |
+
pad=1
|
705 |
+
activation=mish
|
706 |
+
|
707 |
+
[convolutional]
|
708 |
+
batch_normalize=1
|
709 |
+
filters=256
|
710 |
+
size=3
|
711 |
+
stride=1
|
712 |
+
pad=1
|
713 |
+
activation=mish
|
714 |
+
|
715 |
+
[shortcut]
|
716 |
+
from=-3
|
717 |
+
activation=linear
|
718 |
+
|
719 |
+
[convolutional]
|
720 |
+
batch_normalize=1
|
721 |
+
filters=256
|
722 |
+
size=1
|
723 |
+
stride=1
|
724 |
+
pad=1
|
725 |
+
activation=mish
|
726 |
+
|
727 |
+
[convolutional]
|
728 |
+
batch_normalize=1
|
729 |
+
filters=256
|
730 |
+
size=3
|
731 |
+
stride=1
|
732 |
+
pad=1
|
733 |
+
activation=mish
|
734 |
+
|
735 |
+
[shortcut]
|
736 |
+
from=-3
|
737 |
+
activation=linear
|
738 |
+
|
739 |
+
[convolutional]
|
740 |
+
batch_normalize=1
|
741 |
+
filters=256
|
742 |
+
size=1
|
743 |
+
stride=1
|
744 |
+
pad=1
|
745 |
+
activation=mish
|
746 |
+
|
747 |
+
[convolutional]
|
748 |
+
batch_normalize=1
|
749 |
+
filters=256
|
750 |
+
size=3
|
751 |
+
stride=1
|
752 |
+
pad=1
|
753 |
+
activation=mish
|
754 |
+
|
755 |
+
[shortcut]
|
756 |
+
from=-3
|
757 |
+
activation=linear
|
758 |
+
|
759 |
+
[convolutional]
|
760 |
+
batch_normalize=1
|
761 |
+
filters=256
|
762 |
+
size=1
|
763 |
+
stride=1
|
764 |
+
pad=1
|
765 |
+
activation=mish
|
766 |
+
|
767 |
+
[convolutional]
|
768 |
+
batch_normalize=1
|
769 |
+
filters=256
|
770 |
+
size=3
|
771 |
+
stride=1
|
772 |
+
pad=1
|
773 |
+
activation=mish
|
774 |
+
|
775 |
+
[shortcut]
|
776 |
+
from=-3
|
777 |
+
activation=linear
|
778 |
+
|
779 |
+
[convolutional]
|
780 |
+
batch_normalize=1
|
781 |
+
filters=256
|
782 |
+
size=1
|
783 |
+
stride=1
|
784 |
+
pad=1
|
785 |
+
activation=mish
|
786 |
+
|
787 |
+
[convolutional]
|
788 |
+
batch_normalize=1
|
789 |
+
filters=256
|
790 |
+
size=3
|
791 |
+
stride=1
|
792 |
+
pad=1
|
793 |
+
activation=mish
|
794 |
+
|
795 |
+
[shortcut]
|
796 |
+
from=-3
|
797 |
+
activation=linear
|
798 |
+
|
799 |
+
[convolutional]
|
800 |
+
batch_normalize=1
|
801 |
+
filters=256
|
802 |
+
size=1
|
803 |
+
stride=1
|
804 |
+
pad=1
|
805 |
+
activation=mish
|
806 |
+
|
807 |
+
[convolutional]
|
808 |
+
batch_normalize=1
|
809 |
+
filters=256
|
810 |
+
size=3
|
811 |
+
stride=1
|
812 |
+
pad=1
|
813 |
+
activation=mish
|
814 |
+
|
815 |
+
[shortcut]
|
816 |
+
from=-3
|
817 |
+
activation=linear
|
818 |
+
|
819 |
+
[convolutional]
|
820 |
+
batch_normalize=1
|
821 |
+
filters=256
|
822 |
+
size=1
|
823 |
+
stride=1
|
824 |
+
pad=1
|
825 |
+
activation=mish
|
826 |
+
|
827 |
+
[convolutional]
|
828 |
+
batch_normalize=1
|
829 |
+
filters=256
|
830 |
+
size=3
|
831 |
+
stride=1
|
832 |
+
pad=1
|
833 |
+
activation=mish
|
834 |
+
|
835 |
+
[shortcut]
|
836 |
+
from=-3
|
837 |
+
activation=linear
|
838 |
+
|
839 |
+
[convolutional]
|
840 |
+
batch_normalize=1
|
841 |
+
filters=256
|
842 |
+
size=1
|
843 |
+
stride=1
|
844 |
+
pad=1
|
845 |
+
activation=mish
|
846 |
+
|
847 |
+
[convolutional]
|
848 |
+
batch_normalize=1
|
849 |
+
filters=256
|
850 |
+
size=3
|
851 |
+
stride=1
|
852 |
+
pad=1
|
853 |
+
activation=mish
|
854 |
+
|
855 |
+
[shortcut]
|
856 |
+
from=-3
|
857 |
+
activation=linear
|
858 |
+
|
859 |
+
[convolutional]
|
860 |
+
batch_normalize=1
|
861 |
+
filters=256
|
862 |
+
size=1
|
863 |
+
stride=1
|
864 |
+
pad=1
|
865 |
+
activation=mish
|
866 |
+
|
867 |
+
[convolutional]
|
868 |
+
batch_normalize=1
|
869 |
+
filters=256
|
870 |
+
size=3
|
871 |
+
stride=1
|
872 |
+
pad=1
|
873 |
+
activation=mish
|
874 |
+
|
875 |
+
[shortcut]
|
876 |
+
from=-3
|
877 |
+
activation=linear
|
878 |
+
|
879 |
+
[convolutional]
|
880 |
+
batch_normalize=1
|
881 |
+
filters=256
|
882 |
+
size=1
|
883 |
+
stride=1
|
884 |
+
pad=1
|
885 |
+
activation=mish
|
886 |
+
|
887 |
+
[convolutional]
|
888 |
+
batch_normalize=1
|
889 |
+
filters=256
|
890 |
+
size=3
|
891 |
+
stride=1
|
892 |
+
pad=1
|
893 |
+
activation=mish
|
894 |
+
|
895 |
+
[shortcut]
|
896 |
+
from=-3
|
897 |
+
activation=linear
|
898 |
+
|
899 |
+
[convolutional]
|
900 |
+
batch_normalize=1
|
901 |
+
filters=256
|
902 |
+
size=1
|
903 |
+
stride=1
|
904 |
+
pad=1
|
905 |
+
activation=mish
|
906 |
+
|
907 |
+
[convolutional]
|
908 |
+
batch_normalize=1
|
909 |
+
filters=256
|
910 |
+
size=3
|
911 |
+
stride=1
|
912 |
+
pad=1
|
913 |
+
activation=mish
|
914 |
+
|
915 |
+
[shortcut]
|
916 |
+
from=-3
|
917 |
+
activation=linear
|
918 |
+
|
919 |
+
[convolutional]
|
920 |
+
batch_normalize=1
|
921 |
+
filters=256
|
922 |
+
size=1
|
923 |
+
stride=1
|
924 |
+
pad=1
|
925 |
+
activation=mish
|
926 |
+
|
927 |
+
[convolutional]
|
928 |
+
batch_normalize=1
|
929 |
+
filters=256
|
930 |
+
size=3
|
931 |
+
stride=1
|
932 |
+
pad=1
|
933 |
+
activation=mish
|
934 |
+
|
935 |
+
[shortcut]
|
936 |
+
from=-3
|
937 |
+
activation=linear
|
938 |
+
|
939 |
+
# Transition first
|
940 |
+
|
941 |
+
[convolutional]
|
942 |
+
batch_normalize=1
|
943 |
+
filters=256
|
944 |
+
size=1
|
945 |
+
stride=1
|
946 |
+
pad=1
|
947 |
+
activation=mish
|
948 |
+
|
949 |
+
# Merge [-1, -(3k+4)]
|
950 |
+
|
951 |
+
[route]
|
952 |
+
layers = -1,-49
|
953 |
+
|
954 |
+
# Transition last
|
955 |
+
|
956 |
+
# 130 (previous+7+3k)
|
957 |
+
[convolutional]
|
958 |
+
batch_normalize=1
|
959 |
+
filters=512
|
960 |
+
size=1
|
961 |
+
stride=1
|
962 |
+
pad=1
|
963 |
+
activation=mish
|
964 |
+
|
965 |
+
|
966 |
+
# P5
|
967 |
+
|
968 |
+
# Downsample
|
969 |
+
|
970 |
+
[convolutional]
|
971 |
+
batch_normalize=1
|
972 |
+
filters=1024
|
973 |
+
size=3
|
974 |
+
stride=2
|
975 |
+
pad=1
|
976 |
+
activation=mish
|
977 |
+
|
978 |
+
# Split
|
979 |
+
|
980 |
+
[convolutional]
|
981 |
+
batch_normalize=1
|
982 |
+
filters=512
|
983 |
+
size=1
|
984 |
+
stride=1
|
985 |
+
pad=1
|
986 |
+
activation=mish
|
987 |
+
|
988 |
+
[route]
|
989 |
+
layers = -2
|
990 |
+
|
991 |
+
[convolutional]
|
992 |
+
batch_normalize=1
|
993 |
+
filters=512
|
994 |
+
size=1
|
995 |
+
stride=1
|
996 |
+
pad=1
|
997 |
+
activation=mish
|
998 |
+
|
999 |
+
# Residual Block
|
1000 |
+
|
1001 |
+
[convolutional]
|
1002 |
+
batch_normalize=1
|
1003 |
+
filters=512
|
1004 |
+
size=1
|
1005 |
+
stride=1
|
1006 |
+
pad=1
|
1007 |
+
activation=mish
|
1008 |
+
|
1009 |
+
[convolutional]
|
1010 |
+
batch_normalize=1
|
1011 |
+
filters=512
|
1012 |
+
size=3
|
1013 |
+
stride=1
|
1014 |
+
pad=1
|
1015 |
+
activation=mish
|
1016 |
+
|
1017 |
+
[shortcut]
|
1018 |
+
from=-3
|
1019 |
+
activation=linear
|
1020 |
+
|
1021 |
+
[convolutional]
|
1022 |
+
batch_normalize=1
|
1023 |
+
filters=512
|
1024 |
+
size=1
|
1025 |
+
stride=1
|
1026 |
+
pad=1
|
1027 |
+
activation=mish
|
1028 |
+
|
1029 |
+
[convolutional]
|
1030 |
+
batch_normalize=1
|
1031 |
+
filters=512
|
1032 |
+
size=3
|
1033 |
+
stride=1
|
1034 |
+
pad=1
|
1035 |
+
activation=mish
|
1036 |
+
|
1037 |
+
[shortcut]
|
1038 |
+
from=-3
|
1039 |
+
activation=linear
|
1040 |
+
|
1041 |
+
[convolutional]
|
1042 |
+
batch_normalize=1
|
1043 |
+
filters=512
|
1044 |
+
size=1
|
1045 |
+
stride=1
|
1046 |
+
pad=1
|
1047 |
+
activation=mish
|
1048 |
+
|
1049 |
+
[convolutional]
|
1050 |
+
batch_normalize=1
|
1051 |
+
filters=512
|
1052 |
+
size=3
|
1053 |
+
stride=1
|
1054 |
+
pad=1
|
1055 |
+
activation=mish
|
1056 |
+
|
1057 |
+
[shortcut]
|
1058 |
+
from=-3
|
1059 |
+
activation=linear
|
1060 |
+
|
1061 |
+
[convolutional]
|
1062 |
+
batch_normalize=1
|
1063 |
+
filters=512
|
1064 |
+
size=1
|
1065 |
+
stride=1
|
1066 |
+
pad=1
|
1067 |
+
activation=mish
|
1068 |
+
|
1069 |
+
[convolutional]
|
1070 |
+
batch_normalize=1
|
1071 |
+
filters=512
|
1072 |
+
size=3
|
1073 |
+
stride=1
|
1074 |
+
pad=1
|
1075 |
+
activation=mish
|
1076 |
+
|
1077 |
+
[shortcut]
|
1078 |
+
from=-3
|
1079 |
+
activation=linear
|
1080 |
+
|
1081 |
+
[convolutional]
|
1082 |
+
batch_normalize=1
|
1083 |
+
filters=512
|
1084 |
+
size=1
|
1085 |
+
stride=1
|
1086 |
+
pad=1
|
1087 |
+
activation=mish
|
1088 |
+
|
1089 |
+
[convolutional]
|
1090 |
+
batch_normalize=1
|
1091 |
+
filters=512
|
1092 |
+
size=3
|
1093 |
+
stride=1
|
1094 |
+
pad=1
|
1095 |
+
activation=mish
|
1096 |
+
|
1097 |
+
[shortcut]
|
1098 |
+
from=-3
|
1099 |
+
activation=linear
|
1100 |
+
|
1101 |
+
[convolutional]
|
1102 |
+
batch_normalize=1
|
1103 |
+
filters=512
|
1104 |
+
size=1
|
1105 |
+
stride=1
|
1106 |
+
pad=1
|
1107 |
+
activation=mish
|
1108 |
+
|
1109 |
+
[convolutional]
|
1110 |
+
batch_normalize=1
|
1111 |
+
filters=512
|
1112 |
+
size=3
|
1113 |
+
stride=1
|
1114 |
+
pad=1
|
1115 |
+
activation=mish
|
1116 |
+
|
1117 |
+
[shortcut]
|
1118 |
+
from=-3
|
1119 |
+
activation=linear
|
1120 |
+
|
1121 |
+
[convolutional]
|
1122 |
+
batch_normalize=1
|
1123 |
+
filters=512
|
1124 |
+
size=1
|
1125 |
+
stride=1
|
1126 |
+
pad=1
|
1127 |
+
activation=mish
|
1128 |
+
|
1129 |
+
[convolutional]
|
1130 |
+
batch_normalize=1
|
1131 |
+
filters=512
|
1132 |
+
size=3
|
1133 |
+
stride=1
|
1134 |
+
pad=1
|
1135 |
+
activation=mish
|
1136 |
+
|
1137 |
+
[shortcut]
|
1138 |
+
from=-3
|
1139 |
+
activation=linear
|
1140 |
+
|
1141 |
+
# Transition first
|
1142 |
+
|
1143 |
+
[convolutional]
|
1144 |
+
batch_normalize=1
|
1145 |
+
filters=512
|
1146 |
+
size=1
|
1147 |
+
stride=1
|
1148 |
+
pad=1
|
1149 |
+
activation=mish
|
1150 |
+
|
1151 |
+
# Merge [-1, -(3k+4)]
|
1152 |
+
|
1153 |
+
[route]
|
1154 |
+
layers = -1,-25
|
1155 |
+
|
1156 |
+
# Transition last
|
1157 |
+
|
1158 |
+
# 158 (previous+7+3k)
|
1159 |
+
[convolutional]
|
1160 |
+
batch_normalize=1
|
1161 |
+
filters=1024
|
1162 |
+
size=1
|
1163 |
+
stride=1
|
1164 |
+
pad=1
|
1165 |
+
activation=mish
|
1166 |
+
|
1167 |
+
|
1168 |
+
# P6
|
1169 |
+
|
1170 |
+
# Downsample
|
1171 |
+
|
1172 |
+
[convolutional]
|
1173 |
+
batch_normalize=1
|
1174 |
+
filters=1024
|
1175 |
+
size=3
|
1176 |
+
stride=2
|
1177 |
+
pad=1
|
1178 |
+
activation=mish
|
1179 |
+
|
1180 |
+
# Split
|
1181 |
+
|
1182 |
+
[convolutional]
|
1183 |
+
batch_normalize=1
|
1184 |
+
filters=512
|
1185 |
+
size=1
|
1186 |
+
stride=1
|
1187 |
+
pad=1
|
1188 |
+
activation=mish
|
1189 |
+
|
1190 |
+
[route]
|
1191 |
+
layers = -2
|
1192 |
+
|
1193 |
+
[convolutional]
|
1194 |
+
batch_normalize=1
|
1195 |
+
filters=512
|
1196 |
+
size=1
|
1197 |
+
stride=1
|
1198 |
+
pad=1
|
1199 |
+
activation=mish
|
1200 |
+
|
1201 |
+
# Residual Block
|
1202 |
+
|
1203 |
+
[convolutional]
|
1204 |
+
batch_normalize=1
|
1205 |
+
filters=512
|
1206 |
+
size=1
|
1207 |
+
stride=1
|
1208 |
+
pad=1
|
1209 |
+
activation=mish
|
1210 |
+
|
1211 |
+
[convolutional]
|
1212 |
+
batch_normalize=1
|
1213 |
+
filters=512
|
1214 |
+
size=3
|
1215 |
+
stride=1
|
1216 |
+
pad=1
|
1217 |
+
activation=mish
|
1218 |
+
|
1219 |
+
[shortcut]
|
1220 |
+
from=-3
|
1221 |
+
activation=linear
|
1222 |
+
|
1223 |
+
[convolutional]
|
1224 |
+
batch_normalize=1
|
1225 |
+
filters=512
|
1226 |
+
size=1
|
1227 |
+
stride=1
|
1228 |
+
pad=1
|
1229 |
+
activation=mish
|
1230 |
+
|
1231 |
+
[convolutional]
|
1232 |
+
batch_normalize=1
|
1233 |
+
filters=512
|
1234 |
+
size=3
|
1235 |
+
stride=1
|
1236 |
+
pad=1
|
1237 |
+
activation=mish
|
1238 |
+
|
1239 |
+
[shortcut]
|
1240 |
+
from=-3
|
1241 |
+
activation=linear
|
1242 |
+
|
1243 |
+
[convolutional]
|
1244 |
+
batch_normalize=1
|
1245 |
+
filters=512
|
1246 |
+
size=1
|
1247 |
+
stride=1
|
1248 |
+
pad=1
|
1249 |
+
activation=mish
|
1250 |
+
|
1251 |
+
[convolutional]
|
1252 |
+
batch_normalize=1
|
1253 |
+
filters=512
|
1254 |
+
size=3
|
1255 |
+
stride=1
|
1256 |
+
pad=1
|
1257 |
+
activation=mish
|
1258 |
+
|
1259 |
+
[shortcut]
|
1260 |
+
from=-3
|
1261 |
+
activation=linear
|
1262 |
+
|
1263 |
+
[convolutional]
|
1264 |
+
batch_normalize=1
|
1265 |
+
filters=512
|
1266 |
+
size=1
|
1267 |
+
stride=1
|
1268 |
+
pad=1
|
1269 |
+
activation=mish
|
1270 |
+
|
1271 |
+
[convolutional]
|
1272 |
+
batch_normalize=1
|
1273 |
+
filters=512
|
1274 |
+
size=3
|
1275 |
+
stride=1
|
1276 |
+
pad=1
|
1277 |
+
activation=mish
|
1278 |
+
|
1279 |
+
[shortcut]
|
1280 |
+
from=-3
|
1281 |
+
activation=linear
|
1282 |
+
|
1283 |
+
[convolutional]
|
1284 |
+
batch_normalize=1
|
1285 |
+
filters=512
|
1286 |
+
size=1
|
1287 |
+
stride=1
|
1288 |
+
pad=1
|
1289 |
+
activation=mish
|
1290 |
+
|
1291 |
+
[convolutional]
|
1292 |
+
batch_normalize=1
|
1293 |
+
filters=512
|
1294 |
+
size=3
|
1295 |
+
stride=1
|
1296 |
+
pad=1
|
1297 |
+
activation=mish
|
1298 |
+
|
1299 |
+
[shortcut]
|
1300 |
+
from=-3
|
1301 |
+
activation=linear
|
1302 |
+
|
1303 |
+
[convolutional]
|
1304 |
+
batch_normalize=1
|
1305 |
+
filters=512
|
1306 |
+
size=1
|
1307 |
+
stride=1
|
1308 |
+
pad=1
|
1309 |
+
activation=mish
|
1310 |
+
|
1311 |
+
[convolutional]
|
1312 |
+
batch_normalize=1
|
1313 |
+
filters=512
|
1314 |
+
size=3
|
1315 |
+
stride=1
|
1316 |
+
pad=1
|
1317 |
+
activation=mish
|
1318 |
+
|
1319 |
+
[shortcut]
|
1320 |
+
from=-3
|
1321 |
+
activation=linear
|
1322 |
+
|
1323 |
+
[convolutional]
|
1324 |
+
batch_normalize=1
|
1325 |
+
filters=512
|
1326 |
+
size=1
|
1327 |
+
stride=1
|
1328 |
+
pad=1
|
1329 |
+
activation=mish
|
1330 |
+
|
1331 |
+
[convolutional]
|
1332 |
+
batch_normalize=1
|
1333 |
+
filters=512
|
1334 |
+
size=3
|
1335 |
+
stride=1
|
1336 |
+
pad=1
|
1337 |
+
activation=mish
|
1338 |
+
|
1339 |
+
[shortcut]
|
1340 |
+
from=-3
|
1341 |
+
activation=linear
|
1342 |
+
|
1343 |
+
# Transition first
|
1344 |
+
|
1345 |
+
[convolutional]
|
1346 |
+
batch_normalize=1
|
1347 |
+
filters=512
|
1348 |
+
size=1
|
1349 |
+
stride=1
|
1350 |
+
pad=1
|
1351 |
+
activation=mish
|
1352 |
+
|
1353 |
+
# Merge [-1, -(3k+4)]
|
1354 |
+
|
1355 |
+
[route]
|
1356 |
+
layers = -1,-25
|
1357 |
+
|
1358 |
+
# Transition last
|
1359 |
+
|
1360 |
+
# 186 (previous+7+3k)
|
1361 |
+
[convolutional]
|
1362 |
+
batch_normalize=1
|
1363 |
+
filters=1024
|
1364 |
+
size=1
|
1365 |
+
stride=1
|
1366 |
+
pad=1
|
1367 |
+
activation=mish
|
1368 |
+
|
1369 |
+
# ============ End of Backbone ============ #
|
1370 |
+
|
1371 |
+
# ============ Neck ============ #
|
1372 |
+
|
1373 |
+
# CSPSPP
|
1374 |
+
|
1375 |
+
[convolutional]
|
1376 |
+
batch_normalize=1
|
1377 |
+
filters=512
|
1378 |
+
size=1
|
1379 |
+
stride=1
|
1380 |
+
pad=1
|
1381 |
+
activation=mish
|
1382 |
+
|
1383 |
+
[route]
|
1384 |
+
layers = -2
|
1385 |
+
|
1386 |
+
[convolutional]
|
1387 |
+
batch_normalize=1
|
1388 |
+
filters=512
|
1389 |
+
size=1
|
1390 |
+
stride=1
|
1391 |
+
pad=1
|
1392 |
+
activation=mish
|
1393 |
+
|
1394 |
+
[convolutional]
|
1395 |
+
batch_normalize=1
|
1396 |
+
size=3
|
1397 |
+
stride=1
|
1398 |
+
pad=1
|
1399 |
+
filters=512
|
1400 |
+
activation=mish
|
1401 |
+
|
1402 |
+
[convolutional]
|
1403 |
+
batch_normalize=1
|
1404 |
+
filters=512
|
1405 |
+
size=1
|
1406 |
+
stride=1
|
1407 |
+
pad=1
|
1408 |
+
activation=mish
|
1409 |
+
|
1410 |
+
### SPP ###
|
1411 |
+
[maxpool]
|
1412 |
+
stride=1
|
1413 |
+
size=5
|
1414 |
+
|
1415 |
+
[route]
|
1416 |
+
layers=-2
|
1417 |
+
|
1418 |
+
[maxpool]
|
1419 |
+
stride=1
|
1420 |
+
size=9
|
1421 |
+
|
1422 |
+
[route]
|
1423 |
+
layers=-4
|
1424 |
+
|
1425 |
+
[maxpool]
|
1426 |
+
stride=1
|
1427 |
+
size=13
|
1428 |
+
|
1429 |
+
[route]
|
1430 |
+
layers=-1,-3,-5,-6
|
1431 |
+
### End SPP ###
|
1432 |
+
|
1433 |
+
[convolutional]
|
1434 |
+
batch_normalize=1
|
1435 |
+
filters=512
|
1436 |
+
size=1
|
1437 |
+
stride=1
|
1438 |
+
pad=1
|
1439 |
+
activation=mish
|
1440 |
+
|
1441 |
+
[convolutional]
|
1442 |
+
batch_normalize=1
|
1443 |
+
size=3
|
1444 |
+
stride=1
|
1445 |
+
pad=1
|
1446 |
+
filters=512
|
1447 |
+
activation=mish
|
1448 |
+
|
1449 |
+
[route]
|
1450 |
+
layers = -1, -13
|
1451 |
+
|
1452 |
+
# 201 (previous+6+5+2k)
|
1453 |
+
[convolutional]
|
1454 |
+
batch_normalize=1
|
1455 |
+
filters=512
|
1456 |
+
size=1
|
1457 |
+
stride=1
|
1458 |
+
pad=1
|
1459 |
+
activation=mish
|
1460 |
+
|
1461 |
+
# End of CSPSPP
|
1462 |
+
|
1463 |
+
|
1464 |
+
# FPN-5
|
1465 |
+
|
1466 |
+
[convolutional]
|
1467 |
+
batch_normalize=1
|
1468 |
+
filters=512
|
1469 |
+
size=1
|
1470 |
+
stride=1
|
1471 |
+
pad=1
|
1472 |
+
activation=mish
|
1473 |
+
|
1474 |
+
[upsample]
|
1475 |
+
stride=2
|
1476 |
+
|
1477 |
+
[route]
|
1478 |
+
layers = 158
|
1479 |
+
|
1480 |
+
[convolutional]
|
1481 |
+
batch_normalize=1
|
1482 |
+
filters=512
|
1483 |
+
size=1
|
1484 |
+
stride=1
|
1485 |
+
pad=1
|
1486 |
+
activation=mish
|
1487 |
+
|
1488 |
+
[route]
|
1489 |
+
layers = -1, -3
|
1490 |
+
|
1491 |
+
[convolutional]
|
1492 |
+
batch_normalize=1
|
1493 |
+
filters=512
|
1494 |
+
size=1
|
1495 |
+
stride=1
|
1496 |
+
pad=1
|
1497 |
+
activation=mish
|
1498 |
+
|
1499 |
+
# Split
|
1500 |
+
|
1501 |
+
[convolutional]
|
1502 |
+
batch_normalize=1
|
1503 |
+
filters=512
|
1504 |
+
size=1
|
1505 |
+
stride=1
|
1506 |
+
pad=1
|
1507 |
+
activation=mish
|
1508 |
+
|
1509 |
+
[route]
|
1510 |
+
layers = -2
|
1511 |
+
|
1512 |
+
# Plain Block
|
1513 |
+
|
1514 |
+
[convolutional]
|
1515 |
+
batch_normalize=1
|
1516 |
+
filters=512
|
1517 |
+
size=1
|
1518 |
+
stride=1
|
1519 |
+
pad=1
|
1520 |
+
activation=mish
|
1521 |
+
|
1522 |
+
[convolutional]
|
1523 |
+
batch_normalize=1
|
1524 |
+
size=3
|
1525 |
+
stride=1
|
1526 |
+
pad=1
|
1527 |
+
filters=512
|
1528 |
+
activation=mish
|
1529 |
+
|
1530 |
+
[convolutional]
|
1531 |
+
batch_normalize=1
|
1532 |
+
filters=512
|
1533 |
+
size=1
|
1534 |
+
stride=1
|
1535 |
+
pad=1
|
1536 |
+
activation=mish
|
1537 |
+
|
1538 |
+
[convolutional]
|
1539 |
+
batch_normalize=1
|
1540 |
+
size=3
|
1541 |
+
stride=1
|
1542 |
+
pad=1
|
1543 |
+
filters=512
|
1544 |
+
activation=mish
|
1545 |
+
|
1546 |
+
[convolutional]
|
1547 |
+
batch_normalize=1
|
1548 |
+
filters=512
|
1549 |
+
size=1
|
1550 |
+
stride=1
|
1551 |
+
pad=1
|
1552 |
+
activation=mish
|
1553 |
+
|
1554 |
+
[convolutional]
|
1555 |
+
batch_normalize=1
|
1556 |
+
size=3
|
1557 |
+
stride=1
|
1558 |
+
pad=1
|
1559 |
+
filters=512
|
1560 |
+
activation=mish
|
1561 |
+
|
1562 |
+
# Merge [-1, -(2k+2)]
|
1563 |
+
|
1564 |
+
[route]
|
1565 |
+
layers = -1, -8
|
1566 |
+
|
1567 |
+
# Transition last
|
1568 |
+
|
1569 |
+
# 217 (previous+6+4+2k)
|
1570 |
+
[convolutional]
|
1571 |
+
batch_normalize=1
|
1572 |
+
filters=512
|
1573 |
+
size=1
|
1574 |
+
stride=1
|
1575 |
+
pad=1
|
1576 |
+
activation=mish
|
1577 |
+
|
1578 |
+
|
1579 |
+
# FPN-4
|
1580 |
+
|
1581 |
+
[convolutional]
|
1582 |
+
batch_normalize=1
|
1583 |
+
filters=256
|
1584 |
+
size=1
|
1585 |
+
stride=1
|
1586 |
+
pad=1
|
1587 |
+
activation=mish
|
1588 |
+
|
1589 |
+
[upsample]
|
1590 |
+
stride=2
|
1591 |
+
|
1592 |
+
[route]
|
1593 |
+
layers = 130
|
1594 |
+
|
1595 |
+
[convolutional]
|
1596 |
+
batch_normalize=1
|
1597 |
+
filters=256
|
1598 |
+
size=1
|
1599 |
+
stride=1
|
1600 |
+
pad=1
|
1601 |
+
activation=mish
|
1602 |
+
|
1603 |
+
[route]
|
1604 |
+
layers = -1, -3
|
1605 |
+
|
1606 |
+
[convolutional]
|
1607 |
+
batch_normalize=1
|
1608 |
+
filters=256
|
1609 |
+
size=1
|
1610 |
+
stride=1
|
1611 |
+
pad=1
|
1612 |
+
activation=mish
|
1613 |
+
|
1614 |
+
# Split
|
1615 |
+
|
1616 |
+
[convolutional]
|
1617 |
+
batch_normalize=1
|
1618 |
+
filters=256
|
1619 |
+
size=1
|
1620 |
+
stride=1
|
1621 |
+
pad=1
|
1622 |
+
activation=mish
|
1623 |
+
|
1624 |
+
[route]
|
1625 |
+
layers = -2
|
1626 |
+
|
1627 |
+
# Plain Block
|
1628 |
+
|
1629 |
+
[convolutional]
|
1630 |
+
batch_normalize=1
|
1631 |
+
filters=256
|
1632 |
+
size=1
|
1633 |
+
stride=1
|
1634 |
+
pad=1
|
1635 |
+
activation=mish
|
1636 |
+
|
1637 |
+
[convolutional]
|
1638 |
+
batch_normalize=1
|
1639 |
+
size=3
|
1640 |
+
stride=1
|
1641 |
+
pad=1
|
1642 |
+
filters=256
|
1643 |
+
activation=mish
|
1644 |
+
|
1645 |
+
[convolutional]
|
1646 |
+
batch_normalize=1
|
1647 |
+
filters=256
|
1648 |
+
size=1
|
1649 |
+
stride=1
|
1650 |
+
pad=1
|
1651 |
+
activation=mish
|
1652 |
+
|
1653 |
+
[convolutional]
|
1654 |
+
batch_normalize=1
|
1655 |
+
size=3
|
1656 |
+
stride=1
|
1657 |
+
pad=1
|
1658 |
+
filters=256
|
1659 |
+
activation=mish
|
1660 |
+
|
1661 |
+
[convolutional]
|
1662 |
+
batch_normalize=1
|
1663 |
+
filters=256
|
1664 |
+
size=1
|
1665 |
+
stride=1
|
1666 |
+
pad=1
|
1667 |
+
activation=mish
|
1668 |
+
|
1669 |
+
[convolutional]
|
1670 |
+
batch_normalize=1
|
1671 |
+
size=3
|
1672 |
+
stride=1
|
1673 |
+
pad=1
|
1674 |
+
filters=256
|
1675 |
+
activation=mish
|
1676 |
+
|
1677 |
+
# Merge [-1, -(2k+2)]
|
1678 |
+
|
1679 |
+
[route]
|
1680 |
+
layers = -1, -8
|
1681 |
+
|
1682 |
+
# Transition last
|
1683 |
+
|
1684 |
+
# 233 (previous+6+4+2k)
|
1685 |
+
[convolutional]
|
1686 |
+
batch_normalize=1
|
1687 |
+
filters=256
|
1688 |
+
size=1
|
1689 |
+
stride=1
|
1690 |
+
pad=1
|
1691 |
+
activation=mish
|
1692 |
+
|
1693 |
+
|
1694 |
+
# FPN-3
|
1695 |
+
|
1696 |
+
[convolutional]
|
1697 |
+
batch_normalize=1
|
1698 |
+
filters=128
|
1699 |
+
size=1
|
1700 |
+
stride=1
|
1701 |
+
pad=1
|
1702 |
+
activation=mish
|
1703 |
+
|
1704 |
+
[upsample]
|
1705 |
+
stride=2
|
1706 |
+
|
1707 |
+
[route]
|
1708 |
+
layers = 78
|
1709 |
+
|
1710 |
+
[convolutional]
|
1711 |
+
batch_normalize=1
|
1712 |
+
filters=128
|
1713 |
+
size=1
|
1714 |
+
stride=1
|
1715 |
+
pad=1
|
1716 |
+
activation=mish
|
1717 |
+
|
1718 |
+
[route]
|
1719 |
+
layers = -1, -3
|
1720 |
+
|
1721 |
+
[convolutional]
|
1722 |
+
batch_normalize=1
|
1723 |
+
filters=128
|
1724 |
+
size=1
|
1725 |
+
stride=1
|
1726 |
+
pad=1
|
1727 |
+
activation=mish
|
1728 |
+
|
1729 |
+
# Split
|
1730 |
+
|
1731 |
+
[convolutional]
|
1732 |
+
batch_normalize=1
|
1733 |
+
filters=128
|
1734 |
+
size=1
|
1735 |
+
stride=1
|
1736 |
+
pad=1
|
1737 |
+
activation=mish
|
1738 |
+
|
1739 |
+
[route]
|
1740 |
+
layers = -2
|
1741 |
+
|
1742 |
+
# Plain Block
|
1743 |
+
|
1744 |
+
[convolutional]
|
1745 |
+
batch_normalize=1
|
1746 |
+
filters=128
|
1747 |
+
size=1
|
1748 |
+
stride=1
|
1749 |
+
pad=1
|
1750 |
+
activation=mish
|
1751 |
+
|
1752 |
+
[convolutional]
|
1753 |
+
batch_normalize=1
|
1754 |
+
size=3
|
1755 |
+
stride=1
|
1756 |
+
pad=1
|
1757 |
+
filters=128
|
1758 |
+
activation=mish
|
1759 |
+
|
1760 |
+
[convolutional]
|
1761 |
+
batch_normalize=1
|
1762 |
+
filters=128
|
1763 |
+
size=1
|
1764 |
+
stride=1
|
1765 |
+
pad=1
|
1766 |
+
activation=mish
|
1767 |
+
|
1768 |
+
[convolutional]
|
1769 |
+
batch_normalize=1
|
1770 |
+
size=3
|
1771 |
+
stride=1
|
1772 |
+
pad=1
|
1773 |
+
filters=128
|
1774 |
+
activation=mish
|
1775 |
+
|
1776 |
+
[convolutional]
|
1777 |
+
batch_normalize=1
|
1778 |
+
filters=128
|
1779 |
+
size=1
|
1780 |
+
stride=1
|
1781 |
+
pad=1
|
1782 |
+
activation=mish
|
1783 |
+
|
1784 |
+
[convolutional]
|
1785 |
+
batch_normalize=1
|
1786 |
+
size=3
|
1787 |
+
stride=1
|
1788 |
+
pad=1
|
1789 |
+
filters=128
|
1790 |
+
activation=mish
|
1791 |
+
|
1792 |
+
# Merge [-1, -(2k+2)]
|
1793 |
+
|
1794 |
+
[route]
|
1795 |
+
layers = -1, -8
|
1796 |
+
|
1797 |
+
# Transition last
|
1798 |
+
|
1799 |
+
# 249 (previous+6+4+2k)
|
1800 |
+
[convolutional]
|
1801 |
+
batch_normalize=1
|
1802 |
+
filters=128
|
1803 |
+
size=1
|
1804 |
+
stride=1
|
1805 |
+
pad=1
|
1806 |
+
activation=mish
|
1807 |
+
|
1808 |
+
|
1809 |
+
# PAN-4
|
1810 |
+
|
1811 |
+
[convolutional]
|
1812 |
+
batch_normalize=1
|
1813 |
+
size=3
|
1814 |
+
stride=2
|
1815 |
+
pad=1
|
1816 |
+
filters=256
|
1817 |
+
activation=mish
|
1818 |
+
|
1819 |
+
[route]
|
1820 |
+
layers = -1, 233
|
1821 |
+
|
1822 |
+
[convolutional]
|
1823 |
+
batch_normalize=1
|
1824 |
+
filters=256
|
1825 |
+
size=1
|
1826 |
+
stride=1
|
1827 |
+
pad=1
|
1828 |
+
activation=mish
|
1829 |
+
|
1830 |
+
# Split
|
1831 |
+
|
1832 |
+
[convolutional]
|
1833 |
+
batch_normalize=1
|
1834 |
+
filters=256
|
1835 |
+
size=1
|
1836 |
+
stride=1
|
1837 |
+
pad=1
|
1838 |
+
activation=mish
|
1839 |
+
|
1840 |
+
[route]
|
1841 |
+
layers = -2
|
1842 |
+
|
1843 |
+
# Plain Block
|
1844 |
+
|
1845 |
+
[convolutional]
|
1846 |
+
batch_normalize=1
|
1847 |
+
filters=256
|
1848 |
+
size=1
|
1849 |
+
stride=1
|
1850 |
+
pad=1
|
1851 |
+
activation=mish
|
1852 |
+
|
1853 |
+
[convolutional]
|
1854 |
+
batch_normalize=1
|
1855 |
+
size=3
|
1856 |
+
stride=1
|
1857 |
+
pad=1
|
1858 |
+
filters=256
|
1859 |
+
activation=mish
|
1860 |
+
|
1861 |
+
[convolutional]
|
1862 |
+
batch_normalize=1
|
1863 |
+
filters=256
|
1864 |
+
size=1
|
1865 |
+
stride=1
|
1866 |
+
pad=1
|
1867 |
+
activation=mish
|
1868 |
+
|
1869 |
+
[convolutional]
|
1870 |
+
batch_normalize=1
|
1871 |
+
size=3
|
1872 |
+
stride=1
|
1873 |
+
pad=1
|
1874 |
+
filters=256
|
1875 |
+
activation=mish
|
1876 |
+
|
1877 |
+
[convolutional]
|
1878 |
+
batch_normalize=1
|
1879 |
+
filters=256
|
1880 |
+
size=1
|
1881 |
+
stride=1
|
1882 |
+
pad=1
|
1883 |
+
activation=mish
|
1884 |
+
|
1885 |
+
[convolutional]
|
1886 |
+
batch_normalize=1
|
1887 |
+
size=3
|
1888 |
+
stride=1
|
1889 |
+
pad=1
|
1890 |
+
filters=256
|
1891 |
+
activation=mish
|
1892 |
+
|
1893 |
+
[route]
|
1894 |
+
layers = -1,-8
|
1895 |
+
|
1896 |
+
# Transition last
|
1897 |
+
|
1898 |
+
# 262 (previous+3+4+2k)
|
1899 |
+
[convolutional]
|
1900 |
+
batch_normalize=1
|
1901 |
+
filters=256
|
1902 |
+
size=1
|
1903 |
+
stride=1
|
1904 |
+
pad=1
|
1905 |
+
activation=mish
|
1906 |
+
|
1907 |
+
|
1908 |
+
# PAN-5
|
1909 |
+
|
1910 |
+
[convolutional]
|
1911 |
+
batch_normalize=1
|
1912 |
+
size=3
|
1913 |
+
stride=2
|
1914 |
+
pad=1
|
1915 |
+
filters=512
|
1916 |
+
activation=mish
|
1917 |
+
|
1918 |
+
[route]
|
1919 |
+
layers = -1, 217
|
1920 |
+
|
1921 |
+
[convolutional]
|
1922 |
+
batch_normalize=1
|
1923 |
+
filters=512
|
1924 |
+
size=1
|
1925 |
+
stride=1
|
1926 |
+
pad=1
|
1927 |
+
activation=mish
|
1928 |
+
|
1929 |
+
# Split
|
1930 |
+
|
1931 |
+
[convolutional]
|
1932 |
+
batch_normalize=1
|
1933 |
+
filters=512
|
1934 |
+
size=1
|
1935 |
+
stride=1
|
1936 |
+
pad=1
|
1937 |
+
activation=mish
|
1938 |
+
|
1939 |
+
[route]
|
1940 |
+
layers = -2
|
1941 |
+
|
1942 |
+
# Plain Block
|
1943 |
+
|
1944 |
+
[convolutional]
|
1945 |
+
batch_normalize=1
|
1946 |
+
filters=512
|
1947 |
+
size=1
|
1948 |
+
stride=1
|
1949 |
+
pad=1
|
1950 |
+
activation=mish
|
1951 |
+
|
1952 |
+
[convolutional]
|
1953 |
+
batch_normalize=1
|
1954 |
+
size=3
|
1955 |
+
stride=1
|
1956 |
+
pad=1
|
1957 |
+
filters=512
|
1958 |
+
activation=mish
|
1959 |
+
|
1960 |
+
[convolutional]
|
1961 |
+
batch_normalize=1
|
1962 |
+
filters=512
|
1963 |
+
size=1
|
1964 |
+
stride=1
|
1965 |
+
pad=1
|
1966 |
+
activation=mish
|
1967 |
+
|
1968 |
+
[convolutional]
|
1969 |
+
batch_normalize=1
|
1970 |
+
size=3
|
1971 |
+
stride=1
|
1972 |
+
pad=1
|
1973 |
+
filters=512
|
1974 |
+
activation=mish
|
1975 |
+
|
1976 |
+
[convolutional]
|
1977 |
+
batch_normalize=1
|
1978 |
+
filters=512
|
1979 |
+
size=1
|
1980 |
+
stride=1
|
1981 |
+
pad=1
|
1982 |
+
activation=mish
|
1983 |
+
|
1984 |
+
[convolutional]
|
1985 |
+
batch_normalize=1
|
1986 |
+
size=3
|
1987 |
+
stride=1
|
1988 |
+
pad=1
|
1989 |
+
filters=512
|
1990 |
+
activation=mish
|
1991 |
+
|
1992 |
+
[route]
|
1993 |
+
layers = -1,-8
|
1994 |
+
|
1995 |
+
# Transition last
|
1996 |
+
|
1997 |
+
# 275 (previous+3+4+2k)
|
1998 |
+
[convolutional]
|
1999 |
+
batch_normalize=1
|
2000 |
+
filters=512
|
2001 |
+
size=1
|
2002 |
+
stride=1
|
2003 |
+
pad=1
|
2004 |
+
activation=mish
|
2005 |
+
|
2006 |
+
|
2007 |
+
# PAN-6
|
2008 |
+
|
2009 |
+
[convolutional]
|
2010 |
+
batch_normalize=1
|
2011 |
+
size=3
|
2012 |
+
stride=2
|
2013 |
+
pad=1
|
2014 |
+
filters=512
|
2015 |
+
activation=mish
|
2016 |
+
|
2017 |
+
[route]
|
2018 |
+
layers = -1, 201
|
2019 |
+
|
2020 |
+
[convolutional]
|
2021 |
+
batch_normalize=1
|
2022 |
+
filters=512
|
2023 |
+
size=1
|
2024 |
+
stride=1
|
2025 |
+
pad=1
|
2026 |
+
activation=mish
|
2027 |
+
|
2028 |
+
# Split
|
2029 |
+
|
2030 |
+
[convolutional]
|
2031 |
+
batch_normalize=1
|
2032 |
+
filters=512
|
2033 |
+
size=1
|
2034 |
+
stride=1
|
2035 |
+
pad=1
|
2036 |
+
activation=mish
|
2037 |
+
|
2038 |
+
[route]
|
2039 |
+
layers = -2
|
2040 |
+
|
2041 |
+
# Plain Block
|
2042 |
+
|
2043 |
+
[convolutional]
|
2044 |
+
batch_normalize=1
|
2045 |
+
filters=512
|
2046 |
+
size=1
|
2047 |
+
stride=1
|
2048 |
+
pad=1
|
2049 |
+
activation=mish
|
2050 |
+
|
2051 |
+
[convolutional]
|
2052 |
+
batch_normalize=1
|
2053 |
+
size=3
|
2054 |
+
stride=1
|
2055 |
+
pad=1
|
2056 |
+
filters=512
|
2057 |
+
activation=mish
|
2058 |
+
|
2059 |
+
[convolutional]
|
2060 |
+
batch_normalize=1
|
2061 |
+
filters=512
|
2062 |
+
size=1
|
2063 |
+
stride=1
|
2064 |
+
pad=1
|
2065 |
+
activation=mish
|
2066 |
+
|
2067 |
+
[convolutional]
|
2068 |
+
batch_normalize=1
|
2069 |
+
size=3
|
2070 |
+
stride=1
|
2071 |
+
pad=1
|
2072 |
+
filters=512
|
2073 |
+
activation=mish
|
2074 |
+
|
2075 |
+
[convolutional]
|
2076 |
+
batch_normalize=1
|
2077 |
+
filters=512
|
2078 |
+
size=1
|
2079 |
+
stride=1
|
2080 |
+
pad=1
|
2081 |
+
activation=mish
|
2082 |
+
|
2083 |
+
[convolutional]
|
2084 |
+
batch_normalize=1
|
2085 |
+
size=3
|
2086 |
+
stride=1
|
2087 |
+
pad=1
|
2088 |
+
filters=512
|
2089 |
+
activation=mish
|
2090 |
+
|
2091 |
+
[route]
|
2092 |
+
layers = -1,-8
|
2093 |
+
|
2094 |
+
# Transition last
|
2095 |
+
|
2096 |
+
# 288 (previous+3+4+2k)
|
2097 |
+
[convolutional]
|
2098 |
+
batch_normalize=1
|
2099 |
+
filters=512
|
2100 |
+
size=1
|
2101 |
+
stride=1
|
2102 |
+
pad=1
|
2103 |
+
activation=mish
|
2104 |
+
|
2105 |
+
# ============ End of Neck ============ #
|
2106 |
+
|
2107 |
+
# ============ Head ============ #
|
2108 |
+
|
2109 |
+
# YOLO-3
|
2110 |
+
|
2111 |
+
[route]
|
2112 |
+
layers = 249
|
2113 |
+
|
2114 |
+
[convolutional]
|
2115 |
+
batch_normalize=1
|
2116 |
+
size=3
|
2117 |
+
stride=1
|
2118 |
+
pad=1
|
2119 |
+
filters=256
|
2120 |
+
activation=mish
|
2121 |
+
|
2122 |
+
[convolutional]
|
2123 |
+
size=1
|
2124 |
+
stride=1
|
2125 |
+
pad=1
|
2126 |
+
filters=340
|
2127 |
+
activation=linear
|
2128 |
+
|
2129 |
+
[yolo]
|
2130 |
+
mask = 0,1,2,3
|
2131 |
+
anchors = 13,17, 31,25, 24,51, 61,45, 61,45, 48,102, 119,96, 97,189, 97,189, 217,184, 171,384, 324,451, 324,451, 545,357, 616,618, 1024,1024
|
2132 |
+
classes=80
|
2133 |
+
num=16
|
2134 |
+
jitter=.3
|
2135 |
+
ignore_thresh = .7
|
2136 |
+
truth_thresh = 1
|
2137 |
+
random=1
|
2138 |
+
scale_x_y = 1.05
|
2139 |
+
iou_thresh=0.213
|
2140 |
+
cls_normalizer=1.0
|
2141 |
+
iou_normalizer=0.07
|
2142 |
+
iou_loss=ciou
|
2143 |
+
nms_kind=greedynms
|
2144 |
+
beta_nms=0.6
|
2145 |
+
|
2146 |
+
|
2147 |
+
# YOLO-4
|
2148 |
+
|
2149 |
+
[route]
|
2150 |
+
layers = 262
|
2151 |
+
|
2152 |
+
[convolutional]
|
2153 |
+
batch_normalize=1
|
2154 |
+
size=3
|
2155 |
+
stride=1
|
2156 |
+
pad=1
|
2157 |
+
filters=512
|
2158 |
+
activation=mish
|
2159 |
+
|
2160 |
+
[convolutional]
|
2161 |
+
size=1
|
2162 |
+
stride=1
|
2163 |
+
pad=1
|
2164 |
+
filters=340
|
2165 |
+
activation=linear
|
2166 |
+
|
2167 |
+
[yolo]
|
2168 |
+
mask = 4,5,6,7
|
2169 |
+
anchors = 13,17, 31,25, 24,51, 61,45, 61,45, 48,102, 119,96, 97,189, 97,189, 217,184, 171,384, 324,451, 324,451, 545,357, 616,618, 1024,1024
|
2170 |
+
classes=80
|
2171 |
+
num=16
|
2172 |
+
jitter=.3
|
2173 |
+
ignore_thresh = .7
|
2174 |
+
truth_thresh = 1
|
2175 |
+
random=1
|
2176 |
+
scale_x_y = 1.05
|
2177 |
+
iou_thresh=0.213
|
2178 |
+
cls_normalizer=1.0
|
2179 |
+
iou_normalizer=0.07
|
2180 |
+
iou_loss=ciou
|
2181 |
+
nms_kind=greedynms
|
2182 |
+
beta_nms=0.6
|
2183 |
+
|
2184 |
+
|
2185 |
+
# YOLO-5
|
2186 |
+
|
2187 |
+
[route]
|
2188 |
+
layers = 275
|
2189 |
+
|
2190 |
+
[convolutional]
|
2191 |
+
batch_normalize=1
|
2192 |
+
size=3
|
2193 |
+
stride=1
|
2194 |
+
pad=1
|
2195 |
+
filters=1024
|
2196 |
+
activation=mish
|
2197 |
+
|
2198 |
+
[convolutional]
|
2199 |
+
size=1
|
2200 |
+
stride=1
|
2201 |
+
pad=1
|
2202 |
+
filters=340
|
2203 |
+
activation=linear
|
2204 |
+
|
2205 |
+
[yolo]
|
2206 |
+
mask = 8,9,10,11
|
2207 |
+
anchors = 13,17, 31,25, 24,51, 61,45, 61,45, 48,102, 119,96, 97,189, 97,189, 217,184, 171,384, 324,451, 324,451, 545,357, 616,618, 1024,1024
|
2208 |
+
classes=80
|
2209 |
+
num=16
|
2210 |
+
jitter=.3
|
2211 |
+
ignore_thresh = .7
|
2212 |
+
truth_thresh = 1
|
2213 |
+
random=1
|
2214 |
+
scale_x_y = 1.05
|
2215 |
+
iou_thresh=0.213
|
2216 |
+
cls_normalizer=1.0
|
2217 |
+
iou_normalizer=0.07
|
2218 |
+
iou_loss=ciou
|
2219 |
+
nms_kind=greedynms
|
2220 |
+
beta_nms=0.6
|
2221 |
+
|
2222 |
+
|
2223 |
+
# YOLO-6
|
2224 |
+
|
2225 |
+
[route]
|
2226 |
+
layers = 288
|
2227 |
+
|
2228 |
+
[convolutional]
|
2229 |
+
batch_normalize=1
|
2230 |
+
size=3
|
2231 |
+
stride=1
|
2232 |
+
pad=1
|
2233 |
+
filters=1024
|
2234 |
+
activation=mish
|
2235 |
+
|
2236 |
+
[convolutional]
|
2237 |
+
size=1
|
2238 |
+
stride=1
|
2239 |
+
pad=1
|
2240 |
+
filters=340
|
2241 |
+
activation=linear
|
2242 |
+
|
2243 |
+
[yolo]
|
2244 |
+
mask = 12,13,14,15
|
2245 |
+
anchors = 13,17, 31,25, 24,51, 61,45, 61,45, 48,102, 119,96, 97,189, 97,189, 217,184, 171,384, 324,451, 324,451, 545,357, 616,618, 1024,1024
|
2246 |
+
classes=80
|
2247 |
+
num=16
|
2248 |
+
jitter=.3
|
2249 |
+
ignore_thresh = .7
|
2250 |
+
truth_thresh = 1
|
2251 |
+
random=1
|
2252 |
+
scale_x_y = 1.05
|
2253 |
+
iou_thresh=0.213
|
2254 |
+
cls_normalizer=1.0
|
2255 |
+
iou_normalizer=0.07
|
2256 |
+
iou_loss=ciou
|
2257 |
+
nms_kind=greedynms
|
2258 |
+
beta_nms=0.6
|
2259 |
+
|
2260 |
+
# ============ End of Head ============ #
|
cfg/yolov4_p7.cfg
ADDED
@@ -0,0 +1,2714 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[net]
|
2 |
+
batch=64
|
3 |
+
subdivisions=8
|
4 |
+
width=1536
|
5 |
+
height=1536
|
6 |
+
channels=3
|
7 |
+
momentum=0.949
|
8 |
+
decay=0.0005
|
9 |
+
angle=0
|
10 |
+
saturation = 1.5
|
11 |
+
exposure = 1.5
|
12 |
+
hue=.1
|
13 |
+
|
14 |
+
learning_rate=0.00261
|
15 |
+
burn_in=1000
|
16 |
+
max_batches = 500500
|
17 |
+
policy=steps
|
18 |
+
steps=400000,450000
|
19 |
+
scales=.1,.1
|
20 |
+
|
21 |
+
mosaic=1
|
22 |
+
|
23 |
+
|
24 |
+
# ============ Backbone ============ #
|
25 |
+
|
26 |
+
# Stem
|
27 |
+
|
28 |
+
# 0
|
29 |
+
[convolutional]
|
30 |
+
batch_normalize=1
|
31 |
+
filters=40
|
32 |
+
size=3
|
33 |
+
stride=1
|
34 |
+
pad=1
|
35 |
+
activation=mish
|
36 |
+
|
37 |
+
|
38 |
+
# P1
|
39 |
+
|
40 |
+
# Downsample
|
41 |
+
|
42 |
+
[convolutional]
|
43 |
+
batch_normalize=1
|
44 |
+
filters=80
|
45 |
+
size=3
|
46 |
+
stride=2
|
47 |
+
pad=1
|
48 |
+
activation=mish
|
49 |
+
|
50 |
+
# Split
|
51 |
+
|
52 |
+
[convolutional]
|
53 |
+
batch_normalize=1
|
54 |
+
filters=40
|
55 |
+
size=1
|
56 |
+
stride=1
|
57 |
+
pad=1
|
58 |
+
activation=mish
|
59 |
+
|
60 |
+
[route]
|
61 |
+
layers = -2
|
62 |
+
|
63 |
+
[convolutional]
|
64 |
+
batch_normalize=1
|
65 |
+
filters=40
|
66 |
+
size=1
|
67 |
+
stride=1
|
68 |
+
pad=1
|
69 |
+
activation=mish
|
70 |
+
|
71 |
+
# Residual Block
|
72 |
+
|
73 |
+
[convolutional]
|
74 |
+
batch_normalize=1
|
75 |
+
filters=40
|
76 |
+
size=1
|
77 |
+
stride=1
|
78 |
+
pad=1
|
79 |
+
activation=mish
|
80 |
+
|
81 |
+
[convolutional]
|
82 |
+
batch_normalize=1
|
83 |
+
filters=40
|
84 |
+
size=3
|
85 |
+
stride=1
|
86 |
+
pad=1
|
87 |
+
activation=mish
|
88 |
+
|
89 |
+
[shortcut]
|
90 |
+
from=-3
|
91 |
+
activation=linear
|
92 |
+
|
93 |
+
# Transition first
|
94 |
+
|
95 |
+
[convolutional]
|
96 |
+
batch_normalize=1
|
97 |
+
filters=40
|
98 |
+
size=1
|
99 |
+
stride=1
|
100 |
+
pad=1
|
101 |
+
activation=mish
|
102 |
+
|
103 |
+
# Merge [-1, -(3k+4)]
|
104 |
+
|
105 |
+
[route]
|
106 |
+
layers = -1,-7
|
107 |
+
|
108 |
+
# Transition last
|
109 |
+
|
110 |
+
# 10 (previous+7+3k)
|
111 |
+
[convolutional]
|
112 |
+
batch_normalize=1
|
113 |
+
filters=80
|
114 |
+
size=1
|
115 |
+
stride=1
|
116 |
+
pad=1
|
117 |
+
activation=mish
|
118 |
+
|
119 |
+
|
120 |
+
# P2
|
121 |
+
|
122 |
+
# Downsample
|
123 |
+
|
124 |
+
[convolutional]
|
125 |
+
batch_normalize=1
|
126 |
+
filters=160
|
127 |
+
size=3
|
128 |
+
stride=2
|
129 |
+
pad=1
|
130 |
+
activation=mish
|
131 |
+
|
132 |
+
# Split
|
133 |
+
|
134 |
+
[convolutional]
|
135 |
+
batch_normalize=1
|
136 |
+
filters=80
|
137 |
+
size=1
|
138 |
+
stride=1
|
139 |
+
pad=1
|
140 |
+
activation=mish
|
141 |
+
|
142 |
+
[route]
|
143 |
+
layers = -2
|
144 |
+
|
145 |
+
[convolutional]
|
146 |
+
batch_normalize=1
|
147 |
+
filters=80
|
148 |
+
size=1
|
149 |
+
stride=1
|
150 |
+
pad=1
|
151 |
+
activation=mish
|
152 |
+
|
153 |
+
# Residual Block
|
154 |
+
|
155 |
+
[convolutional]
|
156 |
+
batch_normalize=1
|
157 |
+
filters=80
|
158 |
+
size=1
|
159 |
+
stride=1
|
160 |
+
pad=1
|
161 |
+
activation=mish
|
162 |
+
|
163 |
+
[convolutional]
|
164 |
+
batch_normalize=1
|
165 |
+
filters=80
|
166 |
+
size=3
|
167 |
+
stride=1
|
168 |
+
pad=1
|
169 |
+
activation=mish
|
170 |
+
|
171 |
+
[shortcut]
|
172 |
+
from=-3
|
173 |
+
activation=linear
|
174 |
+
|
175 |
+
[convolutional]
|
176 |
+
batch_normalize=1
|
177 |
+
filters=80
|
178 |
+
size=1
|
179 |
+
stride=1
|
180 |
+
pad=1
|
181 |
+
activation=mish
|
182 |
+
|
183 |
+
[convolutional]
|
184 |
+
batch_normalize=1
|
185 |
+
filters=80
|
186 |
+
size=3
|
187 |
+
stride=1
|
188 |
+
pad=1
|
189 |
+
activation=mish
|
190 |
+
|
191 |
+
[shortcut]
|
192 |
+
from=-3
|
193 |
+
activation=linear
|
194 |
+
|
195 |
+
[convolutional]
|
196 |
+
batch_normalize=1
|
197 |
+
filters=80
|
198 |
+
size=1
|
199 |
+
stride=1
|
200 |
+
pad=1
|
201 |
+
activation=mish
|
202 |
+
|
203 |
+
[convolutional]
|
204 |
+
batch_normalize=1
|
205 |
+
filters=80
|
206 |
+
size=3
|
207 |
+
stride=1
|
208 |
+
pad=1
|
209 |
+
activation=mish
|
210 |
+
|
211 |
+
[shortcut]
|
212 |
+
from=-3
|
213 |
+
activation=linear
|
214 |
+
|
215 |
+
# Transition first
|
216 |
+
|
217 |
+
[convolutional]
|
218 |
+
batch_normalize=1
|
219 |
+
filters=80
|
220 |
+
size=1
|
221 |
+
stride=1
|
222 |
+
pad=1
|
223 |
+
activation=mish
|
224 |
+
|
225 |
+
# Merge [-1, -(3k+4)]
|
226 |
+
|
227 |
+
[route]
|
228 |
+
layers = -1,-13
|
229 |
+
|
230 |
+
# Transition last
|
231 |
+
|
232 |
+
# 26 (previous+7+3k)
|
233 |
+
[convolutional]
|
234 |
+
batch_normalize=1
|
235 |
+
filters=160
|
236 |
+
size=1
|
237 |
+
stride=1
|
238 |
+
pad=1
|
239 |
+
activation=mish
|
240 |
+
|
241 |
+
|
242 |
+
# P3
|
243 |
+
|
244 |
+
# Downsample
|
245 |
+
|
246 |
+
[convolutional]
|
247 |
+
batch_normalize=1
|
248 |
+
filters=320
|
249 |
+
size=3
|
250 |
+
stride=2
|
251 |
+
pad=1
|
252 |
+
activation=mish
|
253 |
+
|
254 |
+
# Split
|
255 |
+
|
256 |
+
[convolutional]
|
257 |
+
batch_normalize=1
|
258 |
+
filters=160
|
259 |
+
size=1
|
260 |
+
stride=1
|
261 |
+
pad=1
|
262 |
+
activation=mish
|
263 |
+
|
264 |
+
[route]
|
265 |
+
layers = -2
|
266 |
+
|
267 |
+
[convolutional]
|
268 |
+
batch_normalize=1
|
269 |
+
filters=160
|
270 |
+
size=1
|
271 |
+
stride=1
|
272 |
+
pad=1
|
273 |
+
activation=mish
|
274 |
+
|
275 |
+
# Residual Block
|
276 |
+
|
277 |
+
[convolutional]
|
278 |
+
batch_normalize=1
|
279 |
+
filters=160
|
280 |
+
size=1
|
281 |
+
stride=1
|
282 |
+
pad=1
|
283 |
+
activation=mish
|
284 |
+
|
285 |
+
[convolutional]
|
286 |
+
batch_normalize=1
|
287 |
+
filters=160
|
288 |
+
size=3
|
289 |
+
stride=1
|
290 |
+
pad=1
|
291 |
+
activation=mish
|
292 |
+
|
293 |
+
[shortcut]
|
294 |
+
from=-3
|
295 |
+
activation=linear
|
296 |
+
|
297 |
+
[convolutional]
|
298 |
+
batch_normalize=1
|
299 |
+
filters=160
|
300 |
+
size=1
|
301 |
+
stride=1
|
302 |
+
pad=1
|
303 |
+
activation=mish
|
304 |
+
|
305 |
+
[convolutional]
|
306 |
+
batch_normalize=1
|
307 |
+
filters=160
|
308 |
+
size=3
|
309 |
+
stride=1
|
310 |
+
pad=1
|
311 |
+
activation=mish
|
312 |
+
|
313 |
+
[shortcut]
|
314 |
+
from=-3
|
315 |
+
activation=linear
|
316 |
+
|
317 |
+
[convolutional]
|
318 |
+
batch_normalize=1
|
319 |
+
filters=160
|
320 |
+
size=1
|
321 |
+
stride=1
|
322 |
+
pad=1
|
323 |
+
activation=mish
|
324 |
+
|
325 |
+
[convolutional]
|
326 |
+
batch_normalize=1
|
327 |
+
filters=160
|
328 |
+
size=3
|
329 |
+
stride=1
|
330 |
+
pad=1
|
331 |
+
activation=mish
|
332 |
+
|
333 |
+
[shortcut]
|
334 |
+
from=-3
|
335 |
+
activation=linear
|
336 |
+
|
337 |
+
[convolutional]
|
338 |
+
batch_normalize=1
|
339 |
+
filters=160
|
340 |
+
size=1
|
341 |
+
stride=1
|
342 |
+
pad=1
|
343 |
+
activation=mish
|
344 |
+
|
345 |
+
[convolutional]
|
346 |
+
batch_normalize=1
|
347 |
+
filters=160
|
348 |
+
size=3
|
349 |
+
stride=1
|
350 |
+
pad=1
|
351 |
+
activation=mish
|
352 |
+
|
353 |
+
[shortcut]
|
354 |
+
from=-3
|
355 |
+
activation=linear
|
356 |
+
|
357 |
+
[convolutional]
|
358 |
+
batch_normalize=1
|
359 |
+
filters=160
|
360 |
+
size=1
|
361 |
+
stride=1
|
362 |
+
pad=1
|
363 |
+
activation=mish
|
364 |
+
|
365 |
+
[convolutional]
|
366 |
+
batch_normalize=1
|
367 |
+
filters=160
|
368 |
+
size=3
|
369 |
+
stride=1
|
370 |
+
pad=1
|
371 |
+
activation=mish
|
372 |
+
|
373 |
+
[shortcut]
|
374 |
+
from=-3
|
375 |
+
activation=linear
|
376 |
+
|
377 |
+
[convolutional]
|
378 |
+
batch_normalize=1
|
379 |
+
filters=160
|
380 |
+
size=1
|
381 |
+
stride=1
|
382 |
+
pad=1
|
383 |
+
activation=mish
|
384 |
+
|
385 |
+
[convolutional]
|
386 |
+
batch_normalize=1
|
387 |
+
filters=160
|
388 |
+
size=3
|
389 |
+
stride=1
|
390 |
+
pad=1
|
391 |
+
activation=mish
|
392 |
+
|
393 |
+
[shortcut]
|
394 |
+
from=-3
|
395 |
+
activation=linear
|
396 |
+
|
397 |
+
[convolutional]
|
398 |
+
batch_normalize=1
|
399 |
+
filters=160
|
400 |
+
size=1
|
401 |
+
stride=1
|
402 |
+
pad=1
|
403 |
+
activation=mish
|
404 |
+
|
405 |
+
[convolutional]
|
406 |
+
batch_normalize=1
|
407 |
+
filters=160
|
408 |
+
size=3
|
409 |
+
stride=1
|
410 |
+
pad=1
|
411 |
+
activation=mish
|
412 |
+
|
413 |
+
[shortcut]
|
414 |
+
from=-3
|
415 |
+
activation=linear
|
416 |
+
|
417 |
+
[convolutional]
|
418 |
+
batch_normalize=1
|
419 |
+
filters=160
|
420 |
+
size=1
|
421 |
+
stride=1
|
422 |
+
pad=1
|
423 |
+
activation=mish
|
424 |
+
|
425 |
+
[convolutional]
|
426 |
+
batch_normalize=1
|
427 |
+
filters=160
|
428 |
+
size=3
|
429 |
+
stride=1
|
430 |
+
pad=1
|
431 |
+
activation=mish
|
432 |
+
|
433 |
+
[shortcut]
|
434 |
+
from=-3
|
435 |
+
activation=linear
|
436 |
+
|
437 |
+
[convolutional]
|
438 |
+
batch_normalize=1
|
439 |
+
filters=160
|
440 |
+
size=1
|
441 |
+
stride=1
|
442 |
+
pad=1
|
443 |
+
activation=mish
|
444 |
+
|
445 |
+
[convolutional]
|
446 |
+
batch_normalize=1
|
447 |
+
filters=160
|
448 |
+
size=3
|
449 |
+
stride=1
|
450 |
+
pad=1
|
451 |
+
activation=mish
|
452 |
+
|
453 |
+
[shortcut]
|
454 |
+
from=-3
|
455 |
+
activation=linear
|
456 |
+
|
457 |
+
[convolutional]
|
458 |
+
batch_normalize=1
|
459 |
+
filters=160
|
460 |
+
size=1
|
461 |
+
stride=1
|
462 |
+
pad=1
|
463 |
+
activation=mish
|
464 |
+
|
465 |
+
[convolutional]
|
466 |
+
batch_normalize=1
|
467 |
+
filters=160
|
468 |
+
size=3
|
469 |
+
stride=1
|
470 |
+
pad=1
|
471 |
+
activation=mish
|
472 |
+
|
473 |
+
[shortcut]
|
474 |
+
from=-3
|
475 |
+
activation=linear
|
476 |
+
|
477 |
+
[convolutional]
|
478 |
+
batch_normalize=1
|
479 |
+
filters=160
|
480 |
+
size=1
|
481 |
+
stride=1
|
482 |
+
pad=1
|
483 |
+
activation=mish
|
484 |
+
|
485 |
+
[convolutional]
|
486 |
+
batch_normalize=1
|
487 |
+
filters=160
|
488 |
+
size=3
|
489 |
+
stride=1
|
490 |
+
pad=1
|
491 |
+
activation=mish
|
492 |
+
|
493 |
+
[shortcut]
|
494 |
+
from=-3
|
495 |
+
activation=linear
|
496 |
+
|
497 |
+
[convolutional]
|
498 |
+
batch_normalize=1
|
499 |
+
filters=160
|
500 |
+
size=1
|
501 |
+
stride=1
|
502 |
+
pad=1
|
503 |
+
activation=mish
|
504 |
+
|
505 |
+
[convolutional]
|
506 |
+
batch_normalize=1
|
507 |
+
filters=160
|
508 |
+
size=3
|
509 |
+
stride=1
|
510 |
+
pad=1
|
511 |
+
activation=mish
|
512 |
+
|
513 |
+
[shortcut]
|
514 |
+
from=-3
|
515 |
+
activation=linear
|
516 |
+
|
517 |
+
[convolutional]
|
518 |
+
batch_normalize=1
|
519 |
+
filters=160
|
520 |
+
size=1
|
521 |
+
stride=1
|
522 |
+
pad=1
|
523 |
+
activation=mish
|
524 |
+
|
525 |
+
[convolutional]
|
526 |
+
batch_normalize=1
|
527 |
+
filters=160
|
528 |
+
size=3
|
529 |
+
stride=1
|
530 |
+
pad=1
|
531 |
+
activation=mish
|
532 |
+
|
533 |
+
[shortcut]
|
534 |
+
from=-3
|
535 |
+
activation=linear
|
536 |
+
|
537 |
+
[convolutional]
|
538 |
+
batch_normalize=1
|
539 |
+
filters=160
|
540 |
+
size=1
|
541 |
+
stride=1
|
542 |
+
pad=1
|
543 |
+
activation=mish
|
544 |
+
|
545 |
+
[convolutional]
|
546 |
+
batch_normalize=1
|
547 |
+
filters=160
|
548 |
+
size=3
|
549 |
+
stride=1
|
550 |
+
pad=1
|
551 |
+
activation=mish
|
552 |
+
|
553 |
+
[shortcut]
|
554 |
+
from=-3
|
555 |
+
activation=linear
|
556 |
+
|
557 |
+
[convolutional]
|
558 |
+
batch_normalize=1
|
559 |
+
filters=160
|
560 |
+
size=1
|
561 |
+
stride=1
|
562 |
+
pad=1
|
563 |
+
activation=mish
|
564 |
+
|
565 |
+
[convolutional]
|
566 |
+
batch_normalize=1
|
567 |
+
filters=160
|
568 |
+
size=3
|
569 |
+
stride=1
|
570 |
+
pad=1
|
571 |
+
activation=mish
|
572 |
+
|
573 |
+
[shortcut]
|
574 |
+
from=-3
|
575 |
+
activation=linear
|
576 |
+
|
577 |
+
# Transition first
|
578 |
+
|
579 |
+
[convolutional]
|
580 |
+
batch_normalize=1
|
581 |
+
filters=160
|
582 |
+
size=1
|
583 |
+
stride=1
|
584 |
+
pad=1
|
585 |
+
activation=mish
|
586 |
+
|
587 |
+
# Merge [-1, -(3k+4)]
|
588 |
+
|
589 |
+
[route]
|
590 |
+
layers = -1,-49
|
591 |
+
|
592 |
+
# Transition last
|
593 |
+
|
594 |
+
# 78 (previous+7+3k)
|
595 |
+
[convolutional]
|
596 |
+
batch_normalize=1
|
597 |
+
filters=320
|
598 |
+
size=1
|
599 |
+
stride=1
|
600 |
+
pad=1
|
601 |
+
activation=mish
|
602 |
+
|
603 |
+
|
604 |
+
# P4
|
605 |
+
|
606 |
+
# Downsample
|
607 |
+
|
608 |
+
[convolutional]
|
609 |
+
batch_normalize=1
|
610 |
+
filters=640
|
611 |
+
size=3
|
612 |
+
stride=2
|
613 |
+
pad=1
|
614 |
+
activation=mish
|
615 |
+
|
616 |
+
# Split
|
617 |
+
|
618 |
+
[convolutional]
|
619 |
+
batch_normalize=1
|
620 |
+
filters=320
|
621 |
+
size=1
|
622 |
+
stride=1
|
623 |
+
pad=1
|
624 |
+
activation=mish
|
625 |
+
|
626 |
+
[route]
|
627 |
+
layers = -2
|
628 |
+
|
629 |
+
[convolutional]
|
630 |
+
batch_normalize=1
|
631 |
+
filters=320
|
632 |
+
size=1
|
633 |
+
stride=1
|
634 |
+
pad=1
|
635 |
+
activation=mish
|
636 |
+
|
637 |
+
# Residual Block
|
638 |
+
|
639 |
+
[convolutional]
|
640 |
+
batch_normalize=1
|
641 |
+
filters=320
|
642 |
+
size=1
|
643 |
+
stride=1
|
644 |
+
pad=1
|
645 |
+
activation=mish
|
646 |
+
|
647 |
+
[convolutional]
|
648 |
+
batch_normalize=1
|
649 |
+
filters=320
|
650 |
+
size=3
|
651 |
+
stride=1
|
652 |
+
pad=1
|
653 |
+
activation=mish
|
654 |
+
|
655 |
+
[shortcut]
|
656 |
+
from=-3
|
657 |
+
activation=linear
|
658 |
+
|
659 |
+
[convolutional]
|
660 |
+
batch_normalize=1
|
661 |
+
filters=320
|
662 |
+
size=1
|
663 |
+
stride=1
|
664 |
+
pad=1
|
665 |
+
activation=mish
|
666 |
+
|
667 |
+
[convolutional]
|
668 |
+
batch_normalize=1
|
669 |
+
filters=320
|
670 |
+
size=3
|
671 |
+
stride=1
|
672 |
+
pad=1
|
673 |
+
activation=mish
|
674 |
+
|
675 |
+
[shortcut]
|
676 |
+
from=-3
|
677 |
+
activation=linear
|
678 |
+
|
679 |
+
[convolutional]
|
680 |
+
batch_normalize=1
|
681 |
+
filters=320
|
682 |
+
size=1
|
683 |
+
stride=1
|
684 |
+
pad=1
|
685 |
+
activation=mish
|
686 |
+
|
687 |
+
[convolutional]
|
688 |
+
batch_normalize=1
|
689 |
+
filters=320
|
690 |
+
size=3
|
691 |
+
stride=1
|
692 |
+
pad=1
|
693 |
+
activation=mish
|
694 |
+
|
695 |
+
[shortcut]
|
696 |
+
from=-3
|
697 |
+
activation=linear
|
698 |
+
|
699 |
+
[convolutional]
|
700 |
+
batch_normalize=1
|
701 |
+
filters=320
|
702 |
+
size=1
|
703 |
+
stride=1
|
704 |
+
pad=1
|
705 |
+
activation=mish
|
706 |
+
|
707 |
+
[convolutional]
|
708 |
+
batch_normalize=1
|
709 |
+
filters=320
|
710 |
+
size=3
|
711 |
+
stride=1
|
712 |
+
pad=1
|
713 |
+
activation=mish
|
714 |
+
|
715 |
+
[shortcut]
|
716 |
+
from=-3
|
717 |
+
activation=linear
|
718 |
+
|
719 |
+
[convolutional]
|
720 |
+
batch_normalize=1
|
721 |
+
filters=320
|
722 |
+
size=1
|
723 |
+
stride=1
|
724 |
+
pad=1
|
725 |
+
activation=mish
|
726 |
+
|
727 |
+
[convolutional]
|
728 |
+
batch_normalize=1
|
729 |
+
filters=320
|
730 |
+
size=3
|
731 |
+
stride=1
|
732 |
+
pad=1
|
733 |
+
activation=mish
|
734 |
+
|
735 |
+
[shortcut]
|
736 |
+
from=-3
|
737 |
+
activation=linear
|
738 |
+
|
739 |
+
[convolutional]
|
740 |
+
batch_normalize=1
|
741 |
+
filters=320
|
742 |
+
size=1
|
743 |
+
stride=1
|
744 |
+
pad=1
|
745 |
+
activation=mish
|
746 |
+
|
747 |
+
[convolutional]
|
748 |
+
batch_normalize=1
|
749 |
+
filters=320
|
750 |
+
size=3
|
751 |
+
stride=1
|
752 |
+
pad=1
|
753 |
+
activation=mish
|
754 |
+
|
755 |
+
[shortcut]
|
756 |
+
from=-3
|
757 |
+
activation=linear
|
758 |
+
|
759 |
+
[convolutional]
|
760 |
+
batch_normalize=1
|
761 |
+
filters=320
|
762 |
+
size=1
|
763 |
+
stride=1
|
764 |
+
pad=1
|
765 |
+
activation=mish
|
766 |
+
|
767 |
+
[convolutional]
|
768 |
+
batch_normalize=1
|
769 |
+
filters=320
|
770 |
+
size=3
|
771 |
+
stride=1
|
772 |
+
pad=1
|
773 |
+
activation=mish
|
774 |
+
|
775 |
+
[shortcut]
|
776 |
+
from=-3
|
777 |
+
activation=linear
|
778 |
+
|
779 |
+
[convolutional]
|
780 |
+
batch_normalize=1
|
781 |
+
filters=320
|
782 |
+
size=1
|
783 |
+
stride=1
|
784 |
+
pad=1
|
785 |
+
activation=mish
|
786 |
+
|
787 |
+
[convolutional]
|
788 |
+
batch_normalize=1
|
789 |
+
filters=320
|
790 |
+
size=3
|
791 |
+
stride=1
|
792 |
+
pad=1
|
793 |
+
activation=mish
|
794 |
+
|
795 |
+
[shortcut]
|
796 |
+
from=-3
|
797 |
+
activation=linear
|
798 |
+
|
799 |
+
[convolutional]
|
800 |
+
batch_normalize=1
|
801 |
+
filters=320
|
802 |
+
size=1
|
803 |
+
stride=1
|
804 |
+
pad=1
|
805 |
+
activation=mish
|
806 |
+
|
807 |
+
[convolutional]
|
808 |
+
batch_normalize=1
|
809 |
+
filters=320
|
810 |
+
size=3
|
811 |
+
stride=1
|
812 |
+
pad=1
|
813 |
+
activation=mish
|
814 |
+
|
815 |
+
[shortcut]
|
816 |
+
from=-3
|
817 |
+
activation=linear
|
818 |
+
|
819 |
+
[convolutional]
|
820 |
+
batch_normalize=1
|
821 |
+
filters=320
|
822 |
+
size=1
|
823 |
+
stride=1
|
824 |
+
pad=1
|
825 |
+
activation=mish
|
826 |
+
|
827 |
+
[convolutional]
|
828 |
+
batch_normalize=1
|
829 |
+
filters=320
|
830 |
+
size=3
|
831 |
+
stride=1
|
832 |
+
pad=1
|
833 |
+
activation=mish
|
834 |
+
|
835 |
+
[shortcut]
|
836 |
+
from=-3
|
837 |
+
activation=linear
|
838 |
+
|
839 |
+
[convolutional]
|
840 |
+
batch_normalize=1
|
841 |
+
filters=320
|
842 |
+
size=1
|
843 |
+
stride=1
|
844 |
+
pad=1
|
845 |
+
activation=mish
|
846 |
+
|
847 |
+
[convolutional]
|
848 |
+
batch_normalize=1
|
849 |
+
filters=320
|
850 |
+
size=3
|
851 |
+
stride=1
|
852 |
+
pad=1
|
853 |
+
activation=mish
|
854 |
+
|
855 |
+
[shortcut]
|
856 |
+
from=-3
|
857 |
+
activation=linear
|
858 |
+
|
859 |
+
[convolutional]
|
860 |
+
batch_normalize=1
|
861 |
+
filters=320
|
862 |
+
size=1
|
863 |
+
stride=1
|
864 |
+
pad=1
|
865 |
+
activation=mish
|
866 |
+
|
867 |
+
[convolutional]
|
868 |
+
batch_normalize=1
|
869 |
+
filters=320
|
870 |
+
size=3
|
871 |
+
stride=1
|
872 |
+
pad=1
|
873 |
+
activation=mish
|
874 |
+
|
875 |
+
[shortcut]
|
876 |
+
from=-3
|
877 |
+
activation=linear
|
878 |
+
|
879 |
+
[convolutional]
|
880 |
+
batch_normalize=1
|
881 |
+
filters=320
|
882 |
+
size=1
|
883 |
+
stride=1
|
884 |
+
pad=1
|
885 |
+
activation=mish
|
886 |
+
|
887 |
+
[convolutional]
|
888 |
+
batch_normalize=1
|
889 |
+
filters=320
|
890 |
+
size=3
|
891 |
+
stride=1
|
892 |
+
pad=1
|
893 |
+
activation=mish
|
894 |
+
|
895 |
+
[shortcut]
|
896 |
+
from=-3
|
897 |
+
activation=linear
|
898 |
+
|
899 |
+
[convolutional]
|
900 |
+
batch_normalize=1
|
901 |
+
filters=320
|
902 |
+
size=1
|
903 |
+
stride=1
|
904 |
+
pad=1
|
905 |
+
activation=mish
|
906 |
+
|
907 |
+
[convolutional]
|
908 |
+
batch_normalize=1
|
909 |
+
filters=320
|
910 |
+
size=3
|
911 |
+
stride=1
|
912 |
+
pad=1
|
913 |
+
activation=mish
|
914 |
+
|
915 |
+
[shortcut]
|
916 |
+
from=-3
|
917 |
+
activation=linear
|
918 |
+
|
919 |
+
[convolutional]
|
920 |
+
batch_normalize=1
|
921 |
+
filters=320
|
922 |
+
size=1
|
923 |
+
stride=1
|
924 |
+
pad=1
|
925 |
+
activation=mish
|
926 |
+
|
927 |
+
[convolutional]
|
928 |
+
batch_normalize=1
|
929 |
+
filters=320
|
930 |
+
size=3
|
931 |
+
stride=1
|
932 |
+
pad=1
|
933 |
+
activation=mish
|
934 |
+
|
935 |
+
[shortcut]
|
936 |
+
from=-3
|
937 |
+
activation=linear
|
938 |
+
|
939 |
+
# Transition first
|
940 |
+
|
941 |
+
[convolutional]
|
942 |
+
batch_normalize=1
|
943 |
+
filters=320
|
944 |
+
size=1
|
945 |
+
stride=1
|
946 |
+
pad=1
|
947 |
+
activation=mish
|
948 |
+
|
949 |
+
# Merge [-1, -(3k+4)]
|
950 |
+
|
951 |
+
[route]
|
952 |
+
layers = -1,-49
|
953 |
+
|
954 |
+
# Transition last
|
955 |
+
|
956 |
+
# 130 (previous+7+3k)
|
957 |
+
[convolutional]
|
958 |
+
batch_normalize=1
|
959 |
+
filters=640
|
960 |
+
size=1
|
961 |
+
stride=1
|
962 |
+
pad=1
|
963 |
+
activation=mish
|
964 |
+
|
965 |
+
|
966 |
+
# P5
|
967 |
+
|
968 |
+
# Downsample
|
969 |
+
|
970 |
+
[convolutional]
|
971 |
+
batch_normalize=1
|
972 |
+
filters=1280
|
973 |
+
size=3
|
974 |
+
stride=2
|
975 |
+
pad=1
|
976 |
+
activation=mish
|
977 |
+
|
978 |
+
# Split
|
979 |
+
|
980 |
+
[convolutional]
|
981 |
+
batch_normalize=1
|
982 |
+
filters=640
|
983 |
+
size=1
|
984 |
+
stride=1
|
985 |
+
pad=1
|
986 |
+
activation=mish
|
987 |
+
|
988 |
+
[route]
|
989 |
+
layers = -2
|
990 |
+
|
991 |
+
[convolutional]
|
992 |
+
batch_normalize=1
|
993 |
+
filters=640
|
994 |
+
size=1
|
995 |
+
stride=1
|
996 |
+
pad=1
|
997 |
+
activation=mish
|
998 |
+
|
999 |
+
# Residual Block
|
1000 |
+
|
1001 |
+
[convolutional]
|
1002 |
+
batch_normalize=1
|
1003 |
+
filters=640
|
1004 |
+
size=1
|
1005 |
+
stride=1
|
1006 |
+
pad=1
|
1007 |
+
activation=mish
|
1008 |
+
|
1009 |
+
[convolutional]
|
1010 |
+
batch_normalize=1
|
1011 |
+
filters=640
|
1012 |
+
size=3
|
1013 |
+
stride=1
|
1014 |
+
pad=1
|
1015 |
+
activation=mish
|
1016 |
+
|
1017 |
+
[shortcut]
|
1018 |
+
from=-3
|
1019 |
+
activation=linear
|
1020 |
+
|
1021 |
+
[convolutional]
|
1022 |
+
batch_normalize=1
|
1023 |
+
filters=640
|
1024 |
+
size=1
|
1025 |
+
stride=1
|
1026 |
+
pad=1
|
1027 |
+
activation=mish
|
1028 |
+
|
1029 |
+
[convolutional]
|
1030 |
+
batch_normalize=1
|
1031 |
+
filters=640
|
1032 |
+
size=3
|
1033 |
+
stride=1
|
1034 |
+
pad=1
|
1035 |
+
activation=mish
|
1036 |
+
|
1037 |
+
[shortcut]
|
1038 |
+
from=-3
|
1039 |
+
activation=linear
|
1040 |
+
|
1041 |
+
[convolutional]
|
1042 |
+
batch_normalize=1
|
1043 |
+
filters=640
|
1044 |
+
size=1
|
1045 |
+
stride=1
|
1046 |
+
pad=1
|
1047 |
+
activation=mish
|
1048 |
+
|
1049 |
+
[convolutional]
|
1050 |
+
batch_normalize=1
|
1051 |
+
filters=640
|
1052 |
+
size=3
|
1053 |
+
stride=1
|
1054 |
+
pad=1
|
1055 |
+
activation=mish
|
1056 |
+
|
1057 |
+
[shortcut]
|
1058 |
+
from=-3
|
1059 |
+
activation=linear
|
1060 |
+
|
1061 |
+
[convolutional]
|
1062 |
+
batch_normalize=1
|
1063 |
+
filters=640
|
1064 |
+
size=1
|
1065 |
+
stride=1
|
1066 |
+
pad=1
|
1067 |
+
activation=mish
|
1068 |
+
|
1069 |
+
[convolutional]
|
1070 |
+
batch_normalize=1
|
1071 |
+
filters=640
|
1072 |
+
size=3
|
1073 |
+
stride=1
|
1074 |
+
pad=1
|
1075 |
+
activation=mish
|
1076 |
+
|
1077 |
+
[shortcut]
|
1078 |
+
from=-3
|
1079 |
+
activation=linear
|
1080 |
+
|
1081 |
+
[convolutional]
|
1082 |
+
batch_normalize=1
|
1083 |
+
filters=640
|
1084 |
+
size=1
|
1085 |
+
stride=1
|
1086 |
+
pad=1
|
1087 |
+
activation=mish
|
1088 |
+
|
1089 |
+
[convolutional]
|
1090 |
+
batch_normalize=1
|
1091 |
+
filters=640
|
1092 |
+
size=3
|
1093 |
+
stride=1
|
1094 |
+
pad=1
|
1095 |
+
activation=mish
|
1096 |
+
|
1097 |
+
[shortcut]
|
1098 |
+
from=-3
|
1099 |
+
activation=linear
|
1100 |
+
|
1101 |
+
[convolutional]
|
1102 |
+
batch_normalize=1
|
1103 |
+
filters=640
|
1104 |
+
size=1
|
1105 |
+
stride=1
|
1106 |
+
pad=1
|
1107 |
+
activation=mish
|
1108 |
+
|
1109 |
+
[convolutional]
|
1110 |
+
batch_normalize=1
|
1111 |
+
filters=640
|
1112 |
+
size=3
|
1113 |
+
stride=1
|
1114 |
+
pad=1
|
1115 |
+
activation=mish
|
1116 |
+
|
1117 |
+
[shortcut]
|
1118 |
+
from=-3
|
1119 |
+
activation=linear
|
1120 |
+
|
1121 |
+
[convolutional]
|
1122 |
+
batch_normalize=1
|
1123 |
+
filters=640
|
1124 |
+
size=1
|
1125 |
+
stride=1
|
1126 |
+
pad=1
|
1127 |
+
activation=mish
|
1128 |
+
|
1129 |
+
[convolutional]
|
1130 |
+
batch_normalize=1
|
1131 |
+
filters=640
|
1132 |
+
size=3
|
1133 |
+
stride=1
|
1134 |
+
pad=1
|
1135 |
+
activation=mish
|
1136 |
+
|
1137 |
+
[shortcut]
|
1138 |
+
from=-3
|
1139 |
+
activation=linear
|
1140 |
+
|
1141 |
+
# Transition first
|
1142 |
+
|
1143 |
+
[convolutional]
|
1144 |
+
batch_normalize=1
|
1145 |
+
filters=640
|
1146 |
+
size=1
|
1147 |
+
stride=1
|
1148 |
+
pad=1
|
1149 |
+
activation=mish
|
1150 |
+
|
1151 |
+
# Merge [-1, -(3k+4)]
|
1152 |
+
|
1153 |
+
[route]
|
1154 |
+
layers = -1,-25
|
1155 |
+
|
1156 |
+
# Transition last
|
1157 |
+
|
1158 |
+
# 158 (previous+7+3k)
|
1159 |
+
[convolutional]
|
1160 |
+
batch_normalize=1
|
1161 |
+
filters=1280
|
1162 |
+
size=1
|
1163 |
+
stride=1
|
1164 |
+
pad=1
|
1165 |
+
activation=mish
|
1166 |
+
|
1167 |
+
|
1168 |
+
# P6
|
1169 |
+
|
1170 |
+
# Downsample
|
1171 |
+
|
1172 |
+
[convolutional]
|
1173 |
+
batch_normalize=1
|
1174 |
+
filters=1280
|
1175 |
+
size=3
|
1176 |
+
stride=2
|
1177 |
+
pad=1
|
1178 |
+
activation=mish
|
1179 |
+
|
1180 |
+
# Split
|
1181 |
+
|
1182 |
+
[convolutional]
|
1183 |
+
batch_normalize=1
|
1184 |
+
filters=640
|
1185 |
+
size=1
|
1186 |
+
stride=1
|
1187 |
+
pad=1
|
1188 |
+
activation=mish
|
1189 |
+
|
1190 |
+
[route]
|
1191 |
+
layers = -2
|
1192 |
+
|
1193 |
+
[convolutional]
|
1194 |
+
batch_normalize=1
|
1195 |
+
filters=640
|
1196 |
+
size=1
|
1197 |
+
stride=1
|
1198 |
+
pad=1
|
1199 |
+
activation=mish
|
1200 |
+
|
1201 |
+
# Residual Block
|
1202 |
+
|
1203 |
+
[convolutional]
|
1204 |
+
batch_normalize=1
|
1205 |
+
filters=640
|
1206 |
+
size=1
|
1207 |
+
stride=1
|
1208 |
+
pad=1
|
1209 |
+
activation=mish
|
1210 |
+
|
1211 |
+
[convolutional]
|
1212 |
+
batch_normalize=1
|
1213 |
+
filters=640
|
1214 |
+
size=3
|
1215 |
+
stride=1
|
1216 |
+
pad=1
|
1217 |
+
activation=mish
|
1218 |
+
|
1219 |
+
[shortcut]
|
1220 |
+
from=-3
|
1221 |
+
activation=linear
|
1222 |
+
|
1223 |
+
[convolutional]
|
1224 |
+
batch_normalize=1
|
1225 |
+
filters=640
|
1226 |
+
size=1
|
1227 |
+
stride=1
|
1228 |
+
pad=1
|
1229 |
+
activation=mish
|
1230 |
+
|
1231 |
+
[convolutional]
|
1232 |
+
batch_normalize=1
|
1233 |
+
filters=640
|
1234 |
+
size=3
|
1235 |
+
stride=1
|
1236 |
+
pad=1
|
1237 |
+
activation=mish
|
1238 |
+
|
1239 |
+
[shortcut]
|
1240 |
+
from=-3
|
1241 |
+
activation=linear
|
1242 |
+
|
1243 |
+
[convolutional]
|
1244 |
+
batch_normalize=1
|
1245 |
+
filters=640
|
1246 |
+
size=1
|
1247 |
+
stride=1
|
1248 |
+
pad=1
|
1249 |
+
activation=mish
|
1250 |
+
|
1251 |
+
[convolutional]
|
1252 |
+
batch_normalize=1
|
1253 |
+
filters=640
|
1254 |
+
size=3
|
1255 |
+
stride=1
|
1256 |
+
pad=1
|
1257 |
+
activation=mish
|
1258 |
+
|
1259 |
+
[shortcut]
|
1260 |
+
from=-3
|
1261 |
+
activation=linear
|
1262 |
+
|
1263 |
+
[convolutional]
|
1264 |
+
batch_normalize=1
|
1265 |
+
filters=640
|
1266 |
+
size=1
|
1267 |
+
stride=1
|
1268 |
+
pad=1
|
1269 |
+
activation=mish
|
1270 |
+
|
1271 |
+
[convolutional]
|
1272 |
+
batch_normalize=1
|
1273 |
+
filters=640
|
1274 |
+
size=3
|
1275 |
+
stride=1
|
1276 |
+
pad=1
|
1277 |
+
activation=mish
|
1278 |
+
|
1279 |
+
[shortcut]
|
1280 |
+
from=-3
|
1281 |
+
activation=linear
|
1282 |
+
|
1283 |
+
[convolutional]
|
1284 |
+
batch_normalize=1
|
1285 |
+
filters=640
|
1286 |
+
size=1
|
1287 |
+
stride=1
|
1288 |
+
pad=1
|
1289 |
+
activation=mish
|
1290 |
+
|
1291 |
+
[convolutional]
|
1292 |
+
batch_normalize=1
|
1293 |
+
filters=640
|
1294 |
+
size=3
|
1295 |
+
stride=1
|
1296 |
+
pad=1
|
1297 |
+
activation=mish
|
1298 |
+
|
1299 |
+
[shortcut]
|
1300 |
+
from=-3
|
1301 |
+
activation=linear
|
1302 |
+
|
1303 |
+
[convolutional]
|
1304 |
+
batch_normalize=1
|
1305 |
+
filters=640
|
1306 |
+
size=1
|
1307 |
+
stride=1
|
1308 |
+
pad=1
|
1309 |
+
activation=mish
|
1310 |
+
|
1311 |
+
[convolutional]
|
1312 |
+
batch_normalize=1
|
1313 |
+
filters=640
|
1314 |
+
size=3
|
1315 |
+
stride=1
|
1316 |
+
pad=1
|
1317 |
+
activation=mish
|
1318 |
+
|
1319 |
+
[shortcut]
|
1320 |
+
from=-3
|
1321 |
+
activation=linear
|
1322 |
+
|
1323 |
+
[convolutional]
|
1324 |
+
batch_normalize=1
|
1325 |
+
filters=640
|
1326 |
+
size=1
|
1327 |
+
stride=1
|
1328 |
+
pad=1
|
1329 |
+
activation=mish
|
1330 |
+
|
1331 |
+
[convolutional]
|
1332 |
+
batch_normalize=1
|
1333 |
+
filters=640
|
1334 |
+
size=3
|
1335 |
+
stride=1
|
1336 |
+
pad=1
|
1337 |
+
activation=mish
|
1338 |
+
|
1339 |
+
[shortcut]
|
1340 |
+
from=-3
|
1341 |
+
activation=linear
|
1342 |
+
|
1343 |
+
# Transition first
|
1344 |
+
|
1345 |
+
[convolutional]
|
1346 |
+
batch_normalize=1
|
1347 |
+
filters=640
|
1348 |
+
size=1
|
1349 |
+
stride=1
|
1350 |
+
pad=1
|
1351 |
+
activation=mish
|
1352 |
+
|
1353 |
+
# Merge [-1, -(3k+4)]
|
1354 |
+
|
1355 |
+
[route]
|
1356 |
+
layers = -1,-25
|
1357 |
+
|
1358 |
+
# Transition last
|
1359 |
+
|
1360 |
+
# 186 (previous+7+3k)
|
1361 |
+
[convolutional]
|
1362 |
+
batch_normalize=1
|
1363 |
+
filters=1280
|
1364 |
+
size=1
|
1365 |
+
stride=1
|
1366 |
+
pad=1
|
1367 |
+
activation=mish
|
1368 |
+
|
1369 |
+
|
1370 |
+
# P7
|
1371 |
+
|
1372 |
+
# Downsample
|
1373 |
+
|
1374 |
+
[convolutional]
|
1375 |
+
batch_normalize=1
|
1376 |
+
filters=1280
|
1377 |
+
size=3
|
1378 |
+
stride=2
|
1379 |
+
pad=1
|
1380 |
+
activation=mish
|
1381 |
+
|
1382 |
+
# Split
|
1383 |
+
|
1384 |
+
[convolutional]
|
1385 |
+
batch_normalize=1
|
1386 |
+
filters=640
|
1387 |
+
size=1
|
1388 |
+
stride=1
|
1389 |
+
pad=1
|
1390 |
+
activation=mish
|
1391 |
+
|
1392 |
+
[route]
|
1393 |
+
layers = -2
|
1394 |
+
|
1395 |
+
[convolutional]
|
1396 |
+
batch_normalize=1
|
1397 |
+
filters=640
|
1398 |
+
size=1
|
1399 |
+
stride=1
|
1400 |
+
pad=1
|
1401 |
+
activation=mish
|
1402 |
+
|
1403 |
+
# Residual Block
|
1404 |
+
|
1405 |
+
[convolutional]
|
1406 |
+
batch_normalize=1
|
1407 |
+
filters=640
|
1408 |
+
size=1
|
1409 |
+
stride=1
|
1410 |
+
pad=1
|
1411 |
+
activation=mish
|
1412 |
+
|
1413 |
+
[convolutional]
|
1414 |
+
batch_normalize=1
|
1415 |
+
filters=640
|
1416 |
+
size=3
|
1417 |
+
stride=1
|
1418 |
+
pad=1
|
1419 |
+
activation=mish
|
1420 |
+
|
1421 |
+
[shortcut]
|
1422 |
+
from=-3
|
1423 |
+
activation=linear
|
1424 |
+
|
1425 |
+
[convolutional]
|
1426 |
+
batch_normalize=1
|
1427 |
+
filters=640
|
1428 |
+
size=1
|
1429 |
+
stride=1
|
1430 |
+
pad=1
|
1431 |
+
activation=mish
|
1432 |
+
|
1433 |
+
[convolutional]
|
1434 |
+
batch_normalize=1
|
1435 |
+
filters=640
|
1436 |
+
size=3
|
1437 |
+
stride=1
|
1438 |
+
pad=1
|
1439 |
+
activation=mish
|
1440 |
+
|
1441 |
+
[shortcut]
|
1442 |
+
from=-3
|
1443 |
+
activation=linear
|
1444 |
+
|
1445 |
+
[convolutional]
|
1446 |
+
batch_normalize=1
|
1447 |
+
filters=640
|
1448 |
+
size=1
|
1449 |
+
stride=1
|
1450 |
+
pad=1
|
1451 |
+
activation=mish
|
1452 |
+
|
1453 |
+
[convolutional]
|
1454 |
+
batch_normalize=1
|
1455 |
+
filters=640
|
1456 |
+
size=3
|
1457 |
+
stride=1
|
1458 |
+
pad=1
|
1459 |
+
activation=mish
|
1460 |
+
|
1461 |
+
[shortcut]
|
1462 |
+
from=-3
|
1463 |
+
activation=linear
|
1464 |
+
|
1465 |
+
[convolutional]
|
1466 |
+
batch_normalize=1
|
1467 |
+
filters=640
|
1468 |
+
size=1
|
1469 |
+
stride=1
|
1470 |
+
pad=1
|
1471 |
+
activation=mish
|
1472 |
+
|
1473 |
+
[convolutional]
|
1474 |
+
batch_normalize=1
|
1475 |
+
filters=640
|
1476 |
+
size=3
|
1477 |
+
stride=1
|
1478 |
+
pad=1
|
1479 |
+
activation=mish
|
1480 |
+
|
1481 |
+
[shortcut]
|
1482 |
+
from=-3
|
1483 |
+
activation=linear
|
1484 |
+
|
1485 |
+
[convolutional]
|
1486 |
+
batch_normalize=1
|
1487 |
+
filters=640
|
1488 |
+
size=1
|
1489 |
+
stride=1
|
1490 |
+
pad=1
|
1491 |
+
activation=mish
|
1492 |
+
|
1493 |
+
[convolutional]
|
1494 |
+
batch_normalize=1
|
1495 |
+
filters=640
|
1496 |
+
size=3
|
1497 |
+
stride=1
|
1498 |
+
pad=1
|
1499 |
+
activation=mish
|
1500 |
+
|
1501 |
+
[shortcut]
|
1502 |
+
from=-3
|
1503 |
+
activation=linear
|
1504 |
+
|
1505 |
+
[convolutional]
|
1506 |
+
batch_normalize=1
|
1507 |
+
filters=640
|
1508 |
+
size=1
|
1509 |
+
stride=1
|
1510 |
+
pad=1
|
1511 |
+
activation=mish
|
1512 |
+
|
1513 |
+
[convolutional]
|
1514 |
+
batch_normalize=1
|
1515 |
+
filters=640
|
1516 |
+
size=3
|
1517 |
+
stride=1
|
1518 |
+
pad=1
|
1519 |
+
activation=mish
|
1520 |
+
|
1521 |
+
[shortcut]
|
1522 |
+
from=-3
|
1523 |
+
activation=linear
|
1524 |
+
|
1525 |
+
[convolutional]
|
1526 |
+
batch_normalize=1
|
1527 |
+
filters=640
|
1528 |
+
size=1
|
1529 |
+
stride=1
|
1530 |
+
pad=1
|
1531 |
+
activation=mish
|
1532 |
+
|
1533 |
+
[convolutional]
|
1534 |
+
batch_normalize=1
|
1535 |
+
filters=640
|
1536 |
+
size=3
|
1537 |
+
stride=1
|
1538 |
+
pad=1
|
1539 |
+
activation=mish
|
1540 |
+
|
1541 |
+
[shortcut]
|
1542 |
+
from=-3
|
1543 |
+
activation=linear
|
1544 |
+
|
1545 |
+
# Transition first
|
1546 |
+
|
1547 |
+
[convolutional]
|
1548 |
+
batch_normalize=1
|
1549 |
+
filters=640
|
1550 |
+
size=1
|
1551 |
+
stride=1
|
1552 |
+
pad=1
|
1553 |
+
activation=mish
|
1554 |
+
|
1555 |
+
# Merge [-1, -(3k+4)]
|
1556 |
+
|
1557 |
+
[route]
|
1558 |
+
layers = -1,-25
|
1559 |
+
|
1560 |
+
# Transition last
|
1561 |
+
|
1562 |
+
# 214 (previous+7+3k)
|
1563 |
+
[convolutional]
|
1564 |
+
batch_normalize=1
|
1565 |
+
filters=1280
|
1566 |
+
size=1
|
1567 |
+
stride=1
|
1568 |
+
pad=1
|
1569 |
+
activation=mish
|
1570 |
+
|
1571 |
+
# ============ End of Backbone ============ #
|
1572 |
+
|
1573 |
+
# ============ Neck ============ #
|
1574 |
+
|
1575 |
+
# CSPSPP
|
1576 |
+
|
1577 |
+
[convolutional]
|
1578 |
+
batch_normalize=1
|
1579 |
+
filters=640
|
1580 |
+
size=1
|
1581 |
+
stride=1
|
1582 |
+
pad=1
|
1583 |
+
activation=mish
|
1584 |
+
|
1585 |
+
[route]
|
1586 |
+
layers = -2
|
1587 |
+
|
1588 |
+
[convolutional]
|
1589 |
+
batch_normalize=1
|
1590 |
+
filters=640
|
1591 |
+
size=1
|
1592 |
+
stride=1
|
1593 |
+
pad=1
|
1594 |
+
activation=mish
|
1595 |
+
|
1596 |
+
[convolutional]
|
1597 |
+
batch_normalize=1
|
1598 |
+
size=3
|
1599 |
+
stride=1
|
1600 |
+
pad=1
|
1601 |
+
filters=640
|
1602 |
+
activation=mish
|
1603 |
+
|
1604 |
+
[convolutional]
|
1605 |
+
batch_normalize=1
|
1606 |
+
filters=640
|
1607 |
+
size=1
|
1608 |
+
stride=1
|
1609 |
+
pad=1
|
1610 |
+
activation=mish
|
1611 |
+
|
1612 |
+
### SPP ###
|
1613 |
+
[maxpool]
|
1614 |
+
stride=1
|
1615 |
+
size=5
|
1616 |
+
|
1617 |
+
[route]
|
1618 |
+
layers=-2
|
1619 |
+
|
1620 |
+
[maxpool]
|
1621 |
+
stride=1
|
1622 |
+
size=9
|
1623 |
+
|
1624 |
+
[route]
|
1625 |
+
layers=-4
|
1626 |
+
|
1627 |
+
[maxpool]
|
1628 |
+
stride=1
|
1629 |
+
size=13
|
1630 |
+
|
1631 |
+
[route]
|
1632 |
+
layers=-1,-3,-5,-6
|
1633 |
+
### End SPP ###
|
1634 |
+
|
1635 |
+
[convolutional]
|
1636 |
+
batch_normalize=1
|
1637 |
+
filters=640
|
1638 |
+
size=1
|
1639 |
+
stride=1
|
1640 |
+
pad=1
|
1641 |
+
activation=mish
|
1642 |
+
|
1643 |
+
[convolutional]
|
1644 |
+
batch_normalize=1
|
1645 |
+
size=3
|
1646 |
+
stride=1
|
1647 |
+
pad=1
|
1648 |
+
filters=640
|
1649 |
+
activation=mish
|
1650 |
+
|
1651 |
+
[route]
|
1652 |
+
layers = -1, -13
|
1653 |
+
|
1654 |
+
# 229 (previous+6+5+2k)
|
1655 |
+
[convolutional]
|
1656 |
+
batch_normalize=1
|
1657 |
+
filters=640
|
1658 |
+
size=1
|
1659 |
+
stride=1
|
1660 |
+
pad=1
|
1661 |
+
activation=mish
|
1662 |
+
|
1663 |
+
# End of CSPSPP
|
1664 |
+
|
1665 |
+
|
1666 |
+
# FPN-6
|
1667 |
+
|
1668 |
+
[convolutional]
|
1669 |
+
batch_normalize=1
|
1670 |
+
filters=640
|
1671 |
+
size=1
|
1672 |
+
stride=1
|
1673 |
+
pad=1
|
1674 |
+
activation=mish
|
1675 |
+
|
1676 |
+
[upsample]
|
1677 |
+
stride=2
|
1678 |
+
|
1679 |
+
[route]
|
1680 |
+
layers = 186
|
1681 |
+
|
1682 |
+
[convolutional]
|
1683 |
+
batch_normalize=1
|
1684 |
+
filters=640
|
1685 |
+
size=1
|
1686 |
+
stride=1
|
1687 |
+
pad=1
|
1688 |
+
activation=mish
|
1689 |
+
|
1690 |
+
[route]
|
1691 |
+
layers = -1, -3
|
1692 |
+
|
1693 |
+
[convolutional]
|
1694 |
+
batch_normalize=1
|
1695 |
+
filters=640
|
1696 |
+
size=1
|
1697 |
+
stride=1
|
1698 |
+
pad=1
|
1699 |
+
activation=mish
|
1700 |
+
|
1701 |
+
# Split
|
1702 |
+
|
1703 |
+
[convolutional]
|
1704 |
+
batch_normalize=1
|
1705 |
+
filters=640
|
1706 |
+
size=1
|
1707 |
+
stride=1
|
1708 |
+
pad=1
|
1709 |
+
activation=mish
|
1710 |
+
|
1711 |
+
[route]
|
1712 |
+
layers = -2
|
1713 |
+
|
1714 |
+
# Plain Block
|
1715 |
+
|
1716 |
+
[convolutional]
|
1717 |
+
batch_normalize=1
|
1718 |
+
filters=640
|
1719 |
+
size=1
|
1720 |
+
stride=1
|
1721 |
+
pad=1
|
1722 |
+
activation=mish
|
1723 |
+
|
1724 |
+
[convolutional]
|
1725 |
+
batch_normalize=1
|
1726 |
+
size=3
|
1727 |
+
stride=1
|
1728 |
+
pad=1
|
1729 |
+
filters=640
|
1730 |
+
activation=mish
|
1731 |
+
|
1732 |
+
[convolutional]
|
1733 |
+
batch_normalize=1
|
1734 |
+
filters=640
|
1735 |
+
size=1
|
1736 |
+
stride=1
|
1737 |
+
pad=1
|
1738 |
+
activation=mish
|
1739 |
+
|
1740 |
+
[convolutional]
|
1741 |
+
batch_normalize=1
|
1742 |
+
size=3
|
1743 |
+
stride=1
|
1744 |
+
pad=1
|
1745 |
+
filters=640
|
1746 |
+
activation=mish
|
1747 |
+
|
1748 |
+
[convolutional]
|
1749 |
+
batch_normalize=1
|
1750 |
+
filters=640
|
1751 |
+
size=1
|
1752 |
+
stride=1
|
1753 |
+
pad=1
|
1754 |
+
activation=mish
|
1755 |
+
|
1756 |
+
[convolutional]
|
1757 |
+
batch_normalize=1
|
1758 |
+
size=3
|
1759 |
+
stride=1
|
1760 |
+
pad=1
|
1761 |
+
filters=640
|
1762 |
+
activation=mish
|
1763 |
+
|
1764 |
+
# Merge [-1, -(2k+2)]
|
1765 |
+
|
1766 |
+
[route]
|
1767 |
+
layers = -1, -8
|
1768 |
+
|
1769 |
+
# Transition last
|
1770 |
+
|
1771 |
+
# 245 (previous+6+4+2k)
|
1772 |
+
[convolutional]
|
1773 |
+
batch_normalize=1
|
1774 |
+
filters=640
|
1775 |
+
size=1
|
1776 |
+
stride=1
|
1777 |
+
pad=1
|
1778 |
+
activation=mish
|
1779 |
+
|
1780 |
+
|
1781 |
+
# FPN-5
|
1782 |
+
|
1783 |
+
[convolutional]
|
1784 |
+
batch_normalize=1
|
1785 |
+
filters=640
|
1786 |
+
size=1
|
1787 |
+
stride=1
|
1788 |
+
pad=1
|
1789 |
+
activation=mish
|
1790 |
+
|
1791 |
+
[upsample]
|
1792 |
+
stride=2
|
1793 |
+
|
1794 |
+
[route]
|
1795 |
+
layers = 158
|
1796 |
+
|
1797 |
+
[convolutional]
|
1798 |
+
batch_normalize=1
|
1799 |
+
filters=640
|
1800 |
+
size=1
|
1801 |
+
stride=1
|
1802 |
+
pad=1
|
1803 |
+
activation=mish
|
1804 |
+
|
1805 |
+
[route]
|
1806 |
+
layers = -1, -3
|
1807 |
+
|
1808 |
+
[convolutional]
|
1809 |
+
batch_normalize=1
|
1810 |
+
filters=640
|
1811 |
+
size=1
|
1812 |
+
stride=1
|
1813 |
+
pad=1
|
1814 |
+
activation=mish
|
1815 |
+
|
1816 |
+
# Split
|
1817 |
+
|
1818 |
+
[convolutional]
|
1819 |
+
batch_normalize=1
|
1820 |
+
filters=640
|
1821 |
+
size=1
|
1822 |
+
stride=1
|
1823 |
+
pad=1
|
1824 |
+
activation=mish
|
1825 |
+
|
1826 |
+
[route]
|
1827 |
+
layers = -2
|
1828 |
+
|
1829 |
+
# Plain Block
|
1830 |
+
|
1831 |
+
[convolutional]
|
1832 |
+
batch_normalize=1
|
1833 |
+
filters=640
|
1834 |
+
size=1
|
1835 |
+
stride=1
|
1836 |
+
pad=1
|
1837 |
+
activation=mish
|
1838 |
+
|
1839 |
+
[convolutional]
|
1840 |
+
batch_normalize=1
|
1841 |
+
size=3
|
1842 |
+
stride=1
|
1843 |
+
pad=1
|
1844 |
+
filters=640
|
1845 |
+
activation=mish
|
1846 |
+
|
1847 |
+
[convolutional]
|
1848 |
+
batch_normalize=1
|
1849 |
+
filters=640
|
1850 |
+
size=1
|
1851 |
+
stride=1
|
1852 |
+
pad=1
|
1853 |
+
activation=mish
|
1854 |
+
|
1855 |
+
[convolutional]
|
1856 |
+
batch_normalize=1
|
1857 |
+
size=3
|
1858 |
+
stride=1
|
1859 |
+
pad=1
|
1860 |
+
filters=640
|
1861 |
+
activation=mish
|
1862 |
+
|
1863 |
+
[convolutional]
|
1864 |
+
batch_normalize=1
|
1865 |
+
filters=640
|
1866 |
+
size=1
|
1867 |
+
stride=1
|
1868 |
+
pad=1
|
1869 |
+
activation=mish
|
1870 |
+
|
1871 |
+
[convolutional]
|
1872 |
+
batch_normalize=1
|
1873 |
+
size=3
|
1874 |
+
stride=1
|
1875 |
+
pad=1
|
1876 |
+
filters=640
|
1877 |
+
activation=mish
|
1878 |
+
|
1879 |
+
# Merge [-1, -(2k+2)]
|
1880 |
+
|
1881 |
+
[route]
|
1882 |
+
layers = -1, -8
|
1883 |
+
|
1884 |
+
# Transition last
|
1885 |
+
|
1886 |
+
# 261 (previous+6+4+2k)
|
1887 |
+
[convolutional]
|
1888 |
+
batch_normalize=1
|
1889 |
+
filters=640
|
1890 |
+
size=1
|
1891 |
+
stride=1
|
1892 |
+
pad=1
|
1893 |
+
activation=mish
|
1894 |
+
|
1895 |
+
|
1896 |
+
# FPN-4
|
1897 |
+
|
1898 |
+
[convolutional]
|
1899 |
+
batch_normalize=1
|
1900 |
+
filters=320
|
1901 |
+
size=1
|
1902 |
+
stride=1
|
1903 |
+
pad=1
|
1904 |
+
activation=mish
|
1905 |
+
|
1906 |
+
[upsample]
|
1907 |
+
stride=2
|
1908 |
+
|
1909 |
+
[route]
|
1910 |
+
layers = 130
|
1911 |
+
|
1912 |
+
[convolutional]
|
1913 |
+
batch_normalize=1
|
1914 |
+
filters=320
|
1915 |
+
size=1
|
1916 |
+
stride=1
|
1917 |
+
pad=1
|
1918 |
+
activation=mish
|
1919 |
+
|
1920 |
+
[route]
|
1921 |
+
layers = -1, -3
|
1922 |
+
|
1923 |
+
[convolutional]
|
1924 |
+
batch_normalize=1
|
1925 |
+
filters=320
|
1926 |
+
size=1
|
1927 |
+
stride=1
|
1928 |
+
pad=1
|
1929 |
+
activation=mish
|
1930 |
+
|
1931 |
+
# Split
|
1932 |
+
|
1933 |
+
[convolutional]
|
1934 |
+
batch_normalize=1
|
1935 |
+
filters=320
|
1936 |
+
size=1
|
1937 |
+
stride=1
|
1938 |
+
pad=1
|
1939 |
+
activation=mish
|
1940 |
+
|
1941 |
+
[route]
|
1942 |
+
layers = -2
|
1943 |
+
|
1944 |
+
# Plain Block
|
1945 |
+
|
1946 |
+
[convolutional]
|
1947 |
+
batch_normalize=1
|
1948 |
+
filters=320
|
1949 |
+
size=1
|
1950 |
+
stride=1
|
1951 |
+
pad=1
|
1952 |
+
activation=mish
|
1953 |
+
|
1954 |
+
[convolutional]
|
1955 |
+
batch_normalize=1
|
1956 |
+
size=3
|
1957 |
+
stride=1
|
1958 |
+
pad=1
|
1959 |
+
filters=320
|
1960 |
+
activation=mish
|
1961 |
+
|
1962 |
+
[convolutional]
|
1963 |
+
batch_normalize=1
|
1964 |
+
filters=320
|
1965 |
+
size=1
|
1966 |
+
stride=1
|
1967 |
+
pad=1
|
1968 |
+
activation=mish
|
1969 |
+
|
1970 |
+
[convolutional]
|
1971 |
+
batch_normalize=1
|
1972 |
+
size=3
|
1973 |
+
stride=1
|
1974 |
+
pad=1
|
1975 |
+
filters=320
|
1976 |
+
activation=mish
|
1977 |
+
|
1978 |
+
[convolutional]
|
1979 |
+
batch_normalize=1
|
1980 |
+
filters=320
|
1981 |
+
size=1
|
1982 |
+
stride=1
|
1983 |
+
pad=1
|
1984 |
+
activation=mish
|
1985 |
+
|
1986 |
+
[convolutional]
|
1987 |
+
batch_normalize=1
|
1988 |
+
size=3
|
1989 |
+
stride=1
|
1990 |
+
pad=1
|
1991 |
+
filters=320
|
1992 |
+
activation=mish
|
1993 |
+
|
1994 |
+
# Merge [-1, -(2k+2)]
|
1995 |
+
|
1996 |
+
[route]
|
1997 |
+
layers = -1, -8
|
1998 |
+
|
1999 |
+
# Transition last
|
2000 |
+
|
2001 |
+
# 277 (previous+6+4+2k)
|
2002 |
+
[convolutional]
|
2003 |
+
batch_normalize=1
|
2004 |
+
filters=320
|
2005 |
+
size=1
|
2006 |
+
stride=1
|
2007 |
+
pad=1
|
2008 |
+
activation=mish
|
2009 |
+
|
2010 |
+
|
2011 |
+
# FPN-3
|
2012 |
+
|
2013 |
+
[convolutional]
|
2014 |
+
batch_normalize=1
|
2015 |
+
filters=160
|
2016 |
+
size=1
|
2017 |
+
stride=1
|
2018 |
+
pad=1
|
2019 |
+
activation=mish
|
2020 |
+
|
2021 |
+
[upsample]
|
2022 |
+
stride=2
|
2023 |
+
|
2024 |
+
[route]
|
2025 |
+
layers = 78
|
2026 |
+
|
2027 |
+
[convolutional]
|
2028 |
+
batch_normalize=1
|
2029 |
+
filters=160
|
2030 |
+
size=1
|
2031 |
+
stride=1
|
2032 |
+
pad=1
|
2033 |
+
activation=mish
|
2034 |
+
|
2035 |
+
[route]
|
2036 |
+
layers = -1, -3
|
2037 |
+
|
2038 |
+
[convolutional]
|
2039 |
+
batch_normalize=1
|
2040 |
+
filters=160
|
2041 |
+
size=1
|
2042 |
+
stride=1
|
2043 |
+
pad=1
|
2044 |
+
activation=mish
|
2045 |
+
|
2046 |
+
# Split
|
2047 |
+
|
2048 |
+
[convolutional]
|
2049 |
+
batch_normalize=1
|
2050 |
+
filters=160
|
2051 |
+
size=1
|
2052 |
+
stride=1
|
2053 |
+
pad=1
|
2054 |
+
activation=mish
|
2055 |
+
|
2056 |
+
[route]
|
2057 |
+
layers = -2
|
2058 |
+
|
2059 |
+
# Plain Block
|
2060 |
+
|
2061 |
+
[convolutional]
|
2062 |
+
batch_normalize=1
|
2063 |
+
filters=160
|
2064 |
+
size=1
|
2065 |
+
stride=1
|
2066 |
+
pad=1
|
2067 |
+
activation=mish
|
2068 |
+
|
2069 |
+
[convolutional]
|
2070 |
+
batch_normalize=1
|
2071 |
+
size=3
|
2072 |
+
stride=1
|
2073 |
+
pad=1
|
2074 |
+
filters=160
|
2075 |
+
activation=mish
|
2076 |
+
|
2077 |
+
[convolutional]
|
2078 |
+
batch_normalize=1
|
2079 |
+
filters=160
|
2080 |
+
size=1
|
2081 |
+
stride=1
|
2082 |
+
pad=1
|
2083 |
+
activation=mish
|
2084 |
+
|
2085 |
+
[convolutional]
|
2086 |
+
batch_normalize=1
|
2087 |
+
size=3
|
2088 |
+
stride=1
|
2089 |
+
pad=1
|
2090 |
+
filters=160
|
2091 |
+
activation=mish
|
2092 |
+
|
2093 |
+
[convolutional]
|
2094 |
+
batch_normalize=1
|
2095 |
+
filters=160
|
2096 |
+
size=1
|
2097 |
+
stride=1
|
2098 |
+
pad=1
|
2099 |
+
activation=mish
|
2100 |
+
|
2101 |
+
[convolutional]
|
2102 |
+
batch_normalize=1
|
2103 |
+
size=3
|
2104 |
+
stride=1
|
2105 |
+
pad=1
|
2106 |
+
filters=160
|
2107 |
+
activation=mish
|
2108 |
+
|
2109 |
+
# Merge [-1, -(2k+2)]
|
2110 |
+
|
2111 |
+
[route]
|
2112 |
+
layers = -1, -8
|
2113 |
+
|
2114 |
+
# Transition last
|
2115 |
+
|
2116 |
+
# 293 (previous+6+4+2k)
|
2117 |
+
[convolutional]
|
2118 |
+
batch_normalize=1
|
2119 |
+
filters=160
|
2120 |
+
size=1
|
2121 |
+
stride=1
|
2122 |
+
pad=1
|
2123 |
+
activation=mish
|
2124 |
+
|
2125 |
+
|
2126 |
+
# PAN-4
|
2127 |
+
|
2128 |
+
[convolutional]
|
2129 |
+
batch_normalize=1
|
2130 |
+
size=3
|
2131 |
+
stride=2
|
2132 |
+
pad=1
|
2133 |
+
filters=320
|
2134 |
+
activation=mish
|
2135 |
+
|
2136 |
+
[route]
|
2137 |
+
layers = -1, 277
|
2138 |
+
|
2139 |
+
[convolutional]
|
2140 |
+
batch_normalize=1
|
2141 |
+
filters=320
|
2142 |
+
size=1
|
2143 |
+
stride=1
|
2144 |
+
pad=1
|
2145 |
+
activation=mish
|
2146 |
+
|
2147 |
+
# Split
|
2148 |
+
|
2149 |
+
[convolutional]
|
2150 |
+
batch_normalize=1
|
2151 |
+
filters=320
|
2152 |
+
size=1
|
2153 |
+
stride=1
|
2154 |
+
pad=1
|
2155 |
+
activation=mish
|
2156 |
+
|
2157 |
+
[route]
|
2158 |
+
layers = -2
|
2159 |
+
|
2160 |
+
# Plain Block
|
2161 |
+
|
2162 |
+
[convolutional]
|
2163 |
+
batch_normalize=1
|
2164 |
+
filters=320
|
2165 |
+
size=1
|
2166 |
+
stride=1
|
2167 |
+
pad=1
|
2168 |
+
activation=mish
|
2169 |
+
|
2170 |
+
[convolutional]
|
2171 |
+
batch_normalize=1
|
2172 |
+
size=3
|
2173 |
+
stride=1
|
2174 |
+
pad=1
|
2175 |
+
filters=320
|
2176 |
+
activation=mish
|
2177 |
+
|
2178 |
+
[convolutional]
|
2179 |
+
batch_normalize=1
|
2180 |
+
filters=320
|
2181 |
+
size=1
|
2182 |
+
stride=1
|
2183 |
+
pad=1
|
2184 |
+
activation=mish
|
2185 |
+
|
2186 |
+
[convolutional]
|
2187 |
+
batch_normalize=1
|
2188 |
+
size=3
|
2189 |
+
stride=1
|
2190 |
+
pad=1
|
2191 |
+
filters=320
|
2192 |
+
activation=mish
|
2193 |
+
|
2194 |
+
[convolutional]
|
2195 |
+
batch_normalize=1
|
2196 |
+
filters=320
|
2197 |
+
size=1
|
2198 |
+
stride=1
|
2199 |
+
pad=1
|
2200 |
+
activation=mish
|
2201 |
+
|
2202 |
+
[convolutional]
|
2203 |
+
batch_normalize=1
|
2204 |
+
size=3
|
2205 |
+
stride=1
|
2206 |
+
pad=1
|
2207 |
+
filters=320
|
2208 |
+
activation=mish
|
2209 |
+
|
2210 |
+
[route]
|
2211 |
+
layers = -1,-8
|
2212 |
+
|
2213 |
+
# Transition last
|
2214 |
+
|
2215 |
+
# 306 (previous+3+4+2k)
|
2216 |
+
[convolutional]
|
2217 |
+
batch_normalize=1
|
2218 |
+
filters=320
|
2219 |
+
size=1
|
2220 |
+
stride=1
|
2221 |
+
pad=1
|
2222 |
+
activation=mish
|
2223 |
+
|
2224 |
+
|
2225 |
+
# PAN-5
|
2226 |
+
|
2227 |
+
[convolutional]
|
2228 |
+
batch_normalize=1
|
2229 |
+
size=3
|
2230 |
+
stride=2
|
2231 |
+
pad=1
|
2232 |
+
filters=640
|
2233 |
+
activation=mish
|
2234 |
+
|
2235 |
+
[route]
|
2236 |
+
layers = -1, 261
|
2237 |
+
|
2238 |
+
[convolutional]
|
2239 |
+
batch_normalize=1
|
2240 |
+
filters=640
|
2241 |
+
size=1
|
2242 |
+
stride=1
|
2243 |
+
pad=1
|
2244 |
+
activation=mish
|
2245 |
+
|
2246 |
+
# Split
|
2247 |
+
|
2248 |
+
[convolutional]
|
2249 |
+
batch_normalize=1
|
2250 |
+
filters=640
|
2251 |
+
size=1
|
2252 |
+
stride=1
|
2253 |
+
pad=1
|
2254 |
+
activation=mish
|
2255 |
+
|
2256 |
+
[route]
|
2257 |
+
layers = -2
|
2258 |
+
|
2259 |
+
# Plain Block
|
2260 |
+
|
2261 |
+
[convolutional]
|
2262 |
+
batch_normalize=1
|
2263 |
+
filters=640
|
2264 |
+
size=1
|
2265 |
+
stride=1
|
2266 |
+
pad=1
|
2267 |
+
activation=mish
|
2268 |
+
|
2269 |
+
[convolutional]
|
2270 |
+
batch_normalize=1
|
2271 |
+
size=3
|
2272 |
+
stride=1
|
2273 |
+
pad=1
|
2274 |
+
filters=640
|
2275 |
+
activation=mish
|
2276 |
+
|
2277 |
+
[convolutional]
|
2278 |
+
batch_normalize=1
|
2279 |
+
filters=640
|
2280 |
+
size=1
|
2281 |
+
stride=1
|
2282 |
+
pad=1
|
2283 |
+
activation=mish
|
2284 |
+
|
2285 |
+
[convolutional]
|
2286 |
+
batch_normalize=1
|
2287 |
+
size=3
|
2288 |
+
stride=1
|
2289 |
+
pad=1
|
2290 |
+
filters=640
|
2291 |
+
activation=mish
|
2292 |
+
|
2293 |
+
[convolutional]
|
2294 |
+
batch_normalize=1
|
2295 |
+
filters=640
|
2296 |
+
size=1
|
2297 |
+
stride=1
|
2298 |
+
pad=1
|
2299 |
+
activation=mish
|
2300 |
+
|
2301 |
+
[convolutional]
|
2302 |
+
batch_normalize=1
|
2303 |
+
size=3
|
2304 |
+
stride=1
|
2305 |
+
pad=1
|
2306 |
+
filters=640
|
2307 |
+
activation=mish
|
2308 |
+
|
2309 |
+
[route]
|
2310 |
+
layers = -1,-8
|
2311 |
+
|
2312 |
+
# Transition last
|
2313 |
+
|
2314 |
+
# 319 (previous+3+4+2k)
|
2315 |
+
[convolutional]
|
2316 |
+
batch_normalize=1
|
2317 |
+
filters=640
|
2318 |
+
size=1
|
2319 |
+
stride=1
|
2320 |
+
pad=1
|
2321 |
+
activation=mish
|
2322 |
+
|
2323 |
+
|
2324 |
+
# PAN-6
|
2325 |
+
|
2326 |
+
[convolutional]
|
2327 |
+
batch_normalize=1
|
2328 |
+
size=3
|
2329 |
+
stride=2
|
2330 |
+
pad=1
|
2331 |
+
filters=640
|
2332 |
+
activation=mish
|
2333 |
+
|
2334 |
+
[route]
|
2335 |
+
layers = -1, 245
|
2336 |
+
|
2337 |
+
[convolutional]
|
2338 |
+
batch_normalize=1
|
2339 |
+
filters=640
|
2340 |
+
size=1
|
2341 |
+
stride=1
|
2342 |
+
pad=1
|
2343 |
+
activation=mish
|
2344 |
+
|
2345 |
+
# Split
|
2346 |
+
|
2347 |
+
[convolutional]
|
2348 |
+
batch_normalize=1
|
2349 |
+
filters=640
|
2350 |
+
size=1
|
2351 |
+
stride=1
|
2352 |
+
pad=1
|
2353 |
+
activation=mish
|
2354 |
+
|
2355 |
+
[route]
|
2356 |
+
layers = -2
|
2357 |
+
|
2358 |
+
# Plain Block
|
2359 |
+
|
2360 |
+
[convolutional]
|
2361 |
+
batch_normalize=1
|
2362 |
+
filters=640
|
2363 |
+
size=1
|
2364 |
+
stride=1
|
2365 |
+
pad=1
|
2366 |
+
activation=mish
|
2367 |
+
|
2368 |
+
[convolutional]
|
2369 |
+
batch_normalize=1
|
2370 |
+
size=3
|
2371 |
+
stride=1
|
2372 |
+
pad=1
|
2373 |
+
filters=640
|
2374 |
+
activation=mish
|
2375 |
+
|
2376 |
+
[convolutional]
|
2377 |
+
batch_normalize=1
|
2378 |
+
filters=640
|
2379 |
+
size=1
|
2380 |
+
stride=1
|
2381 |
+
pad=1
|
2382 |
+
activation=mish
|
2383 |
+
|
2384 |
+
[convolutional]
|
2385 |
+
batch_normalize=1
|
2386 |
+
size=3
|
2387 |
+
stride=1
|
2388 |
+
pad=1
|
2389 |
+
filters=640
|
2390 |
+
activation=mish
|
2391 |
+
|
2392 |
+
[convolutional]
|
2393 |
+
batch_normalize=1
|
2394 |
+
filters=640
|
2395 |
+
size=1
|
2396 |
+
stride=1
|
2397 |
+
pad=1
|
2398 |
+
activation=mish
|
2399 |
+
|
2400 |
+
[convolutional]
|
2401 |
+
batch_normalize=1
|
2402 |
+
size=3
|
2403 |
+
stride=1
|
2404 |
+
pad=1
|
2405 |
+
filters=640
|
2406 |
+
activation=mish
|
2407 |
+
|
2408 |
+
[route]
|
2409 |
+
layers = -1,-8
|
2410 |
+
|
2411 |
+
# Transition last
|
2412 |
+
|
2413 |
+
# 332 (previous+3+4+2k)
|
2414 |
+
[convolutional]
|
2415 |
+
batch_normalize=1
|
2416 |
+
filters=640
|
2417 |
+
size=1
|
2418 |
+
stride=1
|
2419 |
+
pad=1
|
2420 |
+
activation=mish
|
2421 |
+
|
2422 |
+
|
2423 |
+
# PAN-7
|
2424 |
+
|
2425 |
+
[convolutional]
|
2426 |
+
batch_normalize=1
|
2427 |
+
size=3
|
2428 |
+
stride=2
|
2429 |
+
pad=1
|
2430 |
+
filters=640
|
2431 |
+
activation=mish
|
2432 |
+
|
2433 |
+
[route]
|
2434 |
+
layers = -1, 229
|
2435 |
+
|
2436 |
+
[convolutional]
|
2437 |
+
batch_normalize=1
|
2438 |
+
filters=640
|
2439 |
+
size=1
|
2440 |
+
stride=1
|
2441 |
+
pad=1
|
2442 |
+
activation=mish
|
2443 |
+
|
2444 |
+
# Split
|
2445 |
+
|
2446 |
+
[convolutional]
|
2447 |
+
batch_normalize=1
|
2448 |
+
filters=640
|
2449 |
+
size=1
|
2450 |
+
stride=1
|
2451 |
+
pad=1
|
2452 |
+
activation=mish
|
2453 |
+
|
2454 |
+
[route]
|
2455 |
+
layers = -2
|
2456 |
+
|
2457 |
+
# Plain Block
|
2458 |
+
|
2459 |
+
[convolutional]
|
2460 |
+
batch_normalize=1
|
2461 |
+
filters=640
|
2462 |
+
size=1
|
2463 |
+
stride=1
|
2464 |
+
pad=1
|
2465 |
+
activation=mish
|
2466 |
+
|
2467 |
+
[convolutional]
|
2468 |
+
batch_normalize=1
|
2469 |
+
size=3
|
2470 |
+
stride=1
|
2471 |
+
pad=1
|
2472 |
+
filters=640
|
2473 |
+
activation=mish
|
2474 |
+
|
2475 |
+
[convolutional]
|
2476 |
+
batch_normalize=1
|
2477 |
+
filters=640
|
2478 |
+
size=1
|
2479 |
+
stride=1
|
2480 |
+
pad=1
|
2481 |
+
activation=mish
|
2482 |
+
|
2483 |
+
[convolutional]
|
2484 |
+
batch_normalize=1
|
2485 |
+
size=3
|
2486 |
+
stride=1
|
2487 |
+
pad=1
|
2488 |
+
filters=640
|
2489 |
+
activation=mish
|
2490 |
+
|
2491 |
+
[convolutional]
|
2492 |
+
batch_normalize=1
|
2493 |
+
filters=640
|
2494 |
+
size=1
|
2495 |
+
stride=1
|
2496 |
+
pad=1
|
2497 |
+
activation=mish
|
2498 |
+
|
2499 |
+
[convolutional]
|
2500 |
+
batch_normalize=1
|
2501 |
+
size=3
|
2502 |
+
stride=1
|
2503 |
+
pad=1
|
2504 |
+
filters=640
|
2505 |
+
activation=mish
|
2506 |
+
|
2507 |
+
[route]
|
2508 |
+
layers = -1,-8
|
2509 |
+
|
2510 |
+
# Transition last
|
2511 |
+
|
2512 |
+
# 345 (previous+3+4+2k)
|
2513 |
+
[convolutional]
|
2514 |
+
batch_normalize=1
|
2515 |
+
filters=640
|
2516 |
+
size=1
|
2517 |
+
stride=1
|
2518 |
+
pad=1
|
2519 |
+
activation=mish
|
2520 |
+
|
2521 |
+
# ============ End of Neck ============ #
|
2522 |
+
|
2523 |
+
# ============ Head ============ #
|
2524 |
+
|
2525 |
+
# YOLO-3
|
2526 |
+
|
2527 |
+
[route]
|
2528 |
+
layers = 293
|
2529 |
+
|
2530 |
+
[convolutional]
|
2531 |
+
batch_normalize=1
|
2532 |
+
size=3
|
2533 |
+
stride=1
|
2534 |
+
pad=1
|
2535 |
+
filters=320
|
2536 |
+
activation=mish
|
2537 |
+
|
2538 |
+
[convolutional]
|
2539 |
+
size=1
|
2540 |
+
stride=1
|
2541 |
+
pad=1
|
2542 |
+
filters=340
|
2543 |
+
activation=linear
|
2544 |
+
|
2545 |
+
[yolo]
|
2546 |
+
mask = 0,1,2,3
|
2547 |
+
anchors = 13,17, 22,25, 27,66, 55,41, 57,88, 112,69, 69,177, 136,138, 136,138, 287,114, 134,275, 268,248, 268,248, 232,504, 445,416, 640,640, 812,393, 477,808, 1070,908, 1408,1408
|
2548 |
+
classes=80
|
2549 |
+
num=20
|
2550 |
+
jitter=.3
|
2551 |
+
ignore_thresh = .7
|
2552 |
+
truth_thresh = 1
|
2553 |
+
random=1
|
2554 |
+
scale_x_y = 1.05
|
2555 |
+
iou_thresh=0.213
|
2556 |
+
cls_normalizer=1.0
|
2557 |
+
iou_normalizer=0.07
|
2558 |
+
iou_loss=ciou
|
2559 |
+
nms_kind=greedynms
|
2560 |
+
beta_nms=0.6
|
2561 |
+
|
2562 |
+
|
2563 |
+
# YOLO-4
|
2564 |
+
|
2565 |
+
[route]
|
2566 |
+
layers = 306
|
2567 |
+
|
2568 |
+
[convolutional]
|
2569 |
+
batch_normalize=1
|
2570 |
+
size=3
|
2571 |
+
stride=1
|
2572 |
+
pad=1
|
2573 |
+
filters=640
|
2574 |
+
activation=mish
|
2575 |
+
|
2576 |
+
[convolutional]
|
2577 |
+
size=1
|
2578 |
+
stride=1
|
2579 |
+
pad=1
|
2580 |
+
filters=340
|
2581 |
+
activation=linear
|
2582 |
+
|
2583 |
+
[yolo]
|
2584 |
+
mask = 4,5,6,7
|
2585 |
+
anchors = 13,17, 22,25, 27,66, 55,41, 57,88, 112,69, 69,177, 136,138, 136,138, 287,114, 134,275, 268,248, 268,248, 232,504, 445,416, 640,640, 812,393, 477,808, 1070,908, 1408,1408
|
2586 |
+
classes=80
|
2587 |
+
num=20
|
2588 |
+
jitter=.3
|
2589 |
+
ignore_thresh = .7
|
2590 |
+
truth_thresh = 1
|
2591 |
+
random=1
|
2592 |
+
scale_x_y = 1.05
|
2593 |
+
iou_thresh=0.213
|
2594 |
+
cls_normalizer=1.0
|
2595 |
+
iou_normalizer=0.07
|
2596 |
+
iou_loss=ciou
|
2597 |
+
nms_kind=greedynms
|
2598 |
+
beta_nms=0.6
|
2599 |
+
|
2600 |
+
|
2601 |
+
# YOLO-5
|
2602 |
+
|
2603 |
+
[route]
|
2604 |
+
layers = 319
|
2605 |
+
|
2606 |
+
[convolutional]
|
2607 |
+
batch_normalize=1
|
2608 |
+
size=3
|
2609 |
+
stride=1
|
2610 |
+
pad=1
|
2611 |
+
filters=1280
|
2612 |
+
activation=mish
|
2613 |
+
|
2614 |
+
[convolutional]
|
2615 |
+
size=1
|
2616 |
+
stride=1
|
2617 |
+
pad=1
|
2618 |
+
filters=340
|
2619 |
+
activation=linear
|
2620 |
+
|
2621 |
+
[yolo]
|
2622 |
+
mask = 8,9,10,11
|
2623 |
+
anchors = 13,17, 22,25, 27,66, 55,41, 57,88, 112,69, 69,177, 136,138, 136,138, 287,114, 134,275, 268,248, 268,248, 232,504, 445,416, 640,640, 812,393, 477,808, 1070,908, 1408,1408
|
2624 |
+
classes=80
|
2625 |
+
num=20
|
2626 |
+
jitter=.3
|
2627 |
+
ignore_thresh = .7
|
2628 |
+
truth_thresh = 1
|
2629 |
+
random=1
|
2630 |
+
scale_x_y = 1.05
|
2631 |
+
iou_thresh=0.213
|
2632 |
+
cls_normalizer=1.0
|
2633 |
+
iou_normalizer=0.07
|
2634 |
+
iou_loss=ciou
|
2635 |
+
nms_kind=greedynms
|
2636 |
+
beta_nms=0.6
|
2637 |
+
|
2638 |
+
|
2639 |
+
# YOLO-6
|
2640 |
+
|
2641 |
+
[route]
|
2642 |
+
layers = 332
|
2643 |
+
|
2644 |
+
[convolutional]
|
2645 |
+
batch_normalize=1
|
2646 |
+
size=3
|
2647 |
+
stride=1
|
2648 |
+
pad=1
|
2649 |
+
filters=1280
|
2650 |
+
activation=mish
|
2651 |
+
|
2652 |
+
[convolutional]
|
2653 |
+
size=1
|
2654 |
+
stride=1
|
2655 |
+
pad=1
|
2656 |
+
filters=340
|
2657 |
+
activation=linear
|
2658 |
+
|
2659 |
+
[yolo]
|
2660 |
+
mask = 12,13,14,15
|
2661 |
+
anchors = 13,17, 22,25, 27,66, 55,41, 57,88, 112,69, 69,177, 136,138, 136,138, 287,114, 134,275, 268,248, 268,248, 232,504, 445,416, 640,640, 812,393, 477,808, 1070,908, 1408,1408
|
2662 |
+
classes=80
|
2663 |
+
num=20
|
2664 |
+
jitter=.3
|
2665 |
+
ignore_thresh = .7
|
2666 |
+
truth_thresh = 1
|
2667 |
+
random=1
|
2668 |
+
scale_x_y = 1.05
|
2669 |
+
iou_thresh=0.213
|
2670 |
+
cls_normalizer=1.0
|
2671 |
+
iou_normalizer=0.07
|
2672 |
+
iou_loss=ciou
|
2673 |
+
nms_kind=greedynms
|
2674 |
+
beta_nms=0.6
|
2675 |
+
|
2676 |
+
|
2677 |
+
# YOLO-7
|
2678 |
+
|
2679 |
+
[route]
|
2680 |
+
layers = 345
|
2681 |
+
|
2682 |
+
[convolutional]
|
2683 |
+
batch_normalize=1
|
2684 |
+
size=3
|
2685 |
+
stride=1
|
2686 |
+
pad=1
|
2687 |
+
filters=1280
|
2688 |
+
activation=mish
|
2689 |
+
|
2690 |
+
[convolutional]
|
2691 |
+
size=1
|
2692 |
+
stride=1
|
2693 |
+
pad=1
|
2694 |
+
filters=340
|
2695 |
+
activation=linear
|
2696 |
+
|
2697 |
+
[yolo]
|
2698 |
+
mask = 16,17,18,19
|
2699 |
+
anchors = 13,17, 22,25, 27,66, 55,41, 57,88, 112,69, 69,177, 136,138, 136,138, 287,114, 134,275, 268,248, 268,248, 232,504, 445,416, 640,640, 812,393, 477,808, 1070,908, 1408,1408
|
2700 |
+
classes=80
|
2701 |
+
num=20
|
2702 |
+
jitter=.3
|
2703 |
+
ignore_thresh = .7
|
2704 |
+
truth_thresh = 1
|
2705 |
+
random=1
|
2706 |
+
scale_x_y = 1.05
|
2707 |
+
iou_thresh=0.213
|
2708 |
+
cls_normalizer=1.0
|
2709 |
+
iou_normalizer=0.07
|
2710 |
+
iou_loss=ciou
|
2711 |
+
nms_kind=greedynms
|
2712 |
+
beta_nms=0.6
|
2713 |
+
|
2714 |
+
# ============ End of Head ============ #
|
darknet/README.md
ADDED
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
## Model Zoo
|
2 |
+
|
3 |
+
| Model | Test Size | AP<sup>val</sup> | AP<sub>50</sub><sup>val</sup> | AP<sub>75</sub><sup>val</sup> | AP<sub>S</sub><sup>val</sup> | AP<sub>M</sub><sup>val</sup> | AP<sub>L</sub><sup>val</sup> | batch1 throughput |
|
4 |
+
| :-- | :-: | :-: | :-: | :-: | :-: | :-: | :-: | :-: |
|
5 |
+
| **YOLOv4-CSP** | 640 | **49.1%** | **67.7%** | **53.8%** | **32.1%** | **54.4%** | **63.2%** | 76 *fps* |
|
6 |
+
| **YOLOR-CSP** | 640 | **49.2%** | **67.6%** | **53.7%** | **32.9%** | **54.4%** | **63.0%** | - |
|
7 |
+
| | | | | | | |
|
8 |
+
| **YOLOv4-CSP-X** | 640 | **50.9%** | **69.3%** | **55.4%** | **35.3%** | **55.8%** | **64.8%** | 53 *fps* |
|
9 |
+
| **YOLOR-CSP-X** | 640 | **51.1%** | **69.6%** | **55.7%** | **35.7%** | **56.0%** | **65.2%** | - |
|
10 |
+
| | | | | | | |
|
11 |
+
|
12 |
+
## Installation
|
13 |
+
|
14 |
+
https://github.com/AlexeyAB/darknet
|
15 |
+
|
16 |
+
Docker environment (recommended)
|
17 |
+
<details><summary> <b>Expand</b> </summary>
|
18 |
+
|
19 |
+
```
|
20 |
+
# get code
|
21 |
+
git clone https://github.com/AlexeyAB/darknet
|
22 |
+
|
23 |
+
# create the docker container, you can change the share memory size if you have more.
|
24 |
+
nvidia-docker run --name yolor -it -v your_coco_path/:/coco/ -v your_code_path/:/yolor --shm-size=64g nvcr.io/nvidia/pytorch:21.02-py3
|
25 |
+
|
26 |
+
# apt install required packages
|
27 |
+
apt update
|
28 |
+
apt install -y libopencv-dev
|
29 |
+
|
30 |
+
# edit Makefile
|
31 |
+
#GPU=1
|
32 |
+
#CUDNN=1
|
33 |
+
#CUDNN_HALF=1
|
34 |
+
#OPENCV=1
|
35 |
+
#AVX=1
|
36 |
+
#OPENMP=1
|
37 |
+
#LIBSO=1
|
38 |
+
#ZED_CAMERA=0
|
39 |
+
#ZED_CAMERA_v2_8=0
|
40 |
+
#
|
41 |
+
#USE_CPP=0
|
42 |
+
#DEBUG=0
|
43 |
+
#
|
44 |
+
#ARCH= -gencode arch=compute_52,code=[sm_70,compute_70] \
|
45 |
+
# -gencode arch=compute_61,code=[sm_75,compute_75] \
|
46 |
+
# -gencode arch=compute_61,code=[sm_80,compute_80] \
|
47 |
+
# -gencode arch=compute_61,code=[sm_86,compute_86]
|
48 |
+
#
|
49 |
+
#...
|
50 |
+
|
51 |
+
# build
|
52 |
+
make -j8
|
53 |
+
```
|
54 |
+
|
55 |
+
</details>
|
56 |
+
|
57 |
+
## Testing
|
58 |
+
|
59 |
+
To reproduce inference speed, using:
|
60 |
+
|
61 |
+
```
|
62 |
+
CUDA_VISIBLE_DEVICES=0 ./darknet detector demo cfg/coco.data cfg/yolov4-csp.cfg weights/yolov4-csp.weights source/test.mp4 -dont_show -benchmark
|
63 |
+
```
|
darknet/cfg/yolov4-csp-x.cfg
ADDED
@@ -0,0 +1,1555 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
[net]
|
3 |
+
# Testing
|
4 |
+
#batch=1
|
5 |
+
#subdivisions=1
|
6 |
+
# Training
|
7 |
+
batch=64
|
8 |
+
subdivisions=8
|
9 |
+
width=640
|
10 |
+
height=640
|
11 |
+
channels=3
|
12 |
+
momentum=0.949
|
13 |
+
decay=0.0005
|
14 |
+
angle=0
|
15 |
+
saturation = 1.5
|
16 |
+
exposure = 1.5
|
17 |
+
hue=.1
|
18 |
+
|
19 |
+
learning_rate=0.001
|
20 |
+
burn_in=1000
|
21 |
+
max_batches = 500500
|
22 |
+
policy=steps
|
23 |
+
steps=400000,450000
|
24 |
+
scales=.1,.1
|
25 |
+
|
26 |
+
mosaic=1
|
27 |
+
|
28 |
+
letter_box=1
|
29 |
+
|
30 |
+
ema_alpha=0.9998
|
31 |
+
|
32 |
+
#optimized_memory=1
|
33 |
+
|
34 |
+
|
35 |
+
# ============ Backbone ============ #
|
36 |
+
|
37 |
+
# Stem
|
38 |
+
|
39 |
+
# 0
|
40 |
+
[convolutional]
|
41 |
+
batch_normalize=1
|
42 |
+
filters=32
|
43 |
+
size=3
|
44 |
+
stride=1
|
45 |
+
pad=1
|
46 |
+
activation=swish
|
47 |
+
|
48 |
+
# P1
|
49 |
+
|
50 |
+
# Downsample
|
51 |
+
|
52 |
+
[convolutional]
|
53 |
+
batch_normalize=1
|
54 |
+
filters=80
|
55 |
+
size=3
|
56 |
+
stride=2
|
57 |
+
pad=1
|
58 |
+
activation=swish
|
59 |
+
|
60 |
+
# Residual Block
|
61 |
+
|
62 |
+
[convolutional]
|
63 |
+
batch_normalize=1
|
64 |
+
filters=40
|
65 |
+
size=1
|
66 |
+
stride=1
|
67 |
+
pad=1
|
68 |
+
activation=swish
|
69 |
+
|
70 |
+
[convolutional]
|
71 |
+
batch_normalize=1
|
72 |
+
filters=80
|
73 |
+
size=3
|
74 |
+
stride=1
|
75 |
+
pad=1
|
76 |
+
activation=swish
|
77 |
+
|
78 |
+
# 4 (previous+1+3k)
|
79 |
+
[shortcut]
|
80 |
+
from=-3
|
81 |
+
activation=linear
|
82 |
+
|
83 |
+
# P2
|
84 |
+
|
85 |
+
# Downsample
|
86 |
+
|
87 |
+
[convolutional]
|
88 |
+
batch_normalize=1
|
89 |
+
filters=160
|
90 |
+
size=3
|
91 |
+
stride=2
|
92 |
+
pad=1
|
93 |
+
activation=swish
|
94 |
+
|
95 |
+
# Split
|
96 |
+
|
97 |
+
[convolutional]
|
98 |
+
batch_normalize=1
|
99 |
+
filters=80
|
100 |
+
size=1
|
101 |
+
stride=1
|
102 |
+
pad=1
|
103 |
+
activation=swish
|
104 |
+
|
105 |
+
[route]
|
106 |
+
layers = -2
|
107 |
+
|
108 |
+
[convolutional]
|
109 |
+
batch_normalize=1
|
110 |
+
filters=80
|
111 |
+
size=1
|
112 |
+
stride=1
|
113 |
+
pad=1
|
114 |
+
activation=swish
|
115 |
+
|
116 |
+
# Residual Block
|
117 |
+
|
118 |
+
[convolutional]
|
119 |
+
batch_normalize=1
|
120 |
+
filters=80
|
121 |
+
size=1
|
122 |
+
stride=1
|
123 |
+
pad=1
|
124 |
+
activation=swish
|
125 |
+
|
126 |
+
[convolutional]
|
127 |
+
batch_normalize=1
|
128 |
+
filters=80
|
129 |
+
size=3
|
130 |
+
stride=1
|
131 |
+
pad=1
|
132 |
+
activation=swish
|
133 |
+
|
134 |
+
[shortcut]
|
135 |
+
from=-3
|
136 |
+
activation=linear
|
137 |
+
|
138 |
+
[convolutional]
|
139 |
+
batch_normalize=1
|
140 |
+
filters=80
|
141 |
+
size=1
|
142 |
+
stride=1
|
143 |
+
pad=1
|
144 |
+
activation=swish
|
145 |
+
|
146 |
+
[convolutional]
|
147 |
+
batch_normalize=1
|
148 |
+
filters=80
|
149 |
+
size=3
|
150 |
+
stride=1
|
151 |
+
pad=1
|
152 |
+
activation=swish
|
153 |
+
|
154 |
+
[shortcut]
|
155 |
+
from=-3
|
156 |
+
activation=linear
|
157 |
+
|
158 |
+
[convolutional]
|
159 |
+
batch_normalize=1
|
160 |
+
filters=80
|
161 |
+
size=1
|
162 |
+
stride=1
|
163 |
+
pad=1
|
164 |
+
activation=swish
|
165 |
+
|
166 |
+
[convolutional]
|
167 |
+
batch_normalize=1
|
168 |
+
filters=80
|
169 |
+
size=3
|
170 |
+
stride=1
|
171 |
+
pad=1
|
172 |
+
activation=swish
|
173 |
+
|
174 |
+
[shortcut]
|
175 |
+
from=-3
|
176 |
+
activation=linear
|
177 |
+
|
178 |
+
# Transition first
|
179 |
+
|
180 |
+
[convolutional]
|
181 |
+
batch_normalize=1
|
182 |
+
filters=80
|
183 |
+
size=1
|
184 |
+
stride=1
|
185 |
+
pad=1
|
186 |
+
activation=swish
|
187 |
+
|
188 |
+
# Merge [-1, -(3k+4)]
|
189 |
+
|
190 |
+
[route]
|
191 |
+
layers = -1,-13
|
192 |
+
|
193 |
+
# Transition last
|
194 |
+
|
195 |
+
# 20 (previous+7+3k)
|
196 |
+
[convolutional]
|
197 |
+
batch_normalize=1
|
198 |
+
filters=160
|
199 |
+
size=1
|
200 |
+
stride=1
|
201 |
+
pad=1
|
202 |
+
activation=swish
|
203 |
+
|
204 |
+
# P3
|
205 |
+
|
206 |
+
# Downsample
|
207 |
+
|
208 |
+
[convolutional]
|
209 |
+
batch_normalize=1
|
210 |
+
filters=320
|
211 |
+
size=3
|
212 |
+
stride=2
|
213 |
+
pad=1
|
214 |
+
activation=swish
|
215 |
+
|
216 |
+
# Split
|
217 |
+
|
218 |
+
[convolutional]
|
219 |
+
batch_normalize=1
|
220 |
+
filters=160
|
221 |
+
size=1
|
222 |
+
stride=1
|
223 |
+
pad=1
|
224 |
+
activation=swish
|
225 |
+
|
226 |
+
[route]
|
227 |
+
layers = -2
|
228 |
+
|
229 |
+
[convolutional]
|
230 |
+
batch_normalize=1
|
231 |
+
filters=160
|
232 |
+
size=1
|
233 |
+
stride=1
|
234 |
+
pad=1
|
235 |
+
activation=swish
|
236 |
+
|
237 |
+
# Residual Block
|
238 |
+
|
239 |
+
[convolutional]
|
240 |
+
batch_normalize=1
|
241 |
+
filters=160
|
242 |
+
size=1
|
243 |
+
stride=1
|
244 |
+
pad=1
|
245 |
+
activation=swish
|
246 |
+
|
247 |
+
[convolutional]
|
248 |
+
batch_normalize=1
|
249 |
+
filters=160
|
250 |
+
size=3
|
251 |
+
stride=1
|
252 |
+
pad=1
|
253 |
+
activation=swish
|
254 |
+
|
255 |
+
[shortcut]
|
256 |
+
from=-3
|
257 |
+
activation=linear
|
258 |
+
|
259 |
+
[convolutional]
|
260 |
+
batch_normalize=1
|
261 |
+
filters=160
|
262 |
+
size=1
|
263 |
+
stride=1
|
264 |
+
pad=1
|
265 |
+
activation=swish
|
266 |
+
|
267 |
+
[convolutional]
|
268 |
+
batch_normalize=1
|
269 |
+
filters=160
|
270 |
+
size=3
|
271 |
+
stride=1
|
272 |
+
pad=1
|
273 |
+
activation=swish
|
274 |
+
|
275 |
+
[shortcut]
|
276 |
+
from=-3
|
277 |
+
activation=linear
|
278 |
+
|
279 |
+
[convolutional]
|
280 |
+
batch_normalize=1
|
281 |
+
filters=160
|
282 |
+
size=1
|
283 |
+
stride=1
|
284 |
+
pad=1
|
285 |
+
activation=swish
|
286 |
+
|
287 |
+
[convolutional]
|
288 |
+
batch_normalize=1
|
289 |
+
filters=160
|
290 |
+
size=3
|
291 |
+
stride=1
|
292 |
+
pad=1
|
293 |
+
activation=swish
|
294 |
+
|
295 |
+
[shortcut]
|
296 |
+
from=-3
|
297 |
+
activation=linear
|
298 |
+
|
299 |
+
[convolutional]
|
300 |
+
batch_normalize=1
|
301 |
+
filters=160
|
302 |
+
size=1
|
303 |
+
stride=1
|
304 |
+
pad=1
|
305 |
+
activation=swish
|
306 |
+
|
307 |
+
[convolutional]
|
308 |
+
batch_normalize=1
|
309 |
+
filters=160
|
310 |
+
size=3
|
311 |
+
stride=1
|
312 |
+
pad=1
|
313 |
+
activation=swish
|
314 |
+
|
315 |
+
[shortcut]
|
316 |
+
from=-3
|
317 |
+
activation=linear
|
318 |
+
|
319 |
+
[convolutional]
|
320 |
+
batch_normalize=1
|
321 |
+
filters=160
|
322 |
+
size=1
|
323 |
+
stride=1
|
324 |
+
pad=1
|
325 |
+
activation=swish
|
326 |
+
|
327 |
+
[convolutional]
|
328 |
+
batch_normalize=1
|
329 |
+
filters=160
|
330 |
+
size=3
|
331 |
+
stride=1
|
332 |
+
pad=1
|
333 |
+
activation=swish
|
334 |
+
|
335 |
+
[shortcut]
|
336 |
+
from=-3
|
337 |
+
activation=linear
|
338 |
+
|
339 |
+
[convolutional]
|
340 |
+
batch_normalize=1
|
341 |
+
filters=160
|
342 |
+
size=1
|
343 |
+
stride=1
|
344 |
+
pad=1
|
345 |
+
activation=swish
|
346 |
+
|
347 |
+
[convolutional]
|
348 |
+
batch_normalize=1
|
349 |
+
filters=160
|
350 |
+
size=3
|
351 |
+
stride=1
|
352 |
+
pad=1
|
353 |
+
activation=swish
|
354 |
+
|
355 |
+
[shortcut]
|
356 |
+
from=-3
|
357 |
+
activation=linear
|
358 |
+
|
359 |
+
[convolutional]
|
360 |
+
batch_normalize=1
|
361 |
+
filters=160
|
362 |
+
size=1
|
363 |
+
stride=1
|
364 |
+
pad=1
|
365 |
+
activation=swish
|
366 |
+
|
367 |
+
[convolutional]
|
368 |
+
batch_normalize=1
|
369 |
+
filters=160
|
370 |
+
size=3
|
371 |
+
stride=1
|
372 |
+
pad=1
|
373 |
+
activation=swish
|
374 |
+
|
375 |
+
[shortcut]
|
376 |
+
from=-3
|
377 |
+
activation=linear
|
378 |
+
|
379 |
+
[convolutional]
|
380 |
+
batch_normalize=1
|
381 |
+
filters=160
|
382 |
+
size=1
|
383 |
+
stride=1
|
384 |
+
pad=1
|
385 |
+
activation=swish
|
386 |
+
|
387 |
+
[convolutional]
|
388 |
+
batch_normalize=1
|
389 |
+
filters=160
|
390 |
+
size=3
|
391 |
+
stride=1
|
392 |
+
pad=1
|
393 |
+
activation=swish
|
394 |
+
|
395 |
+
[shortcut]
|
396 |
+
from=-3
|
397 |
+
activation=linear
|
398 |
+
|
399 |
+
[convolutional]
|
400 |
+
batch_normalize=1
|
401 |
+
filters=160
|
402 |
+
size=1
|
403 |
+
stride=1
|
404 |
+
pad=1
|
405 |
+
activation=swish
|
406 |
+
|
407 |
+
[convolutional]
|
408 |
+
batch_normalize=1
|
409 |
+
filters=160
|
410 |
+
size=3
|
411 |
+
stride=1
|
412 |
+
pad=1
|
413 |
+
activation=swish
|
414 |
+
|
415 |
+
[shortcut]
|
416 |
+
from=-3
|
417 |
+
activation=linear
|
418 |
+
|
419 |
+
[convolutional]
|
420 |
+
batch_normalize=1
|
421 |
+
filters=160
|
422 |
+
size=1
|
423 |
+
stride=1
|
424 |
+
pad=1
|
425 |
+
activation=swish
|
426 |
+
|
427 |
+
[convolutional]
|
428 |
+
batch_normalize=1
|
429 |
+
filters=160
|
430 |
+
size=3
|
431 |
+
stride=1
|
432 |
+
pad=1
|
433 |
+
activation=swish
|
434 |
+
|
435 |
+
[shortcut]
|
436 |
+
from=-3
|
437 |
+
activation=linear
|
438 |
+
|
439 |
+
# Transition first
|
440 |
+
|
441 |
+
[convolutional]
|
442 |
+
batch_normalize=1
|
443 |
+
filters=160
|
444 |
+
size=1
|
445 |
+
stride=1
|
446 |
+
pad=1
|
447 |
+
activation=swish
|
448 |
+
|
449 |
+
# Merge [-1 -(4+3k)]
|
450 |
+
|
451 |
+
[route]
|
452 |
+
layers = -1,-34
|
453 |
+
|
454 |
+
# Transition last
|
455 |
+
|
456 |
+
# 57 (previous+7+3k)
|
457 |
+
[convolutional]
|
458 |
+
batch_normalize=1
|
459 |
+
filters=320
|
460 |
+
size=1
|
461 |
+
stride=1
|
462 |
+
pad=1
|
463 |
+
activation=swish
|
464 |
+
|
465 |
+
# P4
|
466 |
+
|
467 |
+
# Downsample
|
468 |
+
|
469 |
+
[convolutional]
|
470 |
+
batch_normalize=1
|
471 |
+
filters=640
|
472 |
+
size=3
|
473 |
+
stride=2
|
474 |
+
pad=1
|
475 |
+
activation=swish
|
476 |
+
|
477 |
+
# Split
|
478 |
+
|
479 |
+
[convolutional]
|
480 |
+
batch_normalize=1
|
481 |
+
filters=320
|
482 |
+
size=1
|
483 |
+
stride=1
|
484 |
+
pad=1
|
485 |
+
activation=swish
|
486 |
+
|
487 |
+
[route]
|
488 |
+
layers = -2
|
489 |
+
|
490 |
+
[convolutional]
|
491 |
+
batch_normalize=1
|
492 |
+
filters=320
|
493 |
+
size=1
|
494 |
+
stride=1
|
495 |
+
pad=1
|
496 |
+
activation=swish
|
497 |
+
|
498 |
+
# Residual Block
|
499 |
+
|
500 |
+
[convolutional]
|
501 |
+
batch_normalize=1
|
502 |
+
filters=320
|
503 |
+
size=1
|
504 |
+
stride=1
|
505 |
+
pad=1
|
506 |
+
activation=swish
|
507 |
+
|
508 |
+
[convolutional]
|
509 |
+
batch_normalize=1
|
510 |
+
filters=320
|
511 |
+
size=3
|
512 |
+
stride=1
|
513 |
+
pad=1
|
514 |
+
activation=swish
|
515 |
+
|
516 |
+
[shortcut]
|
517 |
+
from=-3
|
518 |
+
activation=linear
|
519 |
+
|
520 |
+
[convolutional]
|
521 |
+
batch_normalize=1
|
522 |
+
filters=320
|
523 |
+
size=1
|
524 |
+
stride=1
|
525 |
+
pad=1
|
526 |
+
activation=swish
|
527 |
+
|
528 |
+
[convolutional]
|
529 |
+
batch_normalize=1
|
530 |
+
filters=320
|
531 |
+
size=3
|
532 |
+
stride=1
|
533 |
+
pad=1
|
534 |
+
activation=swish
|
535 |
+
|
536 |
+
[shortcut]
|
537 |
+
from=-3
|
538 |
+
activation=linear
|
539 |
+
|
540 |
+
[convolutional]
|
541 |
+
batch_normalize=1
|
542 |
+
filters=320
|
543 |
+
size=1
|
544 |
+
stride=1
|
545 |
+
pad=1
|
546 |
+
activation=swish
|
547 |
+
|
548 |
+
[convolutional]
|
549 |
+
batch_normalize=1
|
550 |
+
filters=320
|
551 |
+
size=3
|
552 |
+
stride=1
|
553 |
+
pad=1
|
554 |
+
activation=swish
|
555 |
+
|
556 |
+
[shortcut]
|
557 |
+
from=-3
|
558 |
+
activation=linear
|
559 |
+
|
560 |
+
[convolutional]
|
561 |
+
batch_normalize=1
|
562 |
+
filters=320
|
563 |
+
size=1
|
564 |
+
stride=1
|
565 |
+
pad=1
|
566 |
+
activation=swish
|
567 |
+
|
568 |
+
[convolutional]
|
569 |
+
batch_normalize=1
|
570 |
+
filters=320
|
571 |
+
size=3
|
572 |
+
stride=1
|
573 |
+
pad=1
|
574 |
+
activation=swish
|
575 |
+
|
576 |
+
[shortcut]
|
577 |
+
from=-3
|
578 |
+
activation=linear
|
579 |
+
|
580 |
+
[convolutional]
|
581 |
+
batch_normalize=1
|
582 |
+
filters=320
|
583 |
+
size=1
|
584 |
+
stride=1
|
585 |
+
pad=1
|
586 |
+
activation=swish
|
587 |
+
|
588 |
+
[convolutional]
|
589 |
+
batch_normalize=1
|
590 |
+
filters=320
|
591 |
+
size=3
|
592 |
+
stride=1
|
593 |
+
pad=1
|
594 |
+
activation=swish
|
595 |
+
|
596 |
+
[shortcut]
|
597 |
+
from=-3
|
598 |
+
activation=linear
|
599 |
+
|
600 |
+
[convolutional]
|
601 |
+
batch_normalize=1
|
602 |
+
filters=320
|
603 |
+
size=1
|
604 |
+
stride=1
|
605 |
+
pad=1
|
606 |
+
activation=swish
|
607 |
+
|
608 |
+
[convolutional]
|
609 |
+
batch_normalize=1
|
610 |
+
filters=320
|
611 |
+
size=3
|
612 |
+
stride=1
|
613 |
+
pad=1
|
614 |
+
activation=swish
|
615 |
+
|
616 |
+
[shortcut]
|
617 |
+
from=-3
|
618 |
+
activation=linear
|
619 |
+
|
620 |
+
[convolutional]
|
621 |
+
batch_normalize=1
|
622 |
+
filters=320
|
623 |
+
size=1
|
624 |
+
stride=1
|
625 |
+
pad=1
|
626 |
+
activation=swish
|
627 |
+
|
628 |
+
[convolutional]
|
629 |
+
batch_normalize=1
|
630 |
+
filters=320
|
631 |
+
size=3
|
632 |
+
stride=1
|
633 |
+
pad=1
|
634 |
+
activation=swish
|
635 |
+
|
636 |
+
[shortcut]
|
637 |
+
from=-3
|
638 |
+
activation=linear
|
639 |
+
|
640 |
+
[convolutional]
|
641 |
+
batch_normalize=1
|
642 |
+
filters=320
|
643 |
+
size=1
|
644 |
+
stride=1
|
645 |
+
pad=1
|
646 |
+
activation=swish
|
647 |
+
|
648 |
+
[convolutional]
|
649 |
+
batch_normalize=1
|
650 |
+
filters=320
|
651 |
+
size=3
|
652 |
+
stride=1
|
653 |
+
pad=1
|
654 |
+
activation=swish
|
655 |
+
|
656 |
+
[shortcut]
|
657 |
+
from=-3
|
658 |
+
activation=linear
|
659 |
+
|
660 |
+
[convolutional]
|
661 |
+
batch_normalize=1
|
662 |
+
filters=320
|
663 |
+
size=1
|
664 |
+
stride=1
|
665 |
+
pad=1
|
666 |
+
activation=swish
|
667 |
+
|
668 |
+
[convolutional]
|
669 |
+
batch_normalize=1
|
670 |
+
filters=320
|
671 |
+
size=3
|
672 |
+
stride=1
|
673 |
+
pad=1
|
674 |
+
activation=swish
|
675 |
+
|
676 |
+
[shortcut]
|
677 |
+
from=-3
|
678 |
+
activation=linear
|
679 |
+
|
680 |
+
[convolutional]
|
681 |
+
batch_normalize=1
|
682 |
+
filters=320
|
683 |
+
size=1
|
684 |
+
stride=1
|
685 |
+
pad=1
|
686 |
+
activation=swish
|
687 |
+
|
688 |
+
[convolutional]
|
689 |
+
batch_normalize=1
|
690 |
+
filters=320
|
691 |
+
size=3
|
692 |
+
stride=1
|
693 |
+
pad=1
|
694 |
+
activation=swish
|
695 |
+
|
696 |
+
[shortcut]
|
697 |
+
from=-3
|
698 |
+
activation=linear
|
699 |
+
|
700 |
+
# Transition first
|
701 |
+
|
702 |
+
[convolutional]
|
703 |
+
batch_normalize=1
|
704 |
+
filters=320
|
705 |
+
size=1
|
706 |
+
stride=1
|
707 |
+
pad=1
|
708 |
+
activation=swish
|
709 |
+
|
710 |
+
# Merge [-1 -(3k+4)]
|
711 |
+
|
712 |
+
[route]
|
713 |
+
layers = -1,-34
|
714 |
+
|
715 |
+
# Transition last
|
716 |
+
|
717 |
+
# 94 (previous+7+3k)
|
718 |
+
[convolutional]
|
719 |
+
batch_normalize=1
|
720 |
+
filters=640
|
721 |
+
size=1
|
722 |
+
stride=1
|
723 |
+
pad=1
|
724 |
+
activation=swish
|
725 |
+
|
726 |
+
# P5
|
727 |
+
|
728 |
+
# Downsample
|
729 |
+
|
730 |
+
[convolutional]
|
731 |
+
batch_normalize=1
|
732 |
+
filters=1280
|
733 |
+
size=3
|
734 |
+
stride=2
|
735 |
+
pad=1
|
736 |
+
activation=swish
|
737 |
+
|
738 |
+
# Split
|
739 |
+
|
740 |
+
[convolutional]
|
741 |
+
batch_normalize=1
|
742 |
+
filters=640
|
743 |
+
size=1
|
744 |
+
stride=1
|
745 |
+
pad=1
|
746 |
+
activation=swish
|
747 |
+
|
748 |
+
[route]
|
749 |
+
layers = -2
|
750 |
+
|
751 |
+
[convolutional]
|
752 |
+
batch_normalize=1
|
753 |
+
filters=640
|
754 |
+
size=1
|
755 |
+
stride=1
|
756 |
+
pad=1
|
757 |
+
activation=swish
|
758 |
+
|
759 |
+
# Residual Block
|
760 |
+
|
761 |
+
[convolutional]
|
762 |
+
batch_normalize=1
|
763 |
+
filters=640
|
764 |
+
size=1
|
765 |
+
stride=1
|
766 |
+
pad=1
|
767 |
+
activation=swish
|
768 |
+
|
769 |
+
[convolutional]
|
770 |
+
batch_normalize=1
|
771 |
+
filters=640
|
772 |
+
size=3
|
773 |
+
stride=1
|
774 |
+
pad=1
|
775 |
+
activation=swish
|
776 |
+
|
777 |
+
[shortcut]
|
778 |
+
from=-3
|
779 |
+
activation=linear
|
780 |
+
|
781 |
+
[convolutional]
|
782 |
+
batch_normalize=1
|
783 |
+
filters=640
|
784 |
+
size=1
|
785 |
+
stride=1
|
786 |
+
pad=1
|
787 |
+
activation=swish
|
788 |
+
|
789 |
+
[convolutional]
|
790 |
+
batch_normalize=1
|
791 |
+
filters=640
|
792 |
+
size=3
|
793 |
+
stride=1
|
794 |
+
pad=1
|
795 |
+
activation=swish
|
796 |
+
|
797 |
+
[shortcut]
|
798 |
+
from=-3
|
799 |
+
activation=linear
|
800 |
+
|
801 |
+
[convolutional]
|
802 |
+
batch_normalize=1
|
803 |
+
filters=640
|
804 |
+
size=1
|
805 |
+
stride=1
|
806 |
+
pad=1
|
807 |
+
activation=swish
|
808 |
+
|
809 |
+
[convolutional]
|
810 |
+
batch_normalize=1
|
811 |
+
filters=640
|
812 |
+
size=3
|
813 |
+
stride=1
|
814 |
+
pad=1
|
815 |
+
activation=swish
|
816 |
+
|
817 |
+
[shortcut]
|
818 |
+
from=-3
|
819 |
+
activation=linear
|
820 |
+
|
821 |
+
[convolutional]
|
822 |
+
batch_normalize=1
|
823 |
+
filters=640
|
824 |
+
size=1
|
825 |
+
stride=1
|
826 |
+
pad=1
|
827 |
+
activation=swish
|
828 |
+
|
829 |
+
[convolutional]
|
830 |
+
batch_normalize=1
|
831 |
+
filters=640
|
832 |
+
size=3
|
833 |
+
stride=1
|
834 |
+
pad=1
|
835 |
+
activation=swish
|
836 |
+
|
837 |
+
[shortcut]
|
838 |
+
from=-3
|
839 |
+
activation=linear
|
840 |
+
|
841 |
+
[convolutional]
|
842 |
+
batch_normalize=1
|
843 |
+
filters=640
|
844 |
+
size=1
|
845 |
+
stride=1
|
846 |
+
pad=1
|
847 |
+
activation=swish
|
848 |
+
|
849 |
+
[convolutional]
|
850 |
+
batch_normalize=1
|
851 |
+
filters=640
|
852 |
+
size=3
|
853 |
+
stride=1
|
854 |
+
pad=1
|
855 |
+
activation=swish
|
856 |
+
|
857 |
+
[shortcut]
|
858 |
+
from=-3
|
859 |
+
activation=linear
|
860 |
+
|
861 |
+
# Transition first
|
862 |
+
|
863 |
+
[convolutional]
|
864 |
+
batch_normalize=1
|
865 |
+
filters=640
|
866 |
+
size=1
|
867 |
+
stride=1
|
868 |
+
pad=1
|
869 |
+
activation=swish
|
870 |
+
|
871 |
+
# Merge [-1 -(3k+4)]
|
872 |
+
|
873 |
+
[route]
|
874 |
+
layers = -1,-19
|
875 |
+
|
876 |
+
# Transition last
|
877 |
+
|
878 |
+
# 116 (previous+7+3k)
|
879 |
+
[convolutional]
|
880 |
+
batch_normalize=1
|
881 |
+
filters=1280
|
882 |
+
size=1
|
883 |
+
stride=1
|
884 |
+
pad=1
|
885 |
+
activation=swish
|
886 |
+
|
887 |
+
# ============ End of Backbone ============ #
|
888 |
+
|
889 |
+
# ============ Neck ============ #
|
890 |
+
|
891 |
+
# CSPSPP
|
892 |
+
|
893 |
+
[convolutional]
|
894 |
+
batch_normalize=1
|
895 |
+
filters=640
|
896 |
+
size=1
|
897 |
+
stride=1
|
898 |
+
pad=1
|
899 |
+
activation=swish
|
900 |
+
|
901 |
+
[route]
|
902 |
+
layers = -2
|
903 |
+
|
904 |
+
[convolutional]
|
905 |
+
batch_normalize=1
|
906 |
+
filters=640
|
907 |
+
size=1
|
908 |
+
stride=1
|
909 |
+
pad=1
|
910 |
+
activation=swish
|
911 |
+
|
912 |
+
[convolutional]
|
913 |
+
batch_normalize=1
|
914 |
+
size=3
|
915 |
+
stride=1
|
916 |
+
pad=1
|
917 |
+
filters=640
|
918 |
+
activation=swish
|
919 |
+
|
920 |
+
[convolutional]
|
921 |
+
batch_normalize=1
|
922 |
+
filters=640
|
923 |
+
size=1
|
924 |
+
stride=1
|
925 |
+
pad=1
|
926 |
+
activation=swish
|
927 |
+
|
928 |
+
### SPP ###
|
929 |
+
[maxpool]
|
930 |
+
stride=1
|
931 |
+
size=5
|
932 |
+
|
933 |
+
[route]
|
934 |
+
layers=-2
|
935 |
+
|
936 |
+
[maxpool]
|
937 |
+
stride=1
|
938 |
+
size=9
|
939 |
+
|
940 |
+
[route]
|
941 |
+
layers=-4
|
942 |
+
|
943 |
+
[maxpool]
|
944 |
+
stride=1
|
945 |
+
size=13
|
946 |
+
|
947 |
+
[route]
|
948 |
+
layers=-1,-3,-5,-6
|
949 |
+
### End SPP ###
|
950 |
+
|
951 |
+
[convolutional]
|
952 |
+
batch_normalize=1
|
953 |
+
filters=640
|
954 |
+
size=1
|
955 |
+
stride=1
|
956 |
+
pad=1
|
957 |
+
activation=swish
|
958 |
+
|
959 |
+
[convolutional]
|
960 |
+
batch_normalize=1
|
961 |
+
size=3
|
962 |
+
stride=1
|
963 |
+
pad=1
|
964 |
+
filters=640
|
965 |
+
activation=swish
|
966 |
+
|
967 |
+
[convolutional]
|
968 |
+
batch_normalize=1
|
969 |
+
filters=640
|
970 |
+
size=1
|
971 |
+
stride=1
|
972 |
+
pad=1
|
973 |
+
activation=swish
|
974 |
+
|
975 |
+
[convolutional]
|
976 |
+
batch_normalize=1
|
977 |
+
size=3
|
978 |
+
stride=1
|
979 |
+
pad=1
|
980 |
+
filters=640
|
981 |
+
activation=swish
|
982 |
+
|
983 |
+
[route]
|
984 |
+
layers = -1, -15
|
985 |
+
|
986 |
+
# 133 (previous+6+5+2k)
|
987 |
+
[convolutional]
|
988 |
+
batch_normalize=1
|
989 |
+
filters=640
|
990 |
+
size=1
|
991 |
+
stride=1
|
992 |
+
pad=1
|
993 |
+
activation=swish
|
994 |
+
|
995 |
+
# End of CSPSPP
|
996 |
+
|
997 |
+
|
998 |
+
# FPN-4
|
999 |
+
|
1000 |
+
[convolutional]
|
1001 |
+
batch_normalize=1
|
1002 |
+
filters=320
|
1003 |
+
size=1
|
1004 |
+
stride=1
|
1005 |
+
pad=1
|
1006 |
+
activation=swish
|
1007 |
+
|
1008 |
+
[upsample]
|
1009 |
+
stride=2
|
1010 |
+
|
1011 |
+
[route]
|
1012 |
+
layers = 94
|
1013 |
+
|
1014 |
+
[convolutional]
|
1015 |
+
batch_normalize=1
|
1016 |
+
filters=320
|
1017 |
+
size=1
|
1018 |
+
stride=1
|
1019 |
+
pad=1
|
1020 |
+
activation=swish
|
1021 |
+
|
1022 |
+
[route]
|
1023 |
+
layers = -1, -3
|
1024 |
+
|
1025 |
+
[convolutional]
|
1026 |
+
batch_normalize=1
|
1027 |
+
filters=320
|
1028 |
+
size=1
|
1029 |
+
stride=1
|
1030 |
+
pad=1
|
1031 |
+
activation=swish
|
1032 |
+
|
1033 |
+
# Split
|
1034 |
+
|
1035 |
+
[convolutional]
|
1036 |
+
batch_normalize=1
|
1037 |
+
filters=320
|
1038 |
+
size=1
|
1039 |
+
stride=1
|
1040 |
+
pad=1
|
1041 |
+
activation=swish
|
1042 |
+
|
1043 |
+
[route]
|
1044 |
+
layers = -2
|
1045 |
+
|
1046 |
+
# Plain Block
|
1047 |
+
|
1048 |
+
[convolutional]
|
1049 |
+
batch_normalize=1
|
1050 |
+
filters=320
|
1051 |
+
size=1
|
1052 |
+
stride=1
|
1053 |
+
pad=1
|
1054 |
+
activation=swish
|
1055 |
+
|
1056 |
+
[convolutional]
|
1057 |
+
batch_normalize=1
|
1058 |
+
size=3
|
1059 |
+
stride=1
|
1060 |
+
pad=1
|
1061 |
+
filters=320
|
1062 |
+
activation=swish
|
1063 |
+
|
1064 |
+
[convolutional]
|
1065 |
+
batch_normalize=1
|
1066 |
+
filters=320
|
1067 |
+
size=1
|
1068 |
+
stride=1
|
1069 |
+
pad=1
|
1070 |
+
activation=swish
|
1071 |
+
|
1072 |
+
[convolutional]
|
1073 |
+
batch_normalize=1
|
1074 |
+
size=3
|
1075 |
+
stride=1
|
1076 |
+
pad=1
|
1077 |
+
filters=320
|
1078 |
+
activation=swish
|
1079 |
+
|
1080 |
+
[convolutional]
|
1081 |
+
batch_normalize=1
|
1082 |
+
filters=320
|
1083 |
+
size=1
|
1084 |
+
stride=1
|
1085 |
+
pad=1
|
1086 |
+
activation=swish
|
1087 |
+
|
1088 |
+
[convolutional]
|
1089 |
+
batch_normalize=1
|
1090 |
+
size=3
|
1091 |
+
stride=1
|
1092 |
+
pad=1
|
1093 |
+
filters=320
|
1094 |
+
activation=swish
|
1095 |
+
|
1096 |
+
# Merge [-1, -(2k+2)]
|
1097 |
+
|
1098 |
+
[route]
|
1099 |
+
layers = -1, -8
|
1100 |
+
|
1101 |
+
# Transition last
|
1102 |
+
|
1103 |
+
# 149 (previous+6+4+2k)
|
1104 |
+
[convolutional]
|
1105 |
+
batch_normalize=1
|
1106 |
+
filters=320
|
1107 |
+
size=1
|
1108 |
+
stride=1
|
1109 |
+
pad=1
|
1110 |
+
activation=swish
|
1111 |
+
|
1112 |
+
|
1113 |
+
# FPN-3
|
1114 |
+
|
1115 |
+
[convolutional]
|
1116 |
+
batch_normalize=1
|
1117 |
+
filters=160
|
1118 |
+
size=1
|
1119 |
+
stride=1
|
1120 |
+
pad=1
|
1121 |
+
activation=swish
|
1122 |
+
|
1123 |
+
[upsample]
|
1124 |
+
stride=2
|
1125 |
+
|
1126 |
+
[route]
|
1127 |
+
layers = 57
|
1128 |
+
|
1129 |
+
[convolutional]
|
1130 |
+
batch_normalize=1
|
1131 |
+
filters=160
|
1132 |
+
size=1
|
1133 |
+
stride=1
|
1134 |
+
pad=1
|
1135 |
+
activation=swish
|
1136 |
+
|
1137 |
+
[route]
|
1138 |
+
layers = -1, -3
|
1139 |
+
|
1140 |
+
[convolutional]
|
1141 |
+
batch_normalize=1
|
1142 |
+
filters=160
|
1143 |
+
size=1
|
1144 |
+
stride=1
|
1145 |
+
pad=1
|
1146 |
+
activation=swish
|
1147 |
+
|
1148 |
+
# Split
|
1149 |
+
|
1150 |
+
[convolutional]
|
1151 |
+
batch_normalize=1
|
1152 |
+
filters=160
|
1153 |
+
size=1
|
1154 |
+
stride=1
|
1155 |
+
pad=1
|
1156 |
+
activation=swish
|
1157 |
+
|
1158 |
+
[route]
|
1159 |
+
layers = -2
|
1160 |
+
|
1161 |
+
# Plain Block
|
1162 |
+
|
1163 |
+
[convolutional]
|
1164 |
+
batch_normalize=1
|
1165 |
+
filters=160
|
1166 |
+
size=1
|
1167 |
+
stride=1
|
1168 |
+
pad=1
|
1169 |
+
activation=swish
|
1170 |
+
|
1171 |
+
[convolutional]
|
1172 |
+
batch_normalize=1
|
1173 |
+
size=3
|
1174 |
+
stride=1
|
1175 |
+
pad=1
|
1176 |
+
filters=160
|
1177 |
+
activation=swish
|
1178 |
+
|
1179 |
+
[convolutional]
|
1180 |
+
batch_normalize=1
|
1181 |
+
filters=160
|
1182 |
+
size=1
|
1183 |
+
stride=1
|
1184 |
+
pad=1
|
1185 |
+
activation=swish
|
1186 |
+
|
1187 |
+
[convolutional]
|
1188 |
+
batch_normalize=1
|
1189 |
+
size=3
|
1190 |
+
stride=1
|
1191 |
+
pad=1
|
1192 |
+
filters=160
|
1193 |
+
activation=swish
|
1194 |
+
|
1195 |
+
[convolutional]
|
1196 |
+
batch_normalize=1
|
1197 |
+
filters=160
|
1198 |
+
size=1
|
1199 |
+
stride=1
|
1200 |
+
pad=1
|
1201 |
+
activation=swish
|
1202 |
+
|
1203 |
+
[convolutional]
|
1204 |
+
batch_normalize=1
|
1205 |
+
size=3
|
1206 |
+
stride=1
|
1207 |
+
pad=1
|
1208 |
+
filters=160
|
1209 |
+
activation=swish
|
1210 |
+
|
1211 |
+
# Merge [-1, -(2k+2)]
|
1212 |
+
|
1213 |
+
[route]
|
1214 |
+
layers = -1, -8
|
1215 |
+
|
1216 |
+
# Transition last
|
1217 |
+
|
1218 |
+
# 165 (previous+6+4+2k)
|
1219 |
+
[convolutional]
|
1220 |
+
batch_normalize=1
|
1221 |
+
filters=160
|
1222 |
+
size=1
|
1223 |
+
stride=1
|
1224 |
+
pad=1
|
1225 |
+
activation=swish
|
1226 |
+
|
1227 |
+
|
1228 |
+
# PAN-4
|
1229 |
+
|
1230 |
+
[convolutional]
|
1231 |
+
batch_normalize=1
|
1232 |
+
size=3
|
1233 |
+
stride=2
|
1234 |
+
pad=1
|
1235 |
+
filters=320
|
1236 |
+
activation=swish
|
1237 |
+
|
1238 |
+
[route]
|
1239 |
+
layers = -1, 149
|
1240 |
+
|
1241 |
+
[convolutional]
|
1242 |
+
batch_normalize=1
|
1243 |
+
filters=320
|
1244 |
+
size=1
|
1245 |
+
stride=1
|
1246 |
+
pad=1
|
1247 |
+
activation=swish
|
1248 |
+
|
1249 |
+
# Split
|
1250 |
+
|
1251 |
+
[convolutional]
|
1252 |
+
batch_normalize=1
|
1253 |
+
filters=320
|
1254 |
+
size=1
|
1255 |
+
stride=1
|
1256 |
+
pad=1
|
1257 |
+
activation=swish
|
1258 |
+
|
1259 |
+
[route]
|
1260 |
+
layers = -2
|
1261 |
+
|
1262 |
+
# Plain Block
|
1263 |
+
|
1264 |
+
[convolutional]
|
1265 |
+
batch_normalize=1
|
1266 |
+
filters=320
|
1267 |
+
size=1
|
1268 |
+
stride=1
|
1269 |
+
pad=1
|
1270 |
+
activation=swish
|
1271 |
+
|
1272 |
+
[convolutional]
|
1273 |
+
batch_normalize=1
|
1274 |
+
size=3
|
1275 |
+
stride=1
|
1276 |
+
pad=1
|
1277 |
+
filters=320
|
1278 |
+
activation=swish
|
1279 |
+
|
1280 |
+
[convolutional]
|
1281 |
+
batch_normalize=1
|
1282 |
+
filters=320
|
1283 |
+
size=1
|
1284 |
+
stride=1
|
1285 |
+
pad=1
|
1286 |
+
activation=swish
|
1287 |
+
|
1288 |
+
[convolutional]
|
1289 |
+
batch_normalize=1
|
1290 |
+
size=3
|
1291 |
+
stride=1
|
1292 |
+
pad=1
|
1293 |
+
filters=320
|
1294 |
+
activation=swish
|
1295 |
+
|
1296 |
+
[convolutional]
|
1297 |
+
batch_normalize=1
|
1298 |
+
filters=320
|
1299 |
+
size=1
|
1300 |
+
stride=1
|
1301 |
+
pad=1
|
1302 |
+
activation=swish
|
1303 |
+
|
1304 |
+
[convolutional]
|
1305 |
+
batch_normalize=1
|
1306 |
+
size=3
|
1307 |
+
stride=1
|
1308 |
+
pad=1
|
1309 |
+
filters=320
|
1310 |
+
activation=swish
|
1311 |
+
|
1312 |
+
[route]
|
1313 |
+
layers = -1,-8
|
1314 |
+
|
1315 |
+
# Transition last
|
1316 |
+
|
1317 |
+
# 178 (previous+3+4+2k)
|
1318 |
+
[convolutional]
|
1319 |
+
batch_normalize=1
|
1320 |
+
filters=320
|
1321 |
+
size=1
|
1322 |
+
stride=1
|
1323 |
+
pad=1
|
1324 |
+
activation=swish
|
1325 |
+
|
1326 |
+
|
1327 |
+
# PAN-5
|
1328 |
+
|
1329 |
+
[convolutional]
|
1330 |
+
batch_normalize=1
|
1331 |
+
size=3
|
1332 |
+
stride=2
|
1333 |
+
pad=1
|
1334 |
+
filters=640
|
1335 |
+
activation=swish
|
1336 |
+
|
1337 |
+
[route]
|
1338 |
+
layers = -1, 133
|
1339 |
+
|
1340 |
+
[convolutional]
|
1341 |
+
batch_normalize=1
|
1342 |
+
filters=640
|
1343 |
+
size=1
|
1344 |
+
stride=1
|
1345 |
+
pad=1
|
1346 |
+
activation=swish
|
1347 |
+
|
1348 |
+
# Split
|
1349 |
+
|
1350 |
+
[convolutional]
|
1351 |
+
batch_normalize=1
|
1352 |
+
filters=640
|
1353 |
+
size=1
|
1354 |
+
stride=1
|
1355 |
+
pad=1
|
1356 |
+
activation=swish
|
1357 |
+
|
1358 |
+
[route]
|
1359 |
+
layers = -2
|
1360 |
+
|
1361 |
+
# Plain Block
|
1362 |
+
|
1363 |
+
[convolutional]
|
1364 |
+
batch_normalize=1
|
1365 |
+
filters=640
|
1366 |
+
size=1
|
1367 |
+
stride=1
|
1368 |
+
pad=1
|
1369 |
+
activation=swish
|
1370 |
+
|
1371 |
+
[convolutional]
|
1372 |
+
batch_normalize=1
|
1373 |
+
size=3
|
1374 |
+
stride=1
|
1375 |
+
pad=1
|
1376 |
+
filters=640
|
1377 |
+
activation=swish
|
1378 |
+
|
1379 |
+
[convolutional]
|
1380 |
+
batch_normalize=1
|
1381 |
+
filters=640
|
1382 |
+
size=1
|
1383 |
+
stride=1
|
1384 |
+
pad=1
|
1385 |
+
activation=swish
|
1386 |
+
|
1387 |
+
[convolutional]
|
1388 |
+
batch_normalize=1
|
1389 |
+
size=3
|
1390 |
+
stride=1
|
1391 |
+
pad=1
|
1392 |
+
filters=640
|
1393 |
+
activation=swish
|
1394 |
+
|
1395 |
+
[convolutional]
|
1396 |
+
batch_normalize=1
|
1397 |
+
filters=640
|
1398 |
+
size=1
|
1399 |
+
stride=1
|
1400 |
+
pad=1
|
1401 |
+
activation=swish
|
1402 |
+
|
1403 |
+
[convolutional]
|
1404 |
+
batch_normalize=1
|
1405 |
+
size=3
|
1406 |
+
stride=1
|
1407 |
+
pad=1
|
1408 |
+
filters=640
|
1409 |
+
activation=swish
|
1410 |
+
|
1411 |
+
[route]
|
1412 |
+
layers = -1,-8
|
1413 |
+
|
1414 |
+
# Transition last
|
1415 |
+
|
1416 |
+
# 191 (previous+3+4+2k)
|
1417 |
+
[convolutional]
|
1418 |
+
batch_normalize=1
|
1419 |
+
filters=640
|
1420 |
+
size=1
|
1421 |
+
stride=1
|
1422 |
+
pad=1
|
1423 |
+
activation=swish
|
1424 |
+
|
1425 |
+
# ============ End of Neck ============ #
|
1426 |
+
|
1427 |
+
# ============ Head ============ #
|
1428 |
+
|
1429 |
+
# YOLO-3
|
1430 |
+
|
1431 |
+
[route]
|
1432 |
+
layers = 165
|
1433 |
+
|
1434 |
+
[convolutional]
|
1435 |
+
batch_normalize=1
|
1436 |
+
size=3
|
1437 |
+
stride=1
|
1438 |
+
pad=1
|
1439 |
+
filters=320
|
1440 |
+
activation=swish
|
1441 |
+
|
1442 |
+
[convolutional]
|
1443 |
+
size=1
|
1444 |
+
stride=1
|
1445 |
+
pad=1
|
1446 |
+
filters=255
|
1447 |
+
activation=logistic
|
1448 |
+
|
1449 |
+
[yolo]
|
1450 |
+
mask = 0,1,2
|
1451 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1452 |
+
classes=80
|
1453 |
+
num=9
|
1454 |
+
jitter=.1
|
1455 |
+
scale_x_y = 2.0
|
1456 |
+
objectness_smooth=1
|
1457 |
+
ignore_thresh = .7
|
1458 |
+
truth_thresh = 1
|
1459 |
+
#random=1
|
1460 |
+
resize=1.5
|
1461 |
+
iou_thresh=0.2
|
1462 |
+
iou_normalizer=0.05
|
1463 |
+
cls_normalizer=0.5
|
1464 |
+
obj_normalizer=0.4
|
1465 |
+
iou_loss=ciou
|
1466 |
+
nms_kind=diounms
|
1467 |
+
beta_nms=0.6
|
1468 |
+
new_coords=1
|
1469 |
+
max_delta=2
|
1470 |
+
|
1471 |
+
|
1472 |
+
# YOLO-4
|
1473 |
+
|
1474 |
+
[route]
|
1475 |
+
layers = 178
|
1476 |
+
|
1477 |
+
[convolutional]
|
1478 |
+
batch_normalize=1
|
1479 |
+
size=3
|
1480 |
+
stride=1
|
1481 |
+
pad=1
|
1482 |
+
filters=640
|
1483 |
+
activation=swish
|
1484 |
+
|
1485 |
+
[convolutional]
|
1486 |
+
size=1
|
1487 |
+
stride=1
|
1488 |
+
pad=1
|
1489 |
+
filters=255
|
1490 |
+
activation=logistic
|
1491 |
+
|
1492 |
+
[yolo]
|
1493 |
+
mask = 3,4,5
|
1494 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1495 |
+
classes=80
|
1496 |
+
num=9
|
1497 |
+
jitter=.1
|
1498 |
+
scale_x_y = 2.0
|
1499 |
+
objectness_smooth=1
|
1500 |
+
ignore_thresh = .7
|
1501 |
+
truth_thresh = 1
|
1502 |
+
#random=1
|
1503 |
+
resize=1.5
|
1504 |
+
iou_thresh=0.2
|
1505 |
+
iou_normalizer=0.05
|
1506 |
+
cls_normalizer=0.5
|
1507 |
+
obj_normalizer=0.4
|
1508 |
+
iou_loss=ciou
|
1509 |
+
nms_kind=diounms
|
1510 |
+
beta_nms=0.6
|
1511 |
+
new_coords=1
|
1512 |
+
max_delta=2
|
1513 |
+
|
1514 |
+
|
1515 |
+
# YOLO-5
|
1516 |
+
|
1517 |
+
[route]
|
1518 |
+
layers = 191
|
1519 |
+
|
1520 |
+
[convolutional]
|
1521 |
+
batch_normalize=1
|
1522 |
+
size=3
|
1523 |
+
stride=1
|
1524 |
+
pad=1
|
1525 |
+
filters=1280
|
1526 |
+
activation=swish
|
1527 |
+
|
1528 |
+
[convolutional]
|
1529 |
+
size=1
|
1530 |
+
stride=1
|
1531 |
+
pad=1
|
1532 |
+
filters=255
|
1533 |
+
activation=logistic
|
1534 |
+
|
1535 |
+
[yolo]
|
1536 |
+
mask = 6,7,8
|
1537 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1538 |
+
classes=80
|
1539 |
+
num=9
|
1540 |
+
jitter=.1
|
1541 |
+
scale_x_y = 2.0
|
1542 |
+
objectness_smooth=1
|
1543 |
+
ignore_thresh = .7
|
1544 |
+
truth_thresh = 1
|
1545 |
+
#random=1
|
1546 |
+
resize=1.5
|
1547 |
+
iou_thresh=0.2
|
1548 |
+
iou_normalizer=0.05
|
1549 |
+
cls_normalizer=0.5
|
1550 |
+
obj_normalizer=0.4
|
1551 |
+
iou_loss=ciou
|
1552 |
+
nms_kind=diounms
|
1553 |
+
beta_nms=0.6
|
1554 |
+
new_coords=1
|
1555 |
+
max_delta=2
|
darknet/cfg/yolov4-csp.cfg
ADDED
@@ -0,0 +1,1354 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[net]
|
2 |
+
# Testing
|
3 |
+
#batch=1
|
4 |
+
#subdivisions=1
|
5 |
+
# Training
|
6 |
+
batch=64
|
7 |
+
subdivisions=8
|
8 |
+
width=640
|
9 |
+
height=640
|
10 |
+
channels=3
|
11 |
+
momentum=0.949
|
12 |
+
decay=0.0005
|
13 |
+
angle=0
|
14 |
+
saturation = 1.5
|
15 |
+
exposure = 1.5
|
16 |
+
hue=.1
|
17 |
+
|
18 |
+
learning_rate=0.001
|
19 |
+
burn_in=1000
|
20 |
+
max_batches = 500500
|
21 |
+
policy=steps
|
22 |
+
steps=400000,450000
|
23 |
+
scales=.1,.1
|
24 |
+
|
25 |
+
mosaic=1
|
26 |
+
|
27 |
+
letter_box=1
|
28 |
+
|
29 |
+
ema_alpha=0.9998
|
30 |
+
|
31 |
+
#optimized_memory=1
|
32 |
+
|
33 |
+
|
34 |
+
# ============ Backbone ============ #
|
35 |
+
|
36 |
+
# Stem
|
37 |
+
|
38 |
+
# 0
|
39 |
+
[convolutional]
|
40 |
+
batch_normalize=1
|
41 |
+
filters=32
|
42 |
+
size=3
|
43 |
+
stride=1
|
44 |
+
pad=1
|
45 |
+
activation=swish
|
46 |
+
|
47 |
+
# P1
|
48 |
+
|
49 |
+
# Downsample
|
50 |
+
|
51 |
+
[convolutional]
|
52 |
+
batch_normalize=1
|
53 |
+
filters=64
|
54 |
+
size=3
|
55 |
+
stride=2
|
56 |
+
pad=1
|
57 |
+
activation=swish
|
58 |
+
|
59 |
+
# Residual Block
|
60 |
+
|
61 |
+
[convolutional]
|
62 |
+
batch_normalize=1
|
63 |
+
filters=32
|
64 |
+
size=1
|
65 |
+
stride=1
|
66 |
+
pad=1
|
67 |
+
activation=swish
|
68 |
+
|
69 |
+
[convolutional]
|
70 |
+
batch_normalize=1
|
71 |
+
filters=64
|
72 |
+
size=3
|
73 |
+
stride=1
|
74 |
+
pad=1
|
75 |
+
activation=swish
|
76 |
+
|
77 |
+
# 4 (previous+1+3k)
|
78 |
+
[shortcut]
|
79 |
+
from=-3
|
80 |
+
activation=linear
|
81 |
+
|
82 |
+
# P2
|
83 |
+
|
84 |
+
# Downsample
|
85 |
+
|
86 |
+
[convolutional]
|
87 |
+
batch_normalize=1
|
88 |
+
filters=128
|
89 |
+
size=3
|
90 |
+
stride=2
|
91 |
+
pad=1
|
92 |
+
activation=swish
|
93 |
+
|
94 |
+
# Split
|
95 |
+
|
96 |
+
[convolutional]
|
97 |
+
batch_normalize=1
|
98 |
+
filters=64
|
99 |
+
size=1
|
100 |
+
stride=1
|
101 |
+
pad=1
|
102 |
+
activation=swish
|
103 |
+
|
104 |
+
[route]
|
105 |
+
layers = -2
|
106 |
+
|
107 |
+
[convolutional]
|
108 |
+
batch_normalize=1
|
109 |
+
filters=64
|
110 |
+
size=1
|
111 |
+
stride=1
|
112 |
+
pad=1
|
113 |
+
activation=swish
|
114 |
+
|
115 |
+
# Residual Block
|
116 |
+
|
117 |
+
[convolutional]
|
118 |
+
batch_normalize=1
|
119 |
+
filters=64
|
120 |
+
size=1
|
121 |
+
stride=1
|
122 |
+
pad=1
|
123 |
+
activation=swish
|
124 |
+
|
125 |
+
[convolutional]
|
126 |
+
batch_normalize=1
|
127 |
+
filters=64
|
128 |
+
size=3
|
129 |
+
stride=1
|
130 |
+
pad=1
|
131 |
+
activation=swish
|
132 |
+
|
133 |
+
[shortcut]
|
134 |
+
from=-3
|
135 |
+
activation=linear
|
136 |
+
|
137 |
+
[convolutional]
|
138 |
+
batch_normalize=1
|
139 |
+
filters=64
|
140 |
+
size=1
|
141 |
+
stride=1
|
142 |
+
pad=1
|
143 |
+
activation=swish
|
144 |
+
|
145 |
+
[convolutional]
|
146 |
+
batch_normalize=1
|
147 |
+
filters=64
|
148 |
+
size=3
|
149 |
+
stride=1
|
150 |
+
pad=1
|
151 |
+
activation=swish
|
152 |
+
|
153 |
+
[shortcut]
|
154 |
+
from=-3
|
155 |
+
activation=linear
|
156 |
+
|
157 |
+
# Transition first
|
158 |
+
|
159 |
+
[convolutional]
|
160 |
+
batch_normalize=1
|
161 |
+
filters=64
|
162 |
+
size=1
|
163 |
+
stride=1
|
164 |
+
pad=1
|
165 |
+
activation=swish
|
166 |
+
|
167 |
+
# Merge [-1, -(3k+4)]
|
168 |
+
|
169 |
+
[route]
|
170 |
+
layers = -1,-10
|
171 |
+
|
172 |
+
# Transition last
|
173 |
+
|
174 |
+
# 17 (previous+7+3k)
|
175 |
+
[convolutional]
|
176 |
+
batch_normalize=1
|
177 |
+
filters=128
|
178 |
+
size=1
|
179 |
+
stride=1
|
180 |
+
pad=1
|
181 |
+
activation=swish
|
182 |
+
|
183 |
+
# P3
|
184 |
+
|
185 |
+
# Downsample
|
186 |
+
|
187 |
+
[convolutional]
|
188 |
+
batch_normalize=1
|
189 |
+
filters=256
|
190 |
+
size=3
|
191 |
+
stride=2
|
192 |
+
pad=1
|
193 |
+
activation=swish
|
194 |
+
|
195 |
+
# Split
|
196 |
+
|
197 |
+
[convolutional]
|
198 |
+
batch_normalize=1
|
199 |
+
filters=128
|
200 |
+
size=1
|
201 |
+
stride=1
|
202 |
+
pad=1
|
203 |
+
activation=swish
|
204 |
+
|
205 |
+
[route]
|
206 |
+
layers = -2
|
207 |
+
|
208 |
+
[convolutional]
|
209 |
+
batch_normalize=1
|
210 |
+
filters=128
|
211 |
+
size=1
|
212 |
+
stride=1
|
213 |
+
pad=1
|
214 |
+
activation=swish
|
215 |
+
|
216 |
+
# Residual Block
|
217 |
+
|
218 |
+
[convolutional]
|
219 |
+
batch_normalize=1
|
220 |
+
filters=128
|
221 |
+
size=1
|
222 |
+
stride=1
|
223 |
+
pad=1
|
224 |
+
activation=swish
|
225 |
+
|
226 |
+
[convolutional]
|
227 |
+
batch_normalize=1
|
228 |
+
filters=128
|
229 |
+
size=3
|
230 |
+
stride=1
|
231 |
+
pad=1
|
232 |
+
activation=swish
|
233 |
+
|
234 |
+
[shortcut]
|
235 |
+
from=-3
|
236 |
+
activation=linear
|
237 |
+
|
238 |
+
[convolutional]
|
239 |
+
batch_normalize=1
|
240 |
+
filters=128
|
241 |
+
size=1
|
242 |
+
stride=1
|
243 |
+
pad=1
|
244 |
+
activation=swish
|
245 |
+
|
246 |
+
[convolutional]
|
247 |
+
batch_normalize=1
|
248 |
+
filters=128
|
249 |
+
size=3
|
250 |
+
stride=1
|
251 |
+
pad=1
|
252 |
+
activation=swish
|
253 |
+
|
254 |
+
[shortcut]
|
255 |
+
from=-3
|
256 |
+
activation=linear
|
257 |
+
|
258 |
+
[convolutional]
|
259 |
+
batch_normalize=1
|
260 |
+
filters=128
|
261 |
+
size=1
|
262 |
+
stride=1
|
263 |
+
pad=1
|
264 |
+
activation=swish
|
265 |
+
|
266 |
+
[convolutional]
|
267 |
+
batch_normalize=1
|
268 |
+
filters=128
|
269 |
+
size=3
|
270 |
+
stride=1
|
271 |
+
pad=1
|
272 |
+
activation=swish
|
273 |
+
|
274 |
+
[shortcut]
|
275 |
+
from=-3
|
276 |
+
activation=linear
|
277 |
+
|
278 |
+
[convolutional]
|
279 |
+
batch_normalize=1
|
280 |
+
filters=128
|
281 |
+
size=1
|
282 |
+
stride=1
|
283 |
+
pad=1
|
284 |
+
activation=swish
|
285 |
+
|
286 |
+
[convolutional]
|
287 |
+
batch_normalize=1
|
288 |
+
filters=128
|
289 |
+
size=3
|
290 |
+
stride=1
|
291 |
+
pad=1
|
292 |
+
activation=swish
|
293 |
+
|
294 |
+
[shortcut]
|
295 |
+
from=-3
|
296 |
+
activation=linear
|
297 |
+
|
298 |
+
[convolutional]
|
299 |
+
batch_normalize=1
|
300 |
+
filters=128
|
301 |
+
size=1
|
302 |
+
stride=1
|
303 |
+
pad=1
|
304 |
+
activation=swish
|
305 |
+
|
306 |
+
[convolutional]
|
307 |
+
batch_normalize=1
|
308 |
+
filters=128
|
309 |
+
size=3
|
310 |
+
stride=1
|
311 |
+
pad=1
|
312 |
+
activation=swish
|
313 |
+
|
314 |
+
[shortcut]
|
315 |
+
from=-3
|
316 |
+
activation=linear
|
317 |
+
|
318 |
+
[convolutional]
|
319 |
+
batch_normalize=1
|
320 |
+
filters=128
|
321 |
+
size=1
|
322 |
+
stride=1
|
323 |
+
pad=1
|
324 |
+
activation=swish
|
325 |
+
|
326 |
+
[convolutional]
|
327 |
+
batch_normalize=1
|
328 |
+
filters=128
|
329 |
+
size=3
|
330 |
+
stride=1
|
331 |
+
pad=1
|
332 |
+
activation=swish
|
333 |
+
|
334 |
+
[shortcut]
|
335 |
+
from=-3
|
336 |
+
activation=linear
|
337 |
+
|
338 |
+
[convolutional]
|
339 |
+
batch_normalize=1
|
340 |
+
filters=128
|
341 |
+
size=1
|
342 |
+
stride=1
|
343 |
+
pad=1
|
344 |
+
activation=swish
|
345 |
+
|
346 |
+
[convolutional]
|
347 |
+
batch_normalize=1
|
348 |
+
filters=128
|
349 |
+
size=3
|
350 |
+
stride=1
|
351 |
+
pad=1
|
352 |
+
activation=swish
|
353 |
+
|
354 |
+
[shortcut]
|
355 |
+
from=-3
|
356 |
+
activation=linear
|
357 |
+
|
358 |
+
[convolutional]
|
359 |
+
batch_normalize=1
|
360 |
+
filters=128
|
361 |
+
size=1
|
362 |
+
stride=1
|
363 |
+
pad=1
|
364 |
+
activation=swish
|
365 |
+
|
366 |
+
[convolutional]
|
367 |
+
batch_normalize=1
|
368 |
+
filters=128
|
369 |
+
size=3
|
370 |
+
stride=1
|
371 |
+
pad=1
|
372 |
+
activation=swish
|
373 |
+
|
374 |
+
[shortcut]
|
375 |
+
from=-3
|
376 |
+
activation=linear
|
377 |
+
|
378 |
+
# Transition first
|
379 |
+
|
380 |
+
[convolutional]
|
381 |
+
batch_normalize=1
|
382 |
+
filters=128
|
383 |
+
size=1
|
384 |
+
stride=1
|
385 |
+
pad=1
|
386 |
+
activation=swish
|
387 |
+
|
388 |
+
# Merge [-1 -(4+3k)]
|
389 |
+
|
390 |
+
[route]
|
391 |
+
layers = -1,-28
|
392 |
+
|
393 |
+
# Transition last
|
394 |
+
|
395 |
+
# 48 (previous+7+3k)
|
396 |
+
[convolutional]
|
397 |
+
batch_normalize=1
|
398 |
+
filters=256
|
399 |
+
size=1
|
400 |
+
stride=1
|
401 |
+
pad=1
|
402 |
+
activation=swish
|
403 |
+
|
404 |
+
# P4
|
405 |
+
|
406 |
+
# Downsample
|
407 |
+
|
408 |
+
[convolutional]
|
409 |
+
batch_normalize=1
|
410 |
+
filters=512
|
411 |
+
size=3
|
412 |
+
stride=2
|
413 |
+
pad=1
|
414 |
+
activation=swish
|
415 |
+
|
416 |
+
# Split
|
417 |
+
|
418 |
+
[convolutional]
|
419 |
+
batch_normalize=1
|
420 |
+
filters=256
|
421 |
+
size=1
|
422 |
+
stride=1
|
423 |
+
pad=1
|
424 |
+
activation=swish
|
425 |
+
|
426 |
+
[route]
|
427 |
+
layers = -2
|
428 |
+
|
429 |
+
[convolutional]
|
430 |
+
batch_normalize=1
|
431 |
+
filters=256
|
432 |
+
size=1
|
433 |
+
stride=1
|
434 |
+
pad=1
|
435 |
+
activation=swish
|
436 |
+
|
437 |
+
# Residual Block
|
438 |
+
|
439 |
+
[convolutional]
|
440 |
+
batch_normalize=1
|
441 |
+
filters=256
|
442 |
+
size=1
|
443 |
+
stride=1
|
444 |
+
pad=1
|
445 |
+
activation=swish
|
446 |
+
|
447 |
+
[convolutional]
|
448 |
+
batch_normalize=1
|
449 |
+
filters=256
|
450 |
+
size=3
|
451 |
+
stride=1
|
452 |
+
pad=1
|
453 |
+
activation=swish
|
454 |
+
|
455 |
+
[shortcut]
|
456 |
+
from=-3
|
457 |
+
activation=linear
|
458 |
+
|
459 |
+
[convolutional]
|
460 |
+
batch_normalize=1
|
461 |
+
filters=256
|
462 |
+
size=1
|
463 |
+
stride=1
|
464 |
+
pad=1
|
465 |
+
activation=swish
|
466 |
+
|
467 |
+
[convolutional]
|
468 |
+
batch_normalize=1
|
469 |
+
filters=256
|
470 |
+
size=3
|
471 |
+
stride=1
|
472 |
+
pad=1
|
473 |
+
activation=swish
|
474 |
+
|
475 |
+
[shortcut]
|
476 |
+
from=-3
|
477 |
+
activation=linear
|
478 |
+
|
479 |
+
[convolutional]
|
480 |
+
batch_normalize=1
|
481 |
+
filters=256
|
482 |
+
size=1
|
483 |
+
stride=1
|
484 |
+
pad=1
|
485 |
+
activation=swish
|
486 |
+
|
487 |
+
[convolutional]
|
488 |
+
batch_normalize=1
|
489 |
+
filters=256
|
490 |
+
size=3
|
491 |
+
stride=1
|
492 |
+
pad=1
|
493 |
+
activation=swish
|
494 |
+
|
495 |
+
[shortcut]
|
496 |
+
from=-3
|
497 |
+
activation=linear
|
498 |
+
|
499 |
+
[convolutional]
|
500 |
+
batch_normalize=1
|
501 |
+
filters=256
|
502 |
+
size=1
|
503 |
+
stride=1
|
504 |
+
pad=1
|
505 |
+
activation=swish
|
506 |
+
|
507 |
+
[convolutional]
|
508 |
+
batch_normalize=1
|
509 |
+
filters=256
|
510 |
+
size=3
|
511 |
+
stride=1
|
512 |
+
pad=1
|
513 |
+
activation=swish
|
514 |
+
|
515 |
+
[shortcut]
|
516 |
+
from=-3
|
517 |
+
activation=linear
|
518 |
+
|
519 |
+
[convolutional]
|
520 |
+
batch_normalize=1
|
521 |
+
filters=256
|
522 |
+
size=1
|
523 |
+
stride=1
|
524 |
+
pad=1
|
525 |
+
activation=swish
|
526 |
+
|
527 |
+
[convolutional]
|
528 |
+
batch_normalize=1
|
529 |
+
filters=256
|
530 |
+
size=3
|
531 |
+
stride=1
|
532 |
+
pad=1
|
533 |
+
activation=swish
|
534 |
+
|
535 |
+
[shortcut]
|
536 |
+
from=-3
|
537 |
+
activation=linear
|
538 |
+
|
539 |
+
[convolutional]
|
540 |
+
batch_normalize=1
|
541 |
+
filters=256
|
542 |
+
size=1
|
543 |
+
stride=1
|
544 |
+
pad=1
|
545 |
+
activation=swish
|
546 |
+
|
547 |
+
[convolutional]
|
548 |
+
batch_normalize=1
|
549 |
+
filters=256
|
550 |
+
size=3
|
551 |
+
stride=1
|
552 |
+
pad=1
|
553 |
+
activation=swish
|
554 |
+
|
555 |
+
[shortcut]
|
556 |
+
from=-3
|
557 |
+
activation=linear
|
558 |
+
|
559 |
+
[convolutional]
|
560 |
+
batch_normalize=1
|
561 |
+
filters=256
|
562 |
+
size=1
|
563 |
+
stride=1
|
564 |
+
pad=1
|
565 |
+
activation=swish
|
566 |
+
|
567 |
+
[convolutional]
|
568 |
+
batch_normalize=1
|
569 |
+
filters=256
|
570 |
+
size=3
|
571 |
+
stride=1
|
572 |
+
pad=1
|
573 |
+
activation=swish
|
574 |
+
|
575 |
+
[shortcut]
|
576 |
+
from=-3
|
577 |
+
activation=linear
|
578 |
+
|
579 |
+
[convolutional]
|
580 |
+
batch_normalize=1
|
581 |
+
filters=256
|
582 |
+
size=1
|
583 |
+
stride=1
|
584 |
+
pad=1
|
585 |
+
activation=swish
|
586 |
+
|
587 |
+
[convolutional]
|
588 |
+
batch_normalize=1
|
589 |
+
filters=256
|
590 |
+
size=3
|
591 |
+
stride=1
|
592 |
+
pad=1
|
593 |
+
activation=swish
|
594 |
+
|
595 |
+
[shortcut]
|
596 |
+
from=-3
|
597 |
+
activation=linear
|
598 |
+
|
599 |
+
# Transition first
|
600 |
+
|
601 |
+
[convolutional]
|
602 |
+
batch_normalize=1
|
603 |
+
filters=256
|
604 |
+
size=1
|
605 |
+
stride=1
|
606 |
+
pad=1
|
607 |
+
activation=swish
|
608 |
+
|
609 |
+
# Merge [-1 -(3k+4)]
|
610 |
+
|
611 |
+
[route]
|
612 |
+
layers = -1,-28
|
613 |
+
|
614 |
+
# Transition last
|
615 |
+
|
616 |
+
# 79 (previous+7+3k)
|
617 |
+
[convolutional]
|
618 |
+
batch_normalize=1
|
619 |
+
filters=512
|
620 |
+
size=1
|
621 |
+
stride=1
|
622 |
+
pad=1
|
623 |
+
activation=swish
|
624 |
+
|
625 |
+
# P5
|
626 |
+
|
627 |
+
# Downsample
|
628 |
+
|
629 |
+
[convolutional]
|
630 |
+
batch_normalize=1
|
631 |
+
filters=1024
|
632 |
+
size=3
|
633 |
+
stride=2
|
634 |
+
pad=1
|
635 |
+
activation=swish
|
636 |
+
|
637 |
+
# Split
|
638 |
+
|
639 |
+
[convolutional]
|
640 |
+
batch_normalize=1
|
641 |
+
filters=512
|
642 |
+
size=1
|
643 |
+
stride=1
|
644 |
+
pad=1
|
645 |
+
activation=swish
|
646 |
+
|
647 |
+
[route]
|
648 |
+
layers = -2
|
649 |
+
|
650 |
+
[convolutional]
|
651 |
+
batch_normalize=1
|
652 |
+
filters=512
|
653 |
+
size=1
|
654 |
+
stride=1
|
655 |
+
pad=1
|
656 |
+
activation=swish
|
657 |
+
|
658 |
+
# Residual Block
|
659 |
+
|
660 |
+
[convolutional]
|
661 |
+
batch_normalize=1
|
662 |
+
filters=512
|
663 |
+
size=1
|
664 |
+
stride=1
|
665 |
+
pad=1
|
666 |
+
activation=swish
|
667 |
+
|
668 |
+
[convolutional]
|
669 |
+
batch_normalize=1
|
670 |
+
filters=512
|
671 |
+
size=3
|
672 |
+
stride=1
|
673 |
+
pad=1
|
674 |
+
activation=swish
|
675 |
+
|
676 |
+
[shortcut]
|
677 |
+
from=-3
|
678 |
+
activation=linear
|
679 |
+
|
680 |
+
[convolutional]
|
681 |
+
batch_normalize=1
|
682 |
+
filters=512
|
683 |
+
size=1
|
684 |
+
stride=1
|
685 |
+
pad=1
|
686 |
+
activation=swish
|
687 |
+
|
688 |
+
[convolutional]
|
689 |
+
batch_normalize=1
|
690 |
+
filters=512
|
691 |
+
size=3
|
692 |
+
stride=1
|
693 |
+
pad=1
|
694 |
+
activation=swish
|
695 |
+
|
696 |
+
[shortcut]
|
697 |
+
from=-3
|
698 |
+
activation=linear
|
699 |
+
|
700 |
+
[convolutional]
|
701 |
+
batch_normalize=1
|
702 |
+
filters=512
|
703 |
+
size=1
|
704 |
+
stride=1
|
705 |
+
pad=1
|
706 |
+
activation=swish
|
707 |
+
|
708 |
+
[convolutional]
|
709 |
+
batch_normalize=1
|
710 |
+
filters=512
|
711 |
+
size=3
|
712 |
+
stride=1
|
713 |
+
pad=1
|
714 |
+
activation=swish
|
715 |
+
|
716 |
+
[shortcut]
|
717 |
+
from=-3
|
718 |
+
activation=linear
|
719 |
+
|
720 |
+
[convolutional]
|
721 |
+
batch_normalize=1
|
722 |
+
filters=512
|
723 |
+
size=1
|
724 |
+
stride=1
|
725 |
+
pad=1
|
726 |
+
activation=swish
|
727 |
+
|
728 |
+
[convolutional]
|
729 |
+
batch_normalize=1
|
730 |
+
filters=512
|
731 |
+
size=3
|
732 |
+
stride=1
|
733 |
+
pad=1
|
734 |
+
activation=swish
|
735 |
+
|
736 |
+
[shortcut]
|
737 |
+
from=-3
|
738 |
+
activation=linear
|
739 |
+
|
740 |
+
# Transition first
|
741 |
+
|
742 |
+
[convolutional]
|
743 |
+
batch_normalize=1
|
744 |
+
filters=512
|
745 |
+
size=1
|
746 |
+
stride=1
|
747 |
+
pad=1
|
748 |
+
activation=swish
|
749 |
+
|
750 |
+
# Merge [-1 -(3k+4)]
|
751 |
+
|
752 |
+
[route]
|
753 |
+
layers = -1,-16
|
754 |
+
|
755 |
+
# Transition last
|
756 |
+
|
757 |
+
# 98 (previous+7+3k)
|
758 |
+
[convolutional]
|
759 |
+
batch_normalize=1
|
760 |
+
filters=1024
|
761 |
+
size=1
|
762 |
+
stride=1
|
763 |
+
pad=1
|
764 |
+
activation=swish
|
765 |
+
|
766 |
+
# ============ End of Backbone ============ #
|
767 |
+
|
768 |
+
# ============ Neck ============ #
|
769 |
+
|
770 |
+
# CSPSPP
|
771 |
+
|
772 |
+
[convolutional]
|
773 |
+
batch_normalize=1
|
774 |
+
filters=512
|
775 |
+
size=1
|
776 |
+
stride=1
|
777 |
+
pad=1
|
778 |
+
activation=swish
|
779 |
+
|
780 |
+
[route]
|
781 |
+
layers = -2
|
782 |
+
|
783 |
+
[convolutional]
|
784 |
+
batch_normalize=1
|
785 |
+
filters=512
|
786 |
+
size=1
|
787 |
+
stride=1
|
788 |
+
pad=1
|
789 |
+
activation=swish
|
790 |
+
|
791 |
+
[convolutional]
|
792 |
+
batch_normalize=1
|
793 |
+
size=3
|
794 |
+
stride=1
|
795 |
+
pad=1
|
796 |
+
filters=512
|
797 |
+
activation=swish
|
798 |
+
|
799 |
+
[convolutional]
|
800 |
+
batch_normalize=1
|
801 |
+
filters=512
|
802 |
+
size=1
|
803 |
+
stride=1
|
804 |
+
pad=1
|
805 |
+
activation=swish
|
806 |
+
|
807 |
+
### SPP ###
|
808 |
+
[maxpool]
|
809 |
+
stride=1
|
810 |
+
size=5
|
811 |
+
|
812 |
+
[route]
|
813 |
+
layers=-2
|
814 |
+
|
815 |
+
[maxpool]
|
816 |
+
stride=1
|
817 |
+
size=9
|
818 |
+
|
819 |
+
[route]
|
820 |
+
layers=-4
|
821 |
+
|
822 |
+
[maxpool]
|
823 |
+
stride=1
|
824 |
+
size=13
|
825 |
+
|
826 |
+
[route]
|
827 |
+
layers=-1,-3,-5,-6
|
828 |
+
### End SPP ###
|
829 |
+
|
830 |
+
[convolutional]
|
831 |
+
batch_normalize=1
|
832 |
+
filters=512
|
833 |
+
size=1
|
834 |
+
stride=1
|
835 |
+
pad=1
|
836 |
+
activation=swish
|
837 |
+
|
838 |
+
[convolutional]
|
839 |
+
batch_normalize=1
|
840 |
+
size=3
|
841 |
+
stride=1
|
842 |
+
pad=1
|
843 |
+
filters=512
|
844 |
+
activation=swish
|
845 |
+
|
846 |
+
[route]
|
847 |
+
layers = -1, -13
|
848 |
+
|
849 |
+
# 113 (previous+6+5+2k)
|
850 |
+
[convolutional]
|
851 |
+
batch_normalize=1
|
852 |
+
filters=512
|
853 |
+
size=1
|
854 |
+
stride=1
|
855 |
+
pad=1
|
856 |
+
activation=swish
|
857 |
+
|
858 |
+
# End of CSPSPP
|
859 |
+
|
860 |
+
|
861 |
+
# FPN-4
|
862 |
+
|
863 |
+
[convolutional]
|
864 |
+
batch_normalize=1
|
865 |
+
filters=256
|
866 |
+
size=1
|
867 |
+
stride=1
|
868 |
+
pad=1
|
869 |
+
activation=swish
|
870 |
+
|
871 |
+
[upsample]
|
872 |
+
stride=2
|
873 |
+
|
874 |
+
[route]
|
875 |
+
layers = 79
|
876 |
+
|
877 |
+
[convolutional]
|
878 |
+
batch_normalize=1
|
879 |
+
filters=256
|
880 |
+
size=1
|
881 |
+
stride=1
|
882 |
+
pad=1
|
883 |
+
activation=swish
|
884 |
+
|
885 |
+
[route]
|
886 |
+
layers = -1, -3
|
887 |
+
|
888 |
+
[convolutional]
|
889 |
+
batch_normalize=1
|
890 |
+
filters=256
|
891 |
+
size=1
|
892 |
+
stride=1
|
893 |
+
pad=1
|
894 |
+
activation=swish
|
895 |
+
|
896 |
+
# Split
|
897 |
+
|
898 |
+
[convolutional]
|
899 |
+
batch_normalize=1
|
900 |
+
filters=256
|
901 |
+
size=1
|
902 |
+
stride=1
|
903 |
+
pad=1
|
904 |
+
activation=swish
|
905 |
+
|
906 |
+
[route]
|
907 |
+
layers = -2
|
908 |
+
|
909 |
+
# Plain Block
|
910 |
+
|
911 |
+
[convolutional]
|
912 |
+
batch_normalize=1
|
913 |
+
filters=256
|
914 |
+
size=1
|
915 |
+
stride=1
|
916 |
+
pad=1
|
917 |
+
activation=swish
|
918 |
+
|
919 |
+
[convolutional]
|
920 |
+
batch_normalize=1
|
921 |
+
size=3
|
922 |
+
stride=1
|
923 |
+
pad=1
|
924 |
+
filters=256
|
925 |
+
activation=swish
|
926 |
+
|
927 |
+
[convolutional]
|
928 |
+
batch_normalize=1
|
929 |
+
filters=256
|
930 |
+
size=1
|
931 |
+
stride=1
|
932 |
+
pad=1
|
933 |
+
activation=swish
|
934 |
+
|
935 |
+
[convolutional]
|
936 |
+
batch_normalize=1
|
937 |
+
size=3
|
938 |
+
stride=1
|
939 |
+
pad=1
|
940 |
+
filters=256
|
941 |
+
activation=swish
|
942 |
+
|
943 |
+
# Merge [-1, -(2k+2)]
|
944 |
+
|
945 |
+
[route]
|
946 |
+
layers = -1, -6
|
947 |
+
|
948 |
+
# Transition last
|
949 |
+
|
950 |
+
# 127 (previous+6+4+2k)
|
951 |
+
[convolutional]
|
952 |
+
batch_normalize=1
|
953 |
+
filters=256
|
954 |
+
size=1
|
955 |
+
stride=1
|
956 |
+
pad=1
|
957 |
+
activation=swish
|
958 |
+
|
959 |
+
|
960 |
+
# FPN-3
|
961 |
+
|
962 |
+
[convolutional]
|
963 |
+
batch_normalize=1
|
964 |
+
filters=128
|
965 |
+
size=1
|
966 |
+
stride=1
|
967 |
+
pad=1
|
968 |
+
activation=swish
|
969 |
+
|
970 |
+
[upsample]
|
971 |
+
stride=2
|
972 |
+
|
973 |
+
[route]
|
974 |
+
layers = 48
|
975 |
+
|
976 |
+
[convolutional]
|
977 |
+
batch_normalize=1
|
978 |
+
filters=128
|
979 |
+
size=1
|
980 |
+
stride=1
|
981 |
+
pad=1
|
982 |
+
activation=swish
|
983 |
+
|
984 |
+
[route]
|
985 |
+
layers = -1, -3
|
986 |
+
|
987 |
+
[convolutional]
|
988 |
+
batch_normalize=1
|
989 |
+
filters=128
|
990 |
+
size=1
|
991 |
+
stride=1
|
992 |
+
pad=1
|
993 |
+
activation=swish
|
994 |
+
|
995 |
+
# Split
|
996 |
+
|
997 |
+
[convolutional]
|
998 |
+
batch_normalize=1
|
999 |
+
filters=128
|
1000 |
+
size=1
|
1001 |
+
stride=1
|
1002 |
+
pad=1
|
1003 |
+
activation=swish
|
1004 |
+
|
1005 |
+
[route]
|
1006 |
+
layers = -2
|
1007 |
+
|
1008 |
+
# Plain Block
|
1009 |
+
|
1010 |
+
[convolutional]
|
1011 |
+
batch_normalize=1
|
1012 |
+
filters=128
|
1013 |
+
size=1
|
1014 |
+
stride=1
|
1015 |
+
pad=1
|
1016 |
+
activation=swish
|
1017 |
+
|
1018 |
+
[convolutional]
|
1019 |
+
batch_normalize=1
|
1020 |
+
size=3
|
1021 |
+
stride=1
|
1022 |
+
pad=1
|
1023 |
+
filters=128
|
1024 |
+
activation=swish
|
1025 |
+
|
1026 |
+
[convolutional]
|
1027 |
+
batch_normalize=1
|
1028 |
+
filters=128
|
1029 |
+
size=1
|
1030 |
+
stride=1
|
1031 |
+
pad=1
|
1032 |
+
activation=swish
|
1033 |
+
|
1034 |
+
[convolutional]
|
1035 |
+
batch_normalize=1
|
1036 |
+
size=3
|
1037 |
+
stride=1
|
1038 |
+
pad=1
|
1039 |
+
filters=128
|
1040 |
+
activation=swish
|
1041 |
+
|
1042 |
+
# Merge [-1, -(2k+2)]
|
1043 |
+
|
1044 |
+
[route]
|
1045 |
+
layers = -1, -6
|
1046 |
+
|
1047 |
+
# Transition last
|
1048 |
+
|
1049 |
+
# 141 (previous+6+4+2k)
|
1050 |
+
[convolutional]
|
1051 |
+
batch_normalize=1
|
1052 |
+
filters=128
|
1053 |
+
size=1
|
1054 |
+
stride=1
|
1055 |
+
pad=1
|
1056 |
+
activation=swish
|
1057 |
+
|
1058 |
+
|
1059 |
+
# PAN-4
|
1060 |
+
|
1061 |
+
[convolutional]
|
1062 |
+
batch_normalize=1
|
1063 |
+
size=3
|
1064 |
+
stride=2
|
1065 |
+
pad=1
|
1066 |
+
filters=256
|
1067 |
+
activation=swish
|
1068 |
+
|
1069 |
+
[route]
|
1070 |
+
layers = -1, 127
|
1071 |
+
|
1072 |
+
[convolutional]
|
1073 |
+
batch_normalize=1
|
1074 |
+
filters=256
|
1075 |
+
size=1
|
1076 |
+
stride=1
|
1077 |
+
pad=1
|
1078 |
+
activation=swish
|
1079 |
+
|
1080 |
+
# Split
|
1081 |
+
|
1082 |
+
[convolutional]
|
1083 |
+
batch_normalize=1
|
1084 |
+
filters=256
|
1085 |
+
size=1
|
1086 |
+
stride=1
|
1087 |
+
pad=1
|
1088 |
+
activation=swish
|
1089 |
+
|
1090 |
+
[route]
|
1091 |
+
layers = -2
|
1092 |
+
|
1093 |
+
# Plain Block
|
1094 |
+
|
1095 |
+
[convolutional]
|
1096 |
+
batch_normalize=1
|
1097 |
+
filters=256
|
1098 |
+
size=1
|
1099 |
+
stride=1
|
1100 |
+
pad=1
|
1101 |
+
activation=swish
|
1102 |
+
|
1103 |
+
[convolutional]
|
1104 |
+
batch_normalize=1
|
1105 |
+
size=3
|
1106 |
+
stride=1
|
1107 |
+
pad=1
|
1108 |
+
filters=256
|
1109 |
+
activation=swish
|
1110 |
+
|
1111 |
+
[convolutional]
|
1112 |
+
batch_normalize=1
|
1113 |
+
filters=256
|
1114 |
+
size=1
|
1115 |
+
stride=1
|
1116 |
+
pad=1
|
1117 |
+
activation=swish
|
1118 |
+
|
1119 |
+
[convolutional]
|
1120 |
+
batch_normalize=1
|
1121 |
+
size=3
|
1122 |
+
stride=1
|
1123 |
+
pad=1
|
1124 |
+
filters=256
|
1125 |
+
activation=swish
|
1126 |
+
|
1127 |
+
[route]
|
1128 |
+
layers = -1,-6
|
1129 |
+
|
1130 |
+
# Transition last
|
1131 |
+
|
1132 |
+
# 152 (previous+3+4+2k)
|
1133 |
+
[convolutional]
|
1134 |
+
batch_normalize=1
|
1135 |
+
filters=256
|
1136 |
+
size=1
|
1137 |
+
stride=1
|
1138 |
+
pad=1
|
1139 |
+
activation=swish
|
1140 |
+
|
1141 |
+
|
1142 |
+
# PAN-5
|
1143 |
+
|
1144 |
+
[convolutional]
|
1145 |
+
batch_normalize=1
|
1146 |
+
size=3
|
1147 |
+
stride=2
|
1148 |
+
pad=1
|
1149 |
+
filters=512
|
1150 |
+
activation=swish
|
1151 |
+
|
1152 |
+
[route]
|
1153 |
+
layers = -1, 113
|
1154 |
+
|
1155 |
+
[convolutional]
|
1156 |
+
batch_normalize=1
|
1157 |
+
filters=512
|
1158 |
+
size=1
|
1159 |
+
stride=1
|
1160 |
+
pad=1
|
1161 |
+
activation=swish
|
1162 |
+
|
1163 |
+
# Split
|
1164 |
+
|
1165 |
+
[convolutional]
|
1166 |
+
batch_normalize=1
|
1167 |
+
filters=512
|
1168 |
+
size=1
|
1169 |
+
stride=1
|
1170 |
+
pad=1
|
1171 |
+
activation=swish
|
1172 |
+
|
1173 |
+
[route]
|
1174 |
+
layers = -2
|
1175 |
+
|
1176 |
+
# Plain Block
|
1177 |
+
|
1178 |
+
[convolutional]
|
1179 |
+
batch_normalize=1
|
1180 |
+
filters=512
|
1181 |
+
size=1
|
1182 |
+
stride=1
|
1183 |
+
pad=1
|
1184 |
+
activation=swish
|
1185 |
+
|
1186 |
+
[convolutional]
|
1187 |
+
batch_normalize=1
|
1188 |
+
size=3
|
1189 |
+
stride=1
|
1190 |
+
pad=1
|
1191 |
+
filters=512
|
1192 |
+
activation=swish
|
1193 |
+
|
1194 |
+
[convolutional]
|
1195 |
+
batch_normalize=1
|
1196 |
+
filters=512
|
1197 |
+
size=1
|
1198 |
+
stride=1
|
1199 |
+
pad=1
|
1200 |
+
activation=swish
|
1201 |
+
|
1202 |
+
[convolutional]
|
1203 |
+
batch_normalize=1
|
1204 |
+
size=3
|
1205 |
+
stride=1
|
1206 |
+
pad=1
|
1207 |
+
filters=512
|
1208 |
+
activation=swish
|
1209 |
+
|
1210 |
+
[route]
|
1211 |
+
layers = -1,-6
|
1212 |
+
|
1213 |
+
# Transition last
|
1214 |
+
|
1215 |
+
# 163 (previous+3+4+2k)
|
1216 |
+
[convolutional]
|
1217 |
+
batch_normalize=1
|
1218 |
+
filters=512
|
1219 |
+
size=1
|
1220 |
+
stride=1
|
1221 |
+
pad=1
|
1222 |
+
activation=swish
|
1223 |
+
|
1224 |
+
# ============ End of Neck ============ #
|
1225 |
+
|
1226 |
+
# ============ Head ============ #
|
1227 |
+
|
1228 |
+
# YOLO-3
|
1229 |
+
|
1230 |
+
[route]
|
1231 |
+
layers = 141
|
1232 |
+
|
1233 |
+
[convolutional]
|
1234 |
+
batch_normalize=1
|
1235 |
+
size=3
|
1236 |
+
stride=1
|
1237 |
+
pad=1
|
1238 |
+
filters=256
|
1239 |
+
activation=swish
|
1240 |
+
|
1241 |
+
[convolutional]
|
1242 |
+
size=1
|
1243 |
+
stride=1
|
1244 |
+
pad=1
|
1245 |
+
filters=255
|
1246 |
+
activation=logistic
|
1247 |
+
|
1248 |
+
[yolo]
|
1249 |
+
mask = 0,1,2
|
1250 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1251 |
+
classes=80
|
1252 |
+
num=9
|
1253 |
+
jitter=.1
|
1254 |
+
scale_x_y = 2.0
|
1255 |
+
objectness_smooth=1
|
1256 |
+
ignore_thresh = .7
|
1257 |
+
truth_thresh = 1
|
1258 |
+
#random=1
|
1259 |
+
resize=1.5
|
1260 |
+
iou_thresh=0.2
|
1261 |
+
iou_normalizer=0.05
|
1262 |
+
cls_normalizer=0.5
|
1263 |
+
obj_normalizer=0.4
|
1264 |
+
iou_loss=ciou
|
1265 |
+
nms_kind=diounms
|
1266 |
+
beta_nms=0.6
|
1267 |
+
new_coords=1
|
1268 |
+
max_delta=2
|
1269 |
+
|
1270 |
+
|
1271 |
+
# YOLO-4
|
1272 |
+
|
1273 |
+
[route]
|
1274 |
+
layers = 152
|
1275 |
+
|
1276 |
+
[convolutional]
|
1277 |
+
batch_normalize=1
|
1278 |
+
size=3
|
1279 |
+
stride=1
|
1280 |
+
pad=1
|
1281 |
+
filters=512
|
1282 |
+
activation=swish
|
1283 |
+
|
1284 |
+
[convolutional]
|
1285 |
+
size=1
|
1286 |
+
stride=1
|
1287 |
+
pad=1
|
1288 |
+
filters=255
|
1289 |
+
activation=logistic
|
1290 |
+
|
1291 |
+
[yolo]
|
1292 |
+
mask = 3,4,5
|
1293 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1294 |
+
classes=80
|
1295 |
+
num=9
|
1296 |
+
jitter=.1
|
1297 |
+
scale_x_y = 2.0
|
1298 |
+
objectness_smooth=1
|
1299 |
+
ignore_thresh = .7
|
1300 |
+
truth_thresh = 1
|
1301 |
+
#random=1
|
1302 |
+
resize=1.5
|
1303 |
+
iou_thresh=0.2
|
1304 |
+
iou_normalizer=0.05
|
1305 |
+
cls_normalizer=0.5
|
1306 |
+
obj_normalizer=0.4
|
1307 |
+
iou_loss=ciou
|
1308 |
+
nms_kind=diounms
|
1309 |
+
beta_nms=0.6
|
1310 |
+
new_coords=1
|
1311 |
+
max_delta=2
|
1312 |
+
|
1313 |
+
|
1314 |
+
# YOLO-5
|
1315 |
+
|
1316 |
+
[route]
|
1317 |
+
layers = 163
|
1318 |
+
|
1319 |
+
[convolutional]
|
1320 |
+
batch_normalize=1
|
1321 |
+
size=3
|
1322 |
+
stride=1
|
1323 |
+
pad=1
|
1324 |
+
filters=1024
|
1325 |
+
activation=swish
|
1326 |
+
|
1327 |
+
[convolutional]
|
1328 |
+
size=1
|
1329 |
+
stride=1
|
1330 |
+
pad=1
|
1331 |
+
filters=255
|
1332 |
+
activation=logistic
|
1333 |
+
|
1334 |
+
[yolo]
|
1335 |
+
mask = 6,7,8
|
1336 |
+
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
|
1337 |
+
classes=80
|
1338 |
+
num=9
|
1339 |
+
jitter=.1
|
1340 |
+
scale_x_y = 2.0
|
1341 |
+
objectness_smooth=1
|
1342 |
+
ignore_thresh = .7
|
1343 |
+
truth_thresh = 1
|
1344 |
+
#random=1
|
1345 |
+
resize=1.5
|
1346 |
+
iou_thresh=0.2
|
1347 |
+
iou_normalizer=0.05
|
1348 |
+
cls_normalizer=0.5
|
1349 |
+
obj_normalizer=0.4
|
1350 |
+
iou_loss=ciou
|
1351 |
+
nms_kind=diounms
|
1352 |
+
beta_nms=0.6
|
1353 |
+
new_coords=1
|
1354 |
+
max_delta=2
|
darknet/new_layers.md
ADDED
@@ -0,0 +1,329 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
![Implicit Modeling](https://github.com/WongKinYiu/yolor/blob/main/figure/implicit_modeling.png)
|
2 |
+
|
3 |
+
### 1. silence layer
|
4 |
+
|
5 |
+
Usage:
|
6 |
+
|
7 |
+
```
|
8 |
+
[silence]
|
9 |
+
```
|
10 |
+
|
11 |
+
PyTorch code:
|
12 |
+
|
13 |
+
``` python
|
14 |
+
class Silence(nn.Module):
|
15 |
+
def __init__(self):
|
16 |
+
super(Silence, self).__init__()
|
17 |
+
def forward(self, x):
|
18 |
+
return x
|
19 |
+
```
|
20 |
+
|
21 |
+
|
22 |
+
### 2. implicit_add layer
|
23 |
+
|
24 |
+
Usage:
|
25 |
+
|
26 |
+
```
|
27 |
+
[implicit_add]
|
28 |
+
filters=128
|
29 |
+
```
|
30 |
+
|
31 |
+
PyTorch code:
|
32 |
+
|
33 |
+
``` python
|
34 |
+
class ImplicitA(nn.Module):
|
35 |
+
def __init__(self, channel):
|
36 |
+
super(ImplicitA, self).__init__()
|
37 |
+
self.channel = channel
|
38 |
+
self.implicit = nn.Parameter(torch.zeros(1, channel, 1, 1))
|
39 |
+
nn.init.normal_(self.implicit, std=.02)
|
40 |
+
|
41 |
+
def forward(self):
|
42 |
+
return self.implicit
|
43 |
+
```
|
44 |
+
|
45 |
+
|
46 |
+
### 3. shift_channels layer
|
47 |
+
|
48 |
+
Usage:
|
49 |
+
|
50 |
+
```
|
51 |
+
[shift_channels]
|
52 |
+
from=101
|
53 |
+
```
|
54 |
+
|
55 |
+
PyTorch code:
|
56 |
+
|
57 |
+
``` python
|
58 |
+
class ShiftChannel(nn.Module):
|
59 |
+
def __init__(self, layers):
|
60 |
+
super(ShiftChannel, self).__init__()
|
61 |
+
self.layers = layers # layer indices
|
62 |
+
|
63 |
+
def forward(self, x, outputs):
|
64 |
+
a = outputs[self.layers[0]]
|
65 |
+
return a.expand_as(x) + x
|
66 |
+
```
|
67 |
+
|
68 |
+
|
69 |
+
### 4. implicit_mul layer
|
70 |
+
|
71 |
+
Usage:
|
72 |
+
|
73 |
+
```
|
74 |
+
[implicit_mul]
|
75 |
+
filters=128
|
76 |
+
```
|
77 |
+
|
78 |
+
PyTorch code:
|
79 |
+
|
80 |
+
``` python
|
81 |
+
class ImplicitM(nn.Module):
|
82 |
+
def __init__(self, channel):
|
83 |
+
super(ImplicitM, self).__init__()
|
84 |
+
self.channel = channel
|
85 |
+
self.implicit = nn.Parameter(torch.ones(1, channel, 1, 1))
|
86 |
+
nn.init.normal_(self.implicit, mean=1., std=.02)
|
87 |
+
|
88 |
+
def forward(self):
|
89 |
+
return self.implicit
|
90 |
+
```
|
91 |
+
|
92 |
+
|
93 |
+
### 5. control_channels layer
|
94 |
+
|
95 |
+
Usage:
|
96 |
+
|
97 |
+
```
|
98 |
+
[control_channels]
|
99 |
+
from=101
|
100 |
+
```
|
101 |
+
|
102 |
+
PyTorch code:
|
103 |
+
|
104 |
+
``` python
|
105 |
+
class ControlChannel(nn.Module):
|
106 |
+
def __init__(self, layers):
|
107 |
+
super(ControlChannel, self).__init__()
|
108 |
+
self.layers = layers # layer indices
|
109 |
+
|
110 |
+
def forward(self, x, outputs):
|
111 |
+
a = outputs[self.layers[0]]
|
112 |
+
return a.expand_as(x) * x
|
113 |
+
```
|
114 |
+
|
115 |
+
|
116 |
+
### 6. implicit_cat layer
|
117 |
+
|
118 |
+
Usage:
|
119 |
+
|
120 |
+
```
|
121 |
+
[implicit_cat]
|
122 |
+
filters=128
|
123 |
+
```
|
124 |
+
|
125 |
+
PyTorch code: (same as ImplicitA)
|
126 |
+
|
127 |
+
``` python
|
128 |
+
class ImplicitC(nn.Module):
|
129 |
+
def __init__(self, channel):
|
130 |
+
super(ImplicitC, self).__init__()
|
131 |
+
self.channel = channel
|
132 |
+
self.implicit = nn.Parameter(torch.zeros(1, channel, 1, 1))
|
133 |
+
nn.init.normal_(self.implicit, std=.02)
|
134 |
+
|
135 |
+
def forward(self):
|
136 |
+
return self.implicit
|
137 |
+
```
|
138 |
+
|
139 |
+
|
140 |
+
### 7. alternate_channels layer
|
141 |
+
|
142 |
+
Usage:
|
143 |
+
|
144 |
+
```
|
145 |
+
[alternate_channels]
|
146 |
+
from=101
|
147 |
+
```
|
148 |
+
|
149 |
+
PyTorch code:
|
150 |
+
|
151 |
+
``` python
|
152 |
+
class AlternateChannel(nn.Module):
|
153 |
+
def __init__(self, layers):
|
154 |
+
super(AlternateChannel, self).__init__()
|
155 |
+
self.layers = layers # layer indices
|
156 |
+
|
157 |
+
def forward(self, x, outputs):
|
158 |
+
a = outputs[self.layers[0]]
|
159 |
+
return torch.cat([a.expand_as(x), x], dim=1)
|
160 |
+
```
|
161 |
+
|
162 |
+
|
163 |
+
### 8. implicit_add_2d layer
|
164 |
+
|
165 |
+
Usage:
|
166 |
+
|
167 |
+
```
|
168 |
+
[implicit_add_2d]
|
169 |
+
filters=128
|
170 |
+
atoms=128
|
171 |
+
```
|
172 |
+
|
173 |
+
PyTorch code:
|
174 |
+
|
175 |
+
``` python
|
176 |
+
class Implicit2DA(nn.Module):
|
177 |
+
def __init__(self, atom, channel):
|
178 |
+
super(Implicit2DA, self).__init__()
|
179 |
+
self.channel = channel
|
180 |
+
self.implicit = nn.Parameter(torch.zeros(1, atom, channel, 1))
|
181 |
+
nn.init.normal_(self.implicit, std=.02)
|
182 |
+
|
183 |
+
def forward(self):
|
184 |
+
return self.implicit
|
185 |
+
```
|
186 |
+
|
187 |
+
|
188 |
+
### 9. shift_channels_2d layer
|
189 |
+
|
190 |
+
Usage:
|
191 |
+
|
192 |
+
```
|
193 |
+
[shift_channels_2d]
|
194 |
+
from=101
|
195 |
+
```
|
196 |
+
|
197 |
+
PyTorch code:
|
198 |
+
|
199 |
+
``` python
|
200 |
+
class ShiftChannel2D(nn.Module):
|
201 |
+
def __init__(self, layers):
|
202 |
+
super(ShiftChannel2D, self).__init__()
|
203 |
+
self.layers = layers # layer indices
|
204 |
+
|
205 |
+
def forward(self, x, outputs):
|
206 |
+
a = outputs[self.layers[0]].view(1,-1,1,1)
|
207 |
+
return a.expand_as(x) + x
|
208 |
+
```
|
209 |
+
|
210 |
+
|
211 |
+
### 10. implicit_mul_2d layer
|
212 |
+
|
213 |
+
Usage:
|
214 |
+
|
215 |
+
```
|
216 |
+
[implicit_mul_2d]
|
217 |
+
filters=128
|
218 |
+
atoms=128
|
219 |
+
```
|
220 |
+
|
221 |
+
PyTorch code:
|
222 |
+
|
223 |
+
``` python
|
224 |
+
class Implicit2DM(nn.Module):
|
225 |
+
def __init__(self, atom, channel):
|
226 |
+
super(Implicit2DM, self).__init__()
|
227 |
+
self.channel = channel
|
228 |
+
self.implicit = nn.Parameter(torch.ones(1, atom, channel, 1))
|
229 |
+
nn.init.normal_(self.implicit, mean=1., std=.02)
|
230 |
+
|
231 |
+
def forward(self):
|
232 |
+
return self.implicit
|
233 |
+
```
|
234 |
+
|
235 |
+
|
236 |
+
### 11. control_channels_2d layer
|
237 |
+
|
238 |
+
Usage:
|
239 |
+
|
240 |
+
```
|
241 |
+
[control_channels_2d]
|
242 |
+
from=101
|
243 |
+
```
|
244 |
+
|
245 |
+
PyTorch code:
|
246 |
+
|
247 |
+
``` python
|
248 |
+
class ControlChannel2D(nn.Module):
|
249 |
+
def __init__(self, layers):
|
250 |
+
super(ControlChannel2D, self).__init__()
|
251 |
+
self.layers = layers # layer indices
|
252 |
+
|
253 |
+
def forward(self, x, outputs):
|
254 |
+
a = outputs[self.layers[0]].view(1,-1,1,1)
|
255 |
+
return a.expand_as(x) * x
|
256 |
+
```
|
257 |
+
|
258 |
+
|
259 |
+
### 12. implicit_cat_2d layer
|
260 |
+
|
261 |
+
Usage:
|
262 |
+
|
263 |
+
```
|
264 |
+
[implicit_cat_2d]
|
265 |
+
filters=128
|
266 |
+
atoms=128
|
267 |
+
```
|
268 |
+
|
269 |
+
PyTorch code: (same as Implicit2DA)
|
270 |
+
|
271 |
+
``` python
|
272 |
+
class Implicit2DC(nn.Module):
|
273 |
+
def __init__(self, atom, channel):
|
274 |
+
super(Implicit2DC, self).__init__()
|
275 |
+
self.channel = channel
|
276 |
+
self.implicit = nn.Parameter(torch.zeros(1, atom, channel, 1))
|
277 |
+
nn.init.normal_(self.implicit, std=.02)
|
278 |
+
|
279 |
+
def forward(self):
|
280 |
+
return self.implicit
|
281 |
+
```
|
282 |
+
|
283 |
+
|
284 |
+
### 13. alternate_channels_2d layer
|
285 |
+
|
286 |
+
Usage:
|
287 |
+
|
288 |
+
```
|
289 |
+
[alternate_channels_2d]
|
290 |
+
from=101
|
291 |
+
```
|
292 |
+
|
293 |
+
PyTorch code:
|
294 |
+
|
295 |
+
``` python
|
296 |
+
class AlternateChannel2D(nn.Module):
|
297 |
+
def __init__(self, layers):
|
298 |
+
super(AlternateChannel2D, self).__init__()
|
299 |
+
self.layers = layers # layer indices
|
300 |
+
|
301 |
+
def forward(self, x, outputs):
|
302 |
+
a = outputs[self.layers[0]].view(1,-1,1,1)
|
303 |
+
return torch.cat([a.expand_as(x), x], dim=1)
|
304 |
+
```
|
305 |
+
|
306 |
+
|
307 |
+
### 14. dwt layer
|
308 |
+
|
309 |
+
Usage:
|
310 |
+
|
311 |
+
```
|
312 |
+
[dwt]
|
313 |
+
```
|
314 |
+
|
315 |
+
PyTorch code:
|
316 |
+
|
317 |
+
``` python
|
318 |
+
# https://github.com/fbcotter/pytorch_wavelets
|
319 |
+
from pytorch_wavelets import DWTForward, DWTInverse
|
320 |
+
class DWT(nn.Module):
|
321 |
+
def __init__(self):
|
322 |
+
super(DWT, self).__init__()
|
323 |
+
self.xfm = DWTForward(J=1, wave='db1', mode='zero')
|
324 |
+
|
325 |
+
def forward(self, x):
|
326 |
+
b,c,w,h = x.shape
|
327 |
+
yl, yh = self.xfm(x)
|
328 |
+
return torch.cat([yl/2., yh[0].view(b,-1,w//2,h//2)/2.+.5], 1)
|
329 |
+
```
|
data/coco.names
ADDED
@@ -0,0 +1,80 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
person
|
2 |
+
bicycle
|
3 |
+
car
|
4 |
+
motorcycle
|
5 |
+
airplane
|
6 |
+
bus
|
7 |
+
train
|
8 |
+
truck
|
9 |
+
boat
|
10 |
+
traffic light
|
11 |
+
fire hydrant
|
12 |
+
stop sign
|
13 |
+
parking meter
|
14 |
+
bench
|
15 |
+
bird
|
16 |
+
cat
|
17 |
+
dog
|
18 |
+
horse
|
19 |
+
sheep
|
20 |
+
cow
|
21 |
+
elephant
|
22 |
+
bear
|
23 |
+
zebra
|
24 |
+
giraffe
|
25 |
+
backpack
|
26 |
+
umbrella
|
27 |
+
handbag
|
28 |
+
tie
|
29 |
+
suitcase
|
30 |
+
frisbee
|
31 |
+
skis
|
32 |
+
snowboard
|
33 |
+
sports ball
|
34 |
+
kite
|
35 |
+
baseball bat
|
36 |
+
baseball glove
|
37 |
+
skateboard
|
38 |
+
surfboard
|
39 |
+
tennis racket
|
40 |
+
bottle
|
41 |
+
wine glass
|
42 |
+
cup
|
43 |
+
fork
|
44 |
+
knife
|
45 |
+
spoon
|
46 |
+
bowl
|
47 |
+
banana
|
48 |
+
apple
|
49 |
+
sandwich
|
50 |
+
orange
|
51 |
+
broccoli
|
52 |
+
carrot
|
53 |
+
hot dog
|
54 |
+
pizza
|
55 |
+
donut
|
56 |
+
cake
|
57 |
+
chair
|
58 |
+
couch
|
59 |
+
potted plant
|
60 |
+
bed
|
61 |
+
dining table
|
62 |
+
toilet
|
63 |
+
tv
|
64 |
+
laptop
|
65 |
+
mouse
|
66 |
+
remote
|
67 |
+
keyboard
|
68 |
+
cell phone
|
69 |
+
microwave
|
70 |
+
oven
|
71 |
+
toaster
|
72 |
+
sink
|
73 |
+
refrigerator
|
74 |
+
book
|
75 |
+
clock
|
76 |
+
vase
|
77 |
+
scissors
|
78 |
+
teddy bear
|
79 |
+
hair drier
|
80 |
+
toothbrush
|
data/coco.yaml
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# train and val datasets (image directory or *.txt file with image paths)
|
2 |
+
train: ../coco/train2017.txt # 118k images
|
3 |
+
val: ../coco/val2017.txt # 5k images
|
4 |
+
test: ../coco/test-dev2017.txt # 20k images for submission to https://competitions.codalab.org/competitions/20794
|
5 |
+
|
6 |
+
# number of classes
|
7 |
+
nc: 80
|
8 |
+
|
9 |
+
# class names
|
10 |
+
names: ['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat', 'traffic light',
|
11 |
+
'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow',
|
12 |
+
'elephant', 'bear', 'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee',
|
13 |
+
'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard',
|
14 |
+
'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple',
|
15 |
+
'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch',
|
16 |
+
'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone',
|
17 |
+
'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear',
|
18 |
+
'hair drier', 'toothbrush']
|
data/hyp.finetune.1280.yaml
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
|
2 |
+
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
|
3 |
+
momentum: 0.937 # SGD momentum/Adam beta1
|
4 |
+
weight_decay: 0.0005 # optimizer weight decay 5e-4
|
5 |
+
warmup_epochs: 3.0 # warmup epochs (fractions ok)
|
6 |
+
warmup_momentum: 0.8 # warmup initial momentum
|
7 |
+
warmup_bias_lr: 0.1 # warmup initial bias lr
|
8 |
+
box: 0.05 # box loss gain
|
9 |
+
cls: 0.5 # cls loss gain
|
10 |
+
cls_pw: 1.0 # cls BCELoss positive_weight
|
11 |
+
obj: 1.0 # obj loss gain (scale with pixels)
|
12 |
+
obj_pw: 1.0 # obj BCELoss positive_weight
|
13 |
+
iou_t: 0.20 # IoU training threshold
|
14 |
+
anchor_t: 4.0 # anchor-multiple threshold
|
15 |
+
# anchors: 3 # anchors per output layer (0 to ignore)
|
16 |
+
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
|
17 |
+
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
|
18 |
+
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
|
19 |
+
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
|
20 |
+
degrees: 0.0 # image rotation (+/- deg)
|
21 |
+
translate: 0.5 # image translation (+/- fraction)
|
22 |
+
scale: 0.8 # image scale (+/- gain)
|
23 |
+
shear: 0.0 # image shear (+/- deg)
|
24 |
+
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
|
25 |
+
flipud: 0.0 # image flip up-down (probability)
|
26 |
+
fliplr: 0.5 # image flip left-right (probability)
|
27 |
+
mosaic: 1.0 # image mosaic (probability)
|
28 |
+
mixup: 0.2 # image mixup (probability)
|
data/hyp.scratch.1280.yaml
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
|
2 |
+
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
|
3 |
+
momentum: 0.937 # SGD momentum/Adam beta1
|
4 |
+
weight_decay: 0.0005 # optimizer weight decay 5e-4
|
5 |
+
warmup_epochs: 3.0 # warmup epochs (fractions ok)
|
6 |
+
warmup_momentum: 0.8 # warmup initial momentum
|
7 |
+
warmup_bias_lr: 0.1 # warmup initial bias lr
|
8 |
+
box: 0.05 # box loss gain
|
9 |
+
cls: 0.5 # cls loss gain
|
10 |
+
cls_pw: 1.0 # cls BCELoss positive_weight
|
11 |
+
obj: 1.0 # obj loss gain (scale with pixels)
|
12 |
+
obj_pw: 1.0 # obj BCELoss positive_weight
|
13 |
+
iou_t: 0.20 # IoU training threshold
|
14 |
+
anchor_t: 4.0 # anchor-multiple threshold
|
15 |
+
# anchors: 3 # anchors per output layer (0 to ignore)
|
16 |
+
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
|
17 |
+
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
|
18 |
+
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
|
19 |
+
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
|
20 |
+
degrees: 0.0 # image rotation (+/- deg)
|
21 |
+
translate: 0.5 # image translation (+/- fraction)
|
22 |
+
scale: 0.5 # image scale (+/- gain)
|
23 |
+
shear: 0.0 # image shear (+/- deg)
|
24 |
+
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
|
25 |
+
flipud: 0.0 # image flip up-down (probability)
|
26 |
+
fliplr: 0.5 # image flip left-right (probability)
|
27 |
+
mosaic: 1.0 # image mosaic (probability)
|
28 |
+
mixup: 0.0 # image mixup (probability)
|
data/hyp.scratch.640.yaml
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
|
2 |
+
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
|
3 |
+
momentum: 0.937 # SGD momentum/Adam beta1
|
4 |
+
weight_decay: 0.0005 # optimizer weight decay 5e-4
|
5 |
+
warmup_epochs: 3.0 # warmup epochs (fractions ok)
|
6 |
+
warmup_momentum: 0.8 # warmup initial momentum
|
7 |
+
warmup_bias_lr: 0.1 # warmup initial bias lr
|
8 |
+
box: 0.05 # box loss gain
|
9 |
+
cls: 0.3 # cls loss gain
|
10 |
+
cls_pw: 1.0 # cls BCELoss positive_weight
|
11 |
+
obj: 0.7 # obj loss gain (scale with pixels)
|
12 |
+
obj_pw: 1.0 # obj BCELoss positive_weight
|
13 |
+
iou_t: 0.20 # IoU training threshold
|
14 |
+
anchor_t: 4.0 # anchor-multiple threshold
|
15 |
+
# anchors: 3 # anchors per output layer (0 to ignore)
|
16 |
+
fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
|
17 |
+
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
|
18 |
+
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
|
19 |
+
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
|
20 |
+
degrees: 0.0 # image rotation (+/- deg)
|
21 |
+
translate: 0.1 # image translation (+/- fraction)
|
22 |
+
scale: 0.9 # image scale (+/- gain)
|
23 |
+
shear: 0.0 # image shear (+/- deg)
|
24 |
+
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
|
25 |
+
flipud: 0.0 # image flip up-down (probability)
|
26 |
+
fliplr: 0.5 # image flip left-right (probability)
|
27 |
+
mosaic: 1.0 # image mosaic (probability)
|
28 |
+
mixup: 0.0 # image mixup (probability)
|
figure/implicit_modeling.png
ADDED
figure/performance.png
ADDED
figure/schedule.png
ADDED
figure/unifued_network.png
ADDED
inference/images/horses.jpg
ADDED
inference/output/horses.jpg
ADDED
models/__init__.py
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
|
models/__pycache__/__init__.cpython-37.pyc
ADDED
Binary file (153 Bytes). View file
|
|
models/__pycache__/models.cpython-37.pyc
ADDED
Binary file (20.9 kB). View file
|
|
models/export.py
ADDED
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
|
3 |
+
import torch
|
4 |
+
|
5 |
+
from utils.google_utils import attempt_download
|
6 |
+
|
7 |
+
if __name__ == '__main__':
|
8 |
+
parser = argparse.ArgumentParser()
|
9 |
+
parser.add_argument('--weights', type=str, default='./yolov4.pt', help='weights path')
|
10 |
+
parser.add_argument('--img-size', nargs='+', type=int, default=[640, 640], help='image size')
|
11 |
+
parser.add_argument('--batch-size', type=int, default=1, help='batch size')
|
12 |
+
opt = parser.parse_args()
|
13 |
+
opt.img_size *= 2 if len(opt.img_size) == 1 else 1 # expand
|
14 |
+
print(opt)
|
15 |
+
|
16 |
+
# Input
|
17 |
+
img = torch.zeros((opt.batch_size, 3, *opt.img_size)) # image size(1,3,320,192) iDetection
|
18 |
+
|
19 |
+
# Load PyTorch model
|
20 |
+
attempt_download(opt.weights)
|
21 |
+
model = torch.load(opt.weights, map_location=torch.device('cpu'))['model'].float()
|
22 |
+
model.eval()
|
23 |
+
model.model[-1].export = True # set Detect() layer export=True
|
24 |
+
y = model(img) # dry run
|
25 |
+
|
26 |
+
# TorchScript export
|
27 |
+
try:
|
28 |
+
print('\nStarting TorchScript export with torch %s...' % torch.__version__)
|
29 |
+
f = opt.weights.replace('.pt', '.torchscript.pt') # filename
|
30 |
+
ts = torch.jit.trace(model, img)
|
31 |
+
ts.save(f)
|
32 |
+
print('TorchScript export success, saved as %s' % f)
|
33 |
+
except Exception as e:
|
34 |
+
print('TorchScript export failure: %s' % e)
|
35 |
+
|
36 |
+
# ONNX export
|
37 |
+
try:
|
38 |
+
import onnx
|
39 |
+
|
40 |
+
print('\nStarting ONNX export with onnx %s...' % onnx.__version__)
|
41 |
+
f = opt.weights.replace('.pt', '.onnx') # filename
|
42 |
+
model.fuse() # only for ONNX
|
43 |
+
torch.onnx.export(model, img, f, verbose=False, opset_version=12, input_names=['images'],
|
44 |
+
output_names=['classes', 'boxes'] if y is None else ['output'])
|
45 |
+
|
46 |
+
# Checks
|
47 |
+
onnx_model = onnx.load(f) # load onnx model
|
48 |
+
onnx.checker.check_model(onnx_model) # check onnx model
|
49 |
+
print(onnx.helper.printable_graph(onnx_model.graph)) # print a human readable model
|
50 |
+
print('ONNX export success, saved as %s' % f)
|
51 |
+
except Exception as e:
|
52 |
+
print('ONNX export failure: %s' % e)
|
53 |
+
|
54 |
+
# CoreML export
|
55 |
+
try:
|
56 |
+
import coremltools as ct
|
57 |
+
|
58 |
+
print('\nStarting CoreML export with coremltools %s...' % ct.__version__)
|
59 |
+
# convert model from torchscript and apply pixel scaling as per detect.py
|
60 |
+
model = ct.convert(ts, inputs=[ct.ImageType(name='images', shape=img.shape, scale=1 / 255.0, bias=[0, 0, 0])])
|
61 |
+
f = opt.weights.replace('.pt', '.mlmodel') # filename
|
62 |
+
model.save(f)
|
63 |
+
print('CoreML export success, saved as %s' % f)
|
64 |
+
except Exception as e:
|
65 |
+
print('CoreML export failure: %s' % e)
|
66 |
+
|
67 |
+
# Finish
|
68 |
+
print('\nExport complete. Visualize with https://github.com/lutzroeder/netron.')
|
models/models.py
ADDED
@@ -0,0 +1,761 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from utils.google_utils import *
|
2 |
+
from utils.layers import *
|
3 |
+
from utils.parse_config import *
|
4 |
+
from utils import torch_utils
|
5 |
+
|
6 |
+
ONNX_EXPORT = False
|
7 |
+
|
8 |
+
|
9 |
+
def create_modules(module_defs, img_size, cfg):
|
10 |
+
# Constructs module list of layer blocks from module configuration in module_defs
|
11 |
+
|
12 |
+
img_size = [img_size] * 2 if isinstance(img_size, int) else img_size # expand if necessary
|
13 |
+
_ = module_defs.pop(0) # cfg training hyperparams (unused)
|
14 |
+
output_filters = [3] # input channels
|
15 |
+
module_list = nn.ModuleList()
|
16 |
+
routs = [] # list of layers which rout to deeper layers
|
17 |
+
yolo_index = -1
|
18 |
+
|
19 |
+
for i, mdef in enumerate(module_defs):
|
20 |
+
modules = nn.Sequential()
|
21 |
+
|
22 |
+
if mdef['type'] == 'convolutional':
|
23 |
+
bn = mdef['batch_normalize']
|
24 |
+
filters = mdef['filters']
|
25 |
+
k = mdef['size'] # kernel size
|
26 |
+
stride = mdef['stride'] if 'stride' in mdef else (mdef['stride_y'], mdef['stride_x'])
|
27 |
+
if isinstance(k, int): # single-size conv
|
28 |
+
modules.add_module('Conv2d', nn.Conv2d(in_channels=output_filters[-1],
|
29 |
+
out_channels=filters,
|
30 |
+
kernel_size=k,
|
31 |
+
stride=stride,
|
32 |
+
padding=k // 2 if mdef['pad'] else 0,
|
33 |
+
groups=mdef['groups'] if 'groups' in mdef else 1,
|
34 |
+
bias=not bn))
|
35 |
+
else: # multiple-size conv
|
36 |
+
modules.add_module('MixConv2d', MixConv2d(in_ch=output_filters[-1],
|
37 |
+
out_ch=filters,
|
38 |
+
k=k,
|
39 |
+
stride=stride,
|
40 |
+
bias=not bn))
|
41 |
+
|
42 |
+
if bn:
|
43 |
+
modules.add_module('BatchNorm2d', nn.BatchNorm2d(filters, momentum=0.03, eps=1E-4))
|
44 |
+
else:
|
45 |
+
routs.append(i) # detection output (goes into yolo layer)
|
46 |
+
|
47 |
+
if mdef['activation'] == 'leaky': # activation study https://github.com/ultralytics/yolov3/issues/441
|
48 |
+
modules.add_module('activation', nn.LeakyReLU(0.1, inplace=True))
|
49 |
+
elif mdef['activation'] == 'swish':
|
50 |
+
modules.add_module('activation', Swish())
|
51 |
+
elif mdef['activation'] == 'mish':
|
52 |
+
modules.add_module('activation', Mish())
|
53 |
+
elif mdef['activation'] == 'emb':
|
54 |
+
modules.add_module('activation', F.normalize())
|
55 |
+
elif mdef['activation'] == 'logistic':
|
56 |
+
modules.add_module('activation', nn.Sigmoid())
|
57 |
+
elif mdef['activation'] == 'silu':
|
58 |
+
modules.add_module('activation', nn.SiLU())
|
59 |
+
|
60 |
+
elif mdef['type'] == 'deformableconvolutional':
|
61 |
+
bn = mdef['batch_normalize']
|
62 |
+
filters = mdef['filters']
|
63 |
+
k = mdef['size'] # kernel size
|
64 |
+
stride = mdef['stride'] if 'stride' in mdef else (mdef['stride_y'], mdef['stride_x'])
|
65 |
+
if isinstance(k, int): # single-size conv
|
66 |
+
modules.add_module('DeformConv2d', DeformConv2d(output_filters[-1],
|
67 |
+
filters,
|
68 |
+
kernel_size=k,
|
69 |
+
padding=k // 2 if mdef['pad'] else 0,
|
70 |
+
stride=stride,
|
71 |
+
bias=not bn,
|
72 |
+
modulation=True))
|
73 |
+
else: # multiple-size conv
|
74 |
+
modules.add_module('MixConv2d', MixConv2d(in_ch=output_filters[-1],
|
75 |
+
out_ch=filters,
|
76 |
+
k=k,
|
77 |
+
stride=stride,
|
78 |
+
bias=not bn))
|
79 |
+
|
80 |
+
if bn:
|
81 |
+
modules.add_module('BatchNorm2d', nn.BatchNorm2d(filters, momentum=0.03, eps=1E-4))
|
82 |
+
else:
|
83 |
+
routs.append(i) # detection output (goes into yolo layer)
|
84 |
+
|
85 |
+
if mdef['activation'] == 'leaky': # activation study https://github.com/ultralytics/yolov3/issues/441
|
86 |
+
modules.add_module('activation', nn.LeakyReLU(0.1, inplace=True))
|
87 |
+
elif mdef['activation'] == 'swish':
|
88 |
+
modules.add_module('activation', Swish())
|
89 |
+
elif mdef['activation'] == 'mish':
|
90 |
+
modules.add_module('activation', Mish())
|
91 |
+
elif mdef['activation'] == 'silu':
|
92 |
+
modules.add_module('activation', nn.SiLU())
|
93 |
+
|
94 |
+
elif mdef['type'] == 'dropout':
|
95 |
+
p = mdef['probability']
|
96 |
+
modules = nn.Dropout(p)
|
97 |
+
|
98 |
+
elif mdef['type'] == 'avgpool':
|
99 |
+
modules = GAP()
|
100 |
+
|
101 |
+
elif mdef['type'] == 'silence':
|
102 |
+
filters = output_filters[-1]
|
103 |
+
modules = Silence()
|
104 |
+
|
105 |
+
elif mdef['type'] == 'scale_channels': # nn.Sequential() placeholder for 'shortcut' layer
|
106 |
+
layers = mdef['from']
|
107 |
+
filters = output_filters[-1]
|
108 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
109 |
+
modules = ScaleChannel(layers=layers)
|
110 |
+
|
111 |
+
elif mdef['type'] == 'shift_channels': # nn.Sequential() placeholder for 'shortcut' layer
|
112 |
+
layers = mdef['from']
|
113 |
+
filters = output_filters[-1]
|
114 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
115 |
+
modules = ShiftChannel(layers=layers)
|
116 |
+
|
117 |
+
elif mdef['type'] == 'shift_channels_2d': # nn.Sequential() placeholder for 'shortcut' layer
|
118 |
+
layers = mdef['from']
|
119 |
+
filters = output_filters[-1]
|
120 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
121 |
+
modules = ShiftChannel2D(layers=layers)
|
122 |
+
|
123 |
+
elif mdef['type'] == 'control_channels': # nn.Sequential() placeholder for 'shortcut' layer
|
124 |
+
layers = mdef['from']
|
125 |
+
filters = output_filters[-1]
|
126 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
127 |
+
modules = ControlChannel(layers=layers)
|
128 |
+
|
129 |
+
elif mdef['type'] == 'control_channels_2d': # nn.Sequential() placeholder for 'shortcut' layer
|
130 |
+
layers = mdef['from']
|
131 |
+
filters = output_filters[-1]
|
132 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
133 |
+
modules = ControlChannel2D(layers=layers)
|
134 |
+
|
135 |
+
elif mdef['type'] == 'alternate_channels': # nn.Sequential() placeholder for 'shortcut' layer
|
136 |
+
layers = mdef['from']
|
137 |
+
filters = output_filters[-1] * 2
|
138 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
139 |
+
modules = AlternateChannel(layers=layers)
|
140 |
+
|
141 |
+
elif mdef['type'] == 'alternate_channels_2d': # nn.Sequential() placeholder for 'shortcut' layer
|
142 |
+
layers = mdef['from']
|
143 |
+
filters = output_filters[-1] * 2
|
144 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
145 |
+
modules = AlternateChannel2D(layers=layers)
|
146 |
+
|
147 |
+
elif mdef['type'] == 'select_channels': # nn.Sequential() placeholder for 'shortcut' layer
|
148 |
+
layers = mdef['from']
|
149 |
+
filters = output_filters[-1]
|
150 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
151 |
+
modules = SelectChannel(layers=layers)
|
152 |
+
|
153 |
+
elif mdef['type'] == 'select_channels_2d': # nn.Sequential() placeholder for 'shortcut' layer
|
154 |
+
layers = mdef['from']
|
155 |
+
filters = output_filters[-1]
|
156 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
157 |
+
modules = SelectChannel2D(layers=layers)
|
158 |
+
|
159 |
+
elif mdef['type'] == 'sam': # nn.Sequential() placeholder for 'shortcut' layer
|
160 |
+
layers = mdef['from']
|
161 |
+
filters = output_filters[-1]
|
162 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
163 |
+
modules = ScaleSpatial(layers=layers)
|
164 |
+
|
165 |
+
elif mdef['type'] == 'BatchNorm2d':
|
166 |
+
filters = output_filters[-1]
|
167 |
+
modules = nn.BatchNorm2d(filters, momentum=0.03, eps=1E-4)
|
168 |
+
if i == 0 and filters == 3: # normalize RGB image
|
169 |
+
# imagenet mean and var https://pytorch.org/docs/stable/torchvision/models.html#classification
|
170 |
+
modules.running_mean = torch.tensor([0.485, 0.456, 0.406])
|
171 |
+
modules.running_var = torch.tensor([0.0524, 0.0502, 0.0506])
|
172 |
+
|
173 |
+
elif mdef['type'] == 'maxpool':
|
174 |
+
k = mdef['size'] # kernel size
|
175 |
+
stride = mdef['stride']
|
176 |
+
maxpool = nn.MaxPool2d(kernel_size=k, stride=stride, padding=(k - 1) // 2)
|
177 |
+
if k == 2 and stride == 1: # yolov3-tiny
|
178 |
+
modules.add_module('ZeroPad2d', nn.ZeroPad2d((0, 1, 0, 1)))
|
179 |
+
modules.add_module('MaxPool2d', maxpool)
|
180 |
+
else:
|
181 |
+
modules = maxpool
|
182 |
+
|
183 |
+
elif mdef['type'] == 'local_avgpool':
|
184 |
+
k = mdef['size'] # kernel size
|
185 |
+
stride = mdef['stride']
|
186 |
+
avgpool = nn.AvgPool2d(kernel_size=k, stride=stride, padding=(k - 1) // 2)
|
187 |
+
if k == 2 and stride == 1: # yolov3-tiny
|
188 |
+
modules.add_module('ZeroPad2d', nn.ZeroPad2d((0, 1, 0, 1)))
|
189 |
+
modules.add_module('AvgPool2d', avgpool)
|
190 |
+
else:
|
191 |
+
modules = avgpool
|
192 |
+
|
193 |
+
elif mdef['type'] == 'upsample':
|
194 |
+
if ONNX_EXPORT: # explicitly state size, avoid scale_factor
|
195 |
+
g = (yolo_index + 1) * 2 / 32 # gain
|
196 |
+
modules = nn.Upsample(size=tuple(int(x * g) for x in img_size)) # img_size = (320, 192)
|
197 |
+
else:
|
198 |
+
modules = nn.Upsample(scale_factor=mdef['stride'])
|
199 |
+
|
200 |
+
elif mdef['type'] == 'route': # nn.Sequential() placeholder for 'route' layer
|
201 |
+
layers = mdef['layers']
|
202 |
+
filters = sum([output_filters[l + 1 if l > 0 else l] for l in layers])
|
203 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
204 |
+
modules = FeatureConcat(layers=layers)
|
205 |
+
|
206 |
+
elif mdef['type'] == 'route2': # nn.Sequential() placeholder for 'route' layer
|
207 |
+
layers = mdef['layers']
|
208 |
+
filters = sum([output_filters[l + 1 if l > 0 else l] for l in layers])
|
209 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
210 |
+
modules = FeatureConcat2(layers=layers)
|
211 |
+
|
212 |
+
elif mdef['type'] == 'route3': # nn.Sequential() placeholder for 'route' layer
|
213 |
+
layers = mdef['layers']
|
214 |
+
filters = sum([output_filters[l + 1 if l > 0 else l] for l in layers])
|
215 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
216 |
+
modules = FeatureConcat3(layers=layers)
|
217 |
+
|
218 |
+
elif mdef['type'] == 'route_lhalf': # nn.Sequential() placeholder for 'route' layer
|
219 |
+
layers = mdef['layers']
|
220 |
+
filters = sum([output_filters[l + 1 if l > 0 else l] for l in layers])//2
|
221 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
222 |
+
modules = FeatureConcat_l(layers=layers)
|
223 |
+
|
224 |
+
elif mdef['type'] == 'shortcut': # nn.Sequential() placeholder for 'shortcut' layer
|
225 |
+
layers = mdef['from']
|
226 |
+
filters = output_filters[-1]
|
227 |
+
routs.extend([i + l if l < 0 else l for l in layers])
|
228 |
+
modules = WeightedFeatureFusion(layers=layers, weight='weights_type' in mdef)
|
229 |
+
|
230 |
+
elif mdef['type'] == 'reorg3d': # yolov3-spp-pan-scale
|
231 |
+
pass
|
232 |
+
|
233 |
+
elif mdef['type'] == 'reorg': # yolov3-spp-pan-scale
|
234 |
+
filters = 4 * output_filters[-1]
|
235 |
+
modules.add_module('Reorg', Reorg())
|
236 |
+
|
237 |
+
elif mdef['type'] == 'dwt': # yolov3-spp-pan-scale
|
238 |
+
filters = 4 * output_filters[-1]
|
239 |
+
modules.add_module('DWT', DWT())
|
240 |
+
|
241 |
+
elif mdef['type'] == 'implicit_add': # yolov3-spp-pan-scale
|
242 |
+
filters = mdef['filters']
|
243 |
+
modules = ImplicitA(channel=filters)
|
244 |
+
|
245 |
+
elif mdef['type'] == 'implicit_mul': # yolov3-spp-pan-scale
|
246 |
+
filters = mdef['filters']
|
247 |
+
modules = ImplicitM(channel=filters)
|
248 |
+
|
249 |
+
elif mdef['type'] == 'implicit_cat': # yolov3-spp-pan-scale
|
250 |
+
filters = mdef['filters']
|
251 |
+
modules = ImplicitC(channel=filters)
|
252 |
+
|
253 |
+
elif mdef['type'] == 'implicit_add_2d': # yolov3-spp-pan-scale
|
254 |
+
channels = mdef['filters']
|
255 |
+
filters = mdef['atoms']
|
256 |
+
modules = Implicit2DA(atom=filters, channel=channels)
|
257 |
+
|
258 |
+
elif mdef['type'] == 'implicit_mul_2d': # yolov3-spp-pan-scale
|
259 |
+
channels = mdef['filters']
|
260 |
+
filters = mdef['atoms']
|
261 |
+
modules = Implicit2DM(atom=filters, channel=channels)
|
262 |
+
|
263 |
+
elif mdef['type'] == 'implicit_cat_2d': # yolov3-spp-pan-scale
|
264 |
+
channels = mdef['filters']
|
265 |
+
filters = mdef['atoms']
|
266 |
+
modules = Implicit2DC(atom=filters, channel=channels)
|
267 |
+
|
268 |
+
elif mdef['type'] == 'yolo':
|
269 |
+
yolo_index += 1
|
270 |
+
stride = [8, 16, 32, 64, 128] # P3, P4, P5, P6, P7 strides
|
271 |
+
if any(x in cfg for x in ['yolov4-tiny', 'fpn', 'yolov3']): # P5, P4, P3 strides
|
272 |
+
stride = [32, 16, 8]
|
273 |
+
layers = mdef['from'] if 'from' in mdef else []
|
274 |
+
modules = YOLOLayer(anchors=mdef['anchors'][mdef['mask']], # anchor list
|
275 |
+
nc=mdef['classes'], # number of classes
|
276 |
+
img_size=img_size, # (416, 416)
|
277 |
+
yolo_index=yolo_index, # 0, 1, 2...
|
278 |
+
layers=layers, # output layers
|
279 |
+
stride=stride[yolo_index])
|
280 |
+
|
281 |
+
# Initialize preceding Conv2d() bias (https://arxiv.org/pdf/1708.02002.pdf section 3.3)
|
282 |
+
try:
|
283 |
+
j = layers[yolo_index] if 'from' in mdef else -2
|
284 |
+
bias_ = module_list[j][0].bias # shape(255,)
|
285 |
+
bias = bias_[:modules.no * modules.na].view(modules.na, -1) # shape(3,85)
|
286 |
+
#bias[:, 4] += -4.5 # obj
|
287 |
+
bias.data[:, 4] += math.log(8 / (640 / stride[yolo_index]) ** 2) # obj (8 objects per 640 image)
|
288 |
+
bias.data[:, 5:] += math.log(0.6 / (modules.nc - 0.99)) # cls (sigmoid(p) = 1/nc)
|
289 |
+
module_list[j][0].bias = torch.nn.Parameter(bias_, requires_grad=bias_.requires_grad)
|
290 |
+
|
291 |
+
#j = [-2, -5, -8]
|
292 |
+
#for sj in j:
|
293 |
+
# bias_ = module_list[sj][0].bias
|
294 |
+
# bias = bias_[:modules.no * 1].view(1, -1)
|
295 |
+
# bias.data[:, 4] += math.log(8 / (640 / stride[yolo_index]) ** 2)
|
296 |
+
# bias.data[:, 5:] += math.log(0.6 / (modules.nc - 0.99))
|
297 |
+
# module_list[sj][0].bias = torch.nn.Parameter(bias_, requires_grad=bias_.requires_grad)
|
298 |
+
except:
|
299 |
+
print('WARNING: smart bias initialization failure.')
|
300 |
+
|
301 |
+
elif mdef['type'] == 'jde':
|
302 |
+
yolo_index += 1
|
303 |
+
stride = [8, 16, 32, 64, 128] # P3, P4, P5, P6, P7 strides
|
304 |
+
if any(x in cfg for x in ['yolov4-tiny', 'fpn', 'yolov3']): # P5, P4, P3 strides
|
305 |
+
stride = [32, 16, 8]
|
306 |
+
layers = mdef['from'] if 'from' in mdef else []
|
307 |
+
modules = JDELayer(anchors=mdef['anchors'][mdef['mask']], # anchor list
|
308 |
+
nc=mdef['classes'], # number of classes
|
309 |
+
img_size=img_size, # (416, 416)
|
310 |
+
yolo_index=yolo_index, # 0, 1, 2...
|
311 |
+
layers=layers, # output layers
|
312 |
+
stride=stride[yolo_index])
|
313 |
+
|
314 |
+
# Initialize preceding Conv2d() bias (https://arxiv.org/pdf/1708.02002.pdf section 3.3)
|
315 |
+
try:
|
316 |
+
j = layers[yolo_index] if 'from' in mdef else -1
|
317 |
+
bias_ = module_list[j][0].bias # shape(255,)
|
318 |
+
bias = bias_[:modules.no * modules.na].view(modules.na, -1) # shape(3,85)
|
319 |
+
#bias[:, 4] += -4.5 # obj
|
320 |
+
bias.data[:, 4] += math.log(8 / (640 / stride[yolo_index]) ** 2) # obj (8 objects per 640 image)
|
321 |
+
bias.data[:, 5:] += math.log(0.6 / (modules.nc - 0.99)) # cls (sigmoid(p) = 1/nc)
|
322 |
+
module_list[j][0].bias = torch.nn.Parameter(bias_, requires_grad=bias_.requires_grad)
|
323 |
+
except:
|
324 |
+
print('WARNING: smart bias initialization failure.')
|
325 |
+
|
326 |
+
else:
|
327 |
+
print('Warning: Unrecognized Layer Type: ' + mdef['type'])
|
328 |
+
|
329 |
+
# Register module list and number of output filters
|
330 |
+
module_list.append(modules)
|
331 |
+
output_filters.append(filters)
|
332 |
+
|
333 |
+
routs_binary = [False] * (i + 1)
|
334 |
+
for i in routs:
|
335 |
+
routs_binary[i] = True
|
336 |
+
return module_list, routs_binary
|
337 |
+
|
338 |
+
|
339 |
+
class YOLOLayer(nn.Module):
|
340 |
+
def __init__(self, anchors, nc, img_size, yolo_index, layers, stride):
|
341 |
+
super(YOLOLayer, self).__init__()
|
342 |
+
self.anchors = torch.Tensor(anchors)
|
343 |
+
self.index = yolo_index # index of this layer in layers
|
344 |
+
self.layers = layers # model output layer indices
|
345 |
+
self.stride = stride # layer stride
|
346 |
+
self.nl = len(layers) # number of output layers (3)
|
347 |
+
self.na = len(anchors) # number of anchors (3)
|
348 |
+
self.nc = nc # number of classes (80)
|
349 |
+
self.no = nc + 5 # number of outputs (85)
|
350 |
+
self.nx, self.ny, self.ng = 0, 0, 0 # initialize number of x, y gridpoints
|
351 |
+
self.anchor_vec = self.anchors / self.stride
|
352 |
+
self.anchor_wh = self.anchor_vec.view(1, self.na, 1, 1, 2)
|
353 |
+
|
354 |
+
if ONNX_EXPORT:
|
355 |
+
self.training = False
|
356 |
+
self.create_grids((img_size[1] // stride, img_size[0] // stride)) # number x, y grid points
|
357 |
+
|
358 |
+
def create_grids(self, ng=(13, 13), device='cpu'):
|
359 |
+
self.nx, self.ny = ng # x and y grid size
|
360 |
+
self.ng = torch.tensor(ng, dtype=torch.float)
|
361 |
+
|
362 |
+
# build xy offsets
|
363 |
+
if not self.training:
|
364 |
+
yv, xv = torch.meshgrid([torch.arange(self.ny, device=device), torch.arange(self.nx, device=device)])
|
365 |
+
self.grid = torch.stack((xv, yv), 2).view((1, 1, self.ny, self.nx, 2)).float()
|
366 |
+
|
367 |
+
if self.anchor_vec.device != device:
|
368 |
+
self.anchor_vec = self.anchor_vec.to(device)
|
369 |
+
self.anchor_wh = self.anchor_wh.to(device)
|
370 |
+
|
371 |
+
def forward(self, p, out):
|
372 |
+
ASFF = False # https://arxiv.org/abs/1911.09516
|
373 |
+
if ASFF:
|
374 |
+
i, n = self.index, self.nl # index in layers, number of layers
|
375 |
+
p = out[self.layers[i]]
|
376 |
+
bs, _, ny, nx = p.shape # bs, 255, 13, 13
|
377 |
+
if (self.nx, self.ny) != (nx, ny):
|
378 |
+
self.create_grids((nx, ny), p.device)
|
379 |
+
|
380 |
+
# outputs and weights
|
381 |
+
# w = F.softmax(p[:, -n:], 1) # normalized weights
|
382 |
+
w = torch.sigmoid(p[:, -n:]) * (2 / n) # sigmoid weights (faster)
|
383 |
+
# w = w / w.sum(1).unsqueeze(1) # normalize across layer dimension
|
384 |
+
|
385 |
+
# weighted ASFF sum
|
386 |
+
p = out[self.layers[i]][:, :-n] * w[:, i:i + 1]
|
387 |
+
for j in range(n):
|
388 |
+
if j != i:
|
389 |
+
p += w[:, j:j + 1] * \
|
390 |
+
F.interpolate(out[self.layers[j]][:, :-n], size=[ny, nx], mode='bilinear', align_corners=False)
|
391 |
+
|
392 |
+
elif ONNX_EXPORT:
|
393 |
+
bs = 1 # batch size
|
394 |
+
else:
|
395 |
+
bs, _, ny, nx = p.shape # bs, 255, 13, 13
|
396 |
+
if (self.nx, self.ny) != (nx, ny):
|
397 |
+
self.create_grids((nx, ny), p.device)
|
398 |
+
|
399 |
+
# p.view(bs, 255, 13, 13) -- > (bs, 3, 13, 13, 85) # (bs, anchors, grid, grid, classes + xywh)
|
400 |
+
p = p.view(bs, self.na, self.no, self.ny, self.nx).permute(0, 1, 3, 4, 2).contiguous() # prediction
|
401 |
+
|
402 |
+
if self.training:
|
403 |
+
return p
|
404 |
+
|
405 |
+
elif ONNX_EXPORT:
|
406 |
+
# Avoid broadcasting for ANE operations
|
407 |
+
m = self.na * self.nx * self.ny
|
408 |
+
ng = 1. / self.ng.repeat(m, 1)
|
409 |
+
grid = self.grid.repeat(1, self.na, 1, 1, 1).view(m, 2)
|
410 |
+
anchor_wh = self.anchor_wh.repeat(1, 1, self.nx, self.ny, 1).view(m, 2) * ng
|
411 |
+
|
412 |
+
p = p.view(m, self.no)
|
413 |
+
xy = torch.sigmoid(p[:, 0:2]) + grid # x, y
|
414 |
+
wh = torch.exp(p[:, 2:4]) * anchor_wh # width, height
|
415 |
+
p_cls = torch.sigmoid(p[:, 4:5]) if self.nc == 1 else \
|
416 |
+
torch.sigmoid(p[:, 5:self.no]) * torch.sigmoid(p[:, 4:5]) # conf
|
417 |
+
return p_cls, xy * ng, wh
|
418 |
+
|
419 |
+
else: # inference
|
420 |
+
io = p.sigmoid()
|
421 |
+
io[..., :2] = (io[..., :2] * 2. - 0.5 + self.grid)
|
422 |
+
io[..., 2:4] = (io[..., 2:4] * 2) ** 2 * self.anchor_wh
|
423 |
+
io[..., :4] *= self.stride
|
424 |
+
#io = p.clone() # inference output
|
425 |
+
#io[..., :2] = torch.sigmoid(io[..., :2]) + self.grid # xy
|
426 |
+
#io[..., 2:4] = torch.exp(io[..., 2:4]) * self.anchor_wh # wh yolo method
|
427 |
+
#io[..., :4] *= self.stride
|
428 |
+
#torch.sigmoid_(io[..., 4:])
|
429 |
+
return io.view(bs, -1, self.no), p # view [1, 3, 13, 13, 85] as [1, 507, 85]
|
430 |
+
|
431 |
+
|
432 |
+
class JDELayer(nn.Module):
|
433 |
+
def __init__(self, anchors, nc, img_size, yolo_index, layers, stride):
|
434 |
+
super(JDELayer, self).__init__()
|
435 |
+
self.anchors = torch.Tensor(anchors)
|
436 |
+
self.index = yolo_index # index of this layer in layers
|
437 |
+
self.layers = layers # model output layer indices
|
438 |
+
self.stride = stride # layer stride
|
439 |
+
self.nl = len(layers) # number of output layers (3)
|
440 |
+
self.na = len(anchors) # number of anchors (3)
|
441 |
+
self.nc = nc # number of classes (80)
|
442 |
+
self.no = nc + 5 # number of outputs (85)
|
443 |
+
self.nx, self.ny, self.ng = 0, 0, 0 # initialize number of x, y gridpoints
|
444 |
+
self.anchor_vec = self.anchors / self.stride
|
445 |
+
self.anchor_wh = self.anchor_vec.view(1, self.na, 1, 1, 2)
|
446 |
+
|
447 |
+
if ONNX_EXPORT:
|
448 |
+
self.training = False
|
449 |
+
self.create_grids((img_size[1] // stride, img_size[0] // stride)) # number x, y grid points
|
450 |
+
|
451 |
+
def create_grids(self, ng=(13, 13), device='cpu'):
|
452 |
+
self.nx, self.ny = ng # x and y grid size
|
453 |
+
self.ng = torch.tensor(ng, dtype=torch.float)
|
454 |
+
|
455 |
+
# build xy offsets
|
456 |
+
if not self.training:
|
457 |
+
yv, xv = torch.meshgrid([torch.arange(self.ny, device=device), torch.arange(self.nx, device=device)])
|
458 |
+
self.grid = torch.stack((xv, yv), 2).view((1, 1, self.ny, self.nx, 2)).float()
|
459 |
+
|
460 |
+
if self.anchor_vec.device != device:
|
461 |
+
self.anchor_vec = self.anchor_vec.to(device)
|
462 |
+
self.anchor_wh = self.anchor_wh.to(device)
|
463 |
+
|
464 |
+
def forward(self, p, out):
|
465 |
+
ASFF = False # https://arxiv.org/abs/1911.09516
|
466 |
+
if ASFF:
|
467 |
+
i, n = self.index, self.nl # index in layers, number of layers
|
468 |
+
p = out[self.layers[i]]
|
469 |
+
bs, _, ny, nx = p.shape # bs, 255, 13, 13
|
470 |
+
if (self.nx, self.ny) != (nx, ny):
|
471 |
+
self.create_grids((nx, ny), p.device)
|
472 |
+
|
473 |
+
# outputs and weights
|
474 |
+
# w = F.softmax(p[:, -n:], 1) # normalized weights
|
475 |
+
w = torch.sigmoid(p[:, -n:]) * (2 / n) # sigmoid weights (faster)
|
476 |
+
# w = w / w.sum(1).unsqueeze(1) # normalize across layer dimension
|
477 |
+
|
478 |
+
# weighted ASFF sum
|
479 |
+
p = out[self.layers[i]][:, :-n] * w[:, i:i + 1]
|
480 |
+
for j in range(n):
|
481 |
+
if j != i:
|
482 |
+
p += w[:, j:j + 1] * \
|
483 |
+
F.interpolate(out[self.layers[j]][:, :-n], size=[ny, nx], mode='bilinear', align_corners=False)
|
484 |
+
|
485 |
+
elif ONNX_EXPORT:
|
486 |
+
bs = 1 # batch size
|
487 |
+
else:
|
488 |
+
bs, _, ny, nx = p.shape # bs, 255, 13, 13
|
489 |
+
if (self.nx, self.ny) != (nx, ny):
|
490 |
+
self.create_grids((nx, ny), p.device)
|
491 |
+
|
492 |
+
# p.view(bs, 255, 13, 13) -- > (bs, 3, 13, 13, 85) # (bs, anchors, grid, grid, classes + xywh)
|
493 |
+
p = p.view(bs, self.na, self.no, self.ny, self.nx).permute(0, 1, 3, 4, 2).contiguous() # prediction
|
494 |
+
|
495 |
+
if self.training:
|
496 |
+
return p
|
497 |
+
|
498 |
+
elif ONNX_EXPORT:
|
499 |
+
# Avoid broadcasting for ANE operations
|
500 |
+
m = self.na * self.nx * self.ny
|
501 |
+
ng = 1. / self.ng.repeat(m, 1)
|
502 |
+
grid = self.grid.repeat(1, self.na, 1, 1, 1).view(m, 2)
|
503 |
+
anchor_wh = self.anchor_wh.repeat(1, 1, self.nx, self.ny, 1).view(m, 2) * ng
|
504 |
+
|
505 |
+
p = p.view(m, self.no)
|
506 |
+
xy = torch.sigmoid(p[:, 0:2]) + grid # x, y
|
507 |
+
wh = torch.exp(p[:, 2:4]) * anchor_wh # width, height
|
508 |
+
p_cls = torch.sigmoid(p[:, 4:5]) if self.nc == 1 else \
|
509 |
+
torch.sigmoid(p[:, 5:self.no]) * torch.sigmoid(p[:, 4:5]) # conf
|
510 |
+
return p_cls, xy * ng, wh
|
511 |
+
|
512 |
+
else: # inference
|
513 |
+
#io = p.sigmoid()
|
514 |
+
#io[..., :2] = (io[..., :2] * 2. - 0.5 + self.grid)
|
515 |
+
#io[..., 2:4] = (io[..., 2:4] * 2) ** 2 * self.anchor_wh
|
516 |
+
#io[..., :4] *= self.stride
|
517 |
+
io = p.clone() # inference output
|
518 |
+
io[..., :2] = torch.sigmoid(io[..., :2]) * 2. - 0.5 + self.grid # xy
|
519 |
+
io[..., 2:4] = (torch.sigmoid(io[..., 2:4]) * 2) ** 2 * self.anchor_wh # wh yolo method
|
520 |
+
io[..., :4] *= self.stride
|
521 |
+
io[..., 4:] = F.softmax(io[..., 4:])
|
522 |
+
return io.view(bs, -1, self.no), p # view [1, 3, 13, 13, 85] as [1, 507, 85]
|
523 |
+
|
524 |
+
class Darknet(nn.Module):
|
525 |
+
# YOLOv3 object detection model
|
526 |
+
|
527 |
+
def __init__(self, cfg, img_size=(416, 416), verbose=False):
|
528 |
+
super(Darknet, self).__init__()
|
529 |
+
|
530 |
+
self.module_defs = parse_model_cfg(cfg)
|
531 |
+
self.module_list, self.routs = create_modules(self.module_defs, img_size, cfg)
|
532 |
+
self.yolo_layers = get_yolo_layers(self)
|
533 |
+
# torch_utils.initialize_weights(self)
|
534 |
+
|
535 |
+
# Darknet Header https://github.com/AlexeyAB/darknet/issues/2914#issuecomment-496675346
|
536 |
+
self.version = np.array([0, 2, 5], dtype=np.int32) # (int32) version info: major, minor, revision
|
537 |
+
self.seen = np.array([0], dtype=np.int64) # (int64) number of images seen during training
|
538 |
+
self.info(verbose) if not ONNX_EXPORT else None # print model description
|
539 |
+
|
540 |
+
def forward(self, x, augment=False, verbose=False):
|
541 |
+
|
542 |
+
if not augment:
|
543 |
+
return self.forward_once(x)
|
544 |
+
else: # Augment images (inference and test only) https://github.com/ultralytics/yolov3/issues/931
|
545 |
+
img_size = x.shape[-2:] # height, width
|
546 |
+
s = [0.83, 0.67] # scales
|
547 |
+
y = []
|
548 |
+
for i, xi in enumerate((x,
|
549 |
+
torch_utils.scale_img(x.flip(3), s[0], same_shape=False), # flip-lr and scale
|
550 |
+
torch_utils.scale_img(x, s[1], same_shape=False), # scale
|
551 |
+
)):
|
552 |
+
# cv2.imwrite('img%g.jpg' % i, 255 * xi[0].numpy().transpose((1, 2, 0))[:, :, ::-1])
|
553 |
+
y.append(self.forward_once(xi)[0])
|
554 |
+
|
555 |
+
y[1][..., :4] /= s[0] # scale
|
556 |
+
y[1][..., 0] = img_size[1] - y[1][..., 0] # flip lr
|
557 |
+
y[2][..., :4] /= s[1] # scale
|
558 |
+
|
559 |
+
# for i, yi in enumerate(y): # coco small, medium, large = < 32**2 < 96**2 <
|
560 |
+
# area = yi[..., 2:4].prod(2)[:, :, None]
|
561 |
+
# if i == 1:
|
562 |
+
# yi *= (area < 96. ** 2).float()
|
563 |
+
# elif i == 2:
|
564 |
+
# yi *= (area > 32. ** 2).float()
|
565 |
+
# y[i] = yi
|
566 |
+
|
567 |
+
y = torch.cat(y, 1)
|
568 |
+
return y, None
|
569 |
+
|
570 |
+
def forward_once(self, x, augment=False, verbose=False):
|
571 |
+
img_size = x.shape[-2:] # height, width
|
572 |
+
yolo_out, out = [], []
|
573 |
+
if verbose:
|
574 |
+
print('0', x.shape)
|
575 |
+
str = ''
|
576 |
+
|
577 |
+
# Augment images (inference and test only)
|
578 |
+
if augment: # https://github.com/ultralytics/yolov3/issues/931
|
579 |
+
nb = x.shape[0] # batch size
|
580 |
+
s = [0.83, 0.67] # scales
|
581 |
+
x = torch.cat((x,
|
582 |
+
torch_utils.scale_img(x.flip(3), s[0]), # flip-lr and scale
|
583 |
+
torch_utils.scale_img(x, s[1]), # scale
|
584 |
+
), 0)
|
585 |
+
|
586 |
+
for i, module in enumerate(self.module_list):
|
587 |
+
name = module.__class__.__name__
|
588 |
+
#print(name)
|
589 |
+
if name in ['WeightedFeatureFusion', 'FeatureConcat', 'FeatureConcat2', 'FeatureConcat3', 'FeatureConcat_l', 'ScaleChannel', 'ShiftChannel', 'ShiftChannel2D', 'ControlChannel', 'ControlChannel2D', 'AlternateChannel', 'AlternateChannel2D', 'SelectChannel', 'SelectChannel2D', 'ScaleSpatial']: # sum, concat
|
590 |
+
if verbose:
|
591 |
+
l = [i - 1] + module.layers # layers
|
592 |
+
sh = [list(x.shape)] + [list(out[i].shape) for i in module.layers] # shapes
|
593 |
+
str = ' >> ' + ' + '.join(['layer %g %s' % x for x in zip(l, sh)])
|
594 |
+
x = module(x, out) # WeightedFeatureFusion(), FeatureConcat()
|
595 |
+
elif name in ['ImplicitA', 'ImplicitM', 'ImplicitC', 'Implicit2DA', 'Implicit2DM', 'Implicit2DC']:
|
596 |
+
x = module()
|
597 |
+
elif name == 'YOLOLayer':
|
598 |
+
yolo_out.append(module(x, out))
|
599 |
+
elif name == 'JDELayer':
|
600 |
+
yolo_out.append(module(x, out))
|
601 |
+
else: # run module directly, i.e. mtype = 'convolutional', 'upsample', 'maxpool', 'batchnorm2d' etc.
|
602 |
+
#print(module)
|
603 |
+
#print(x.shape)
|
604 |
+
x = module(x)
|
605 |
+
|
606 |
+
out.append(x if self.routs[i] else [])
|
607 |
+
if verbose:
|
608 |
+
print('%g/%g %s -' % (i, len(self.module_list), name), list(x.shape), str)
|
609 |
+
str = ''
|
610 |
+
|
611 |
+
if self.training: # train
|
612 |
+
return yolo_out
|
613 |
+
elif ONNX_EXPORT: # export
|
614 |
+
x = [torch.cat(x, 0) for x in zip(*yolo_out)]
|
615 |
+
return x[0], torch.cat(x[1:3], 1) # scores, boxes: 3780x80, 3780x4
|
616 |
+
else: # inference or test
|
617 |
+
x, p = zip(*yolo_out) # inference output, training output
|
618 |
+
x = torch.cat(x, 1) # cat yolo outputs
|
619 |
+
if augment: # de-augment results
|
620 |
+
x = torch.split(x, nb, dim=0)
|
621 |
+
x[1][..., :4] /= s[0] # scale
|
622 |
+
x[1][..., 0] = img_size[1] - x[1][..., 0] # flip lr
|
623 |
+
x[2][..., :4] /= s[1] # scale
|
624 |
+
x = torch.cat(x, 1)
|
625 |
+
return x, p
|
626 |
+
|
627 |
+
def fuse(self):
|
628 |
+
# Fuse Conv2d + BatchNorm2d layers throughout model
|
629 |
+
print('Fusing layers...')
|
630 |
+
fused_list = nn.ModuleList()
|
631 |
+
for a in list(self.children())[0]:
|
632 |
+
if isinstance(a, nn.Sequential):
|
633 |
+
for i, b in enumerate(a):
|
634 |
+
if isinstance(b, nn.modules.batchnorm.BatchNorm2d):
|
635 |
+
# fuse this bn layer with the previous conv2d layer
|
636 |
+
conv = a[i - 1]
|
637 |
+
fused = torch_utils.fuse_conv_and_bn(conv, b)
|
638 |
+
a = nn.Sequential(fused, *list(a.children())[i + 1:])
|
639 |
+
break
|
640 |
+
fused_list.append(a)
|
641 |
+
self.module_list = fused_list
|
642 |
+
self.info() if not ONNX_EXPORT else None # yolov3-spp reduced from 225 to 152 layers
|
643 |
+
|
644 |
+
def info(self, verbose=False):
|
645 |
+
torch_utils.model_info(self, verbose)
|
646 |
+
|
647 |
+
|
648 |
+
def get_yolo_layers(model):
|
649 |
+
return [i for i, m in enumerate(model.module_list) if m.__class__.__name__ in ['YOLOLayer', 'JDELayer']] # [89, 101, 113]
|
650 |
+
|
651 |
+
|
652 |
+
def load_darknet_weights(self, weights, cutoff=-1):
|
653 |
+
# Parses and loads the weights stored in 'weights'
|
654 |
+
|
655 |
+
# Establish cutoffs (load layers between 0 and cutoff. if cutoff = -1 all are loaded)
|
656 |
+
file = Path(weights).name
|
657 |
+
if file == 'darknet53.conv.74':
|
658 |
+
cutoff = 75
|
659 |
+
elif file == 'yolov3-tiny.conv.15':
|
660 |
+
cutoff = 15
|
661 |
+
|
662 |
+
# Read weights file
|
663 |
+
with open(weights, 'rb') as f:
|
664 |
+
# Read Header https://github.com/AlexeyAB/darknet/issues/2914#issuecomment-496675346
|
665 |
+
self.version = np.fromfile(f, dtype=np.int32, count=3) # (int32) version info: major, minor, revision
|
666 |
+
self.seen = np.fromfile(f, dtype=np.int64, count=1) # (int64) number of images seen during training
|
667 |
+
|
668 |
+
weights = np.fromfile(f, dtype=np.float32) # the rest are weights
|
669 |
+
|
670 |
+
ptr = 0
|
671 |
+
for i, (mdef, module) in enumerate(zip(self.module_defs[:cutoff], self.module_list[:cutoff])):
|
672 |
+
if mdef['type'] == 'convolutional':
|
673 |
+
conv = module[0]
|
674 |
+
if mdef['batch_normalize']:
|
675 |
+
# Load BN bias, weights, running mean and running variance
|
676 |
+
bn = module[1]
|
677 |
+
nb = bn.bias.numel() # number of biases
|
678 |
+
# Bias
|
679 |
+
bn.bias.data.copy_(torch.from_numpy(weights[ptr:ptr + nb]).view_as(bn.bias))
|
680 |
+
ptr += nb
|
681 |
+
# Weight
|
682 |
+
bn.weight.data.copy_(torch.from_numpy(weights[ptr:ptr + nb]).view_as(bn.weight))
|
683 |
+
ptr += nb
|
684 |
+
# Running Mean
|
685 |
+
bn.running_mean.data.copy_(torch.from_numpy(weights[ptr:ptr + nb]).view_as(bn.running_mean))
|
686 |
+
ptr += nb
|
687 |
+
# Running Var
|
688 |
+
bn.running_var.data.copy_(torch.from_numpy(weights[ptr:ptr + nb]).view_as(bn.running_var))
|
689 |
+
ptr += nb
|
690 |
+
else:
|
691 |
+
# Load conv. bias
|
692 |
+
nb = conv.bias.numel()
|
693 |
+
conv_b = torch.from_numpy(weights[ptr:ptr + nb]).view_as(conv.bias)
|
694 |
+
conv.bias.data.copy_(conv_b)
|
695 |
+
ptr += nb
|
696 |
+
# Load conv. weights
|
697 |
+
nw = conv.weight.numel() # number of weights
|
698 |
+
conv.weight.data.copy_(torch.from_numpy(weights[ptr:ptr + nw]).view_as(conv.weight))
|
699 |
+
ptr += nw
|
700 |
+
|
701 |
+
|
702 |
+
def save_weights(self, path='model.weights', cutoff=-1):
|
703 |
+
# Converts a PyTorch model to Darket format (*.pt to *.weights)
|
704 |
+
# Note: Does not work if model.fuse() is applied
|
705 |
+
with open(path, 'wb') as f:
|
706 |
+
# Write Header https://github.com/AlexeyAB/darknet/issues/2914#issuecomment-496675346
|
707 |
+
self.version.tofile(f) # (int32) version info: major, minor, revision
|
708 |
+
self.seen.tofile(f) # (int64) number of images seen during training
|
709 |
+
|
710 |
+
# Iterate through layers
|
711 |
+
for i, (mdef, module) in enumerate(zip(self.module_defs[:cutoff], self.module_list[:cutoff])):
|
712 |
+
if mdef['type'] == 'convolutional':
|
713 |
+
conv_layer = module[0]
|
714 |
+
# If batch norm, load bn first
|
715 |
+
if mdef['batch_normalize']:
|
716 |
+
bn_layer = module[1]
|
717 |
+
bn_layer.bias.data.cpu().numpy().tofile(f)
|
718 |
+
bn_layer.weight.data.cpu().numpy().tofile(f)
|
719 |
+
bn_layer.running_mean.data.cpu().numpy().tofile(f)
|
720 |
+
bn_layer.running_var.data.cpu().numpy().tofile(f)
|
721 |
+
# Load conv bias
|
722 |
+
else:
|
723 |
+
conv_layer.bias.data.cpu().numpy().tofile(f)
|
724 |
+
# Load conv weights
|
725 |
+
conv_layer.weight.data.cpu().numpy().tofile(f)
|
726 |
+
|
727 |
+
|
728 |
+
def convert(cfg='cfg/yolov3-spp.cfg', weights='weights/yolov3-spp.weights', saveto='converted.weights'):
|
729 |
+
# Converts between PyTorch and Darknet format per extension (i.e. *.weights convert to *.pt and vice versa)
|
730 |
+
# from models import *; convert('cfg/yolov3-spp.cfg', 'weights/yolov3-spp.weights')
|
731 |
+
|
732 |
+
# Initialize model
|
733 |
+
model = Darknet(cfg)
|
734 |
+
ckpt = torch.load(weights) # load checkpoint
|
735 |
+
try:
|
736 |
+
ckpt['model'] = {k: v for k, v in ckpt['model'].items() if model.state_dict()[k].numel() == v.numel()}
|
737 |
+
model.load_state_dict(ckpt['model'], strict=False)
|
738 |
+
save_weights(model, path=saveto, cutoff=-1)
|
739 |
+
except KeyError as e:
|
740 |
+
print(e)
|
741 |
+
|
742 |
+
def attempt_download(weights):
|
743 |
+
# Attempt to download pretrained weights if not found locally
|
744 |
+
weights = weights.strip()
|
745 |
+
msg = weights + ' missing, try downloading from https://drive.google.com/open?id=1LezFG5g3BCW6iYaV89B2i64cqEUZD7e0'
|
746 |
+
|
747 |
+
if len(weights) > 0 and not os.path.isfile(weights):
|
748 |
+
d = {''}
|
749 |
+
|
750 |
+
file = Path(weights).name
|
751 |
+
if file in d:
|
752 |
+
r = gdrive_download(id=d[file], name=weights)
|
753 |
+
else: # download from pjreddie.com
|
754 |
+
url = 'https://pjreddie.com/media/files/' + file
|
755 |
+
print('Downloading ' + url)
|
756 |
+
r = os.system('curl -f ' + url + ' -o ' + weights)
|
757 |
+
|
758 |
+
# Error check
|
759 |
+
if not (r == 0 and os.path.exists(weights) and os.path.getsize(weights) > 1E6): # weights exist and > 1MB
|
760 |
+
os.system('rm ' + weights) # remove partial downloads
|
761 |
+
raise Exception(msg)
|
requirements.txt
ADDED
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# pip install -qr requirements.txt
|
2 |
+
|
3 |
+
# base ----------------------------------------
|
4 |
+
Cython
|
5 |
+
matplotlib>=3.2.2
|
6 |
+
numpy>=1.18.5
|
7 |
+
opencv-python>=4.1.2
|
8 |
+
Pillow
|
9 |
+
PyYAML>=5.3.1
|
10 |
+
scipy>=1.4.1
|
11 |
+
tensorboard>=1.5
|
12 |
+
torch==1.7.0
|
13 |
+
torchvision==0.8.1
|
14 |
+
tqdm>=4.41.0
|
15 |
+
|
16 |
+
# logging -------------------------------------
|
17 |
+
# wandb
|
18 |
+
|
19 |
+
# plotting ------------------------------------
|
20 |
+
seaborn>=0.11.0
|
21 |
+
pandas
|
22 |
+
|
23 |
+
# export --------------------------------------
|
24 |
+
# coremltools>=4.1
|
25 |
+
# onnx>=1.8.1
|
26 |
+
# scikit-learn==0.19.2 # for coreml quantization
|
27 |
+
|
28 |
+
# extras --------------------------------------
|
29 |
+
thop # FLOPS computation
|
30 |
+
pycocotools==2.0 # COCO mAP
|
31 |
+
|
32 |
+
|
33 |
+
gdown
|
scripts/get_coco.sh
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/bin/bash
|
2 |
+
# Script credit to https://github.com/ultralytics/yolov5
|
3 |
+
# COCO 2017 dataset http://cocodataset.org
|
4 |
+
# Download command: bash scripts/get_coco.sh
|
5 |
+
# Default dataset location is next to /yolor:
|
6 |
+
# /parent_folder
|
7 |
+
# /coco
|
8 |
+
# /yolor
|
9 |
+
|
10 |
+
# Download/unzip labels
|
11 |
+
d='../' # unzip directory
|
12 |
+
url=https://github.com/ultralytics/yolov5/releases/download/v1.0/
|
13 |
+
f='coco2017labels.zip' # or 'coco2017labels-segments.zip', 68 MB
|
14 |
+
echo 'Downloading' $url$f ' ...'
|
15 |
+
curl -L $url$f -o $f && unzip -q $f -d $d && rm $f & # download, unzip, remove in background
|
16 |
+
|
17 |
+
# Download/unzip images
|
18 |
+
d='../coco/images' # unzip directory
|
19 |
+
url=http://images.cocodataset.org/zips/
|
20 |
+
f1='train2017.zip' # 19G, 118k images
|
21 |
+
f2='val2017.zip' # 1G, 5k images
|
22 |
+
f3='test2017.zip' # 7G, 41k images (optional)
|
23 |
+
for f in $f1 $f2 $f3; do
|
24 |
+
echo 'Downloading' $url$f '...'
|
25 |
+
curl -L $url$f -o $f && unzip -q $f -d $d && rm $f & # download, unzip, remove in background
|
26 |
+
done
|
27 |
+
wait # finish background tasks
|
scripts/get_pretrain.sh
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
curl -c ./cookie -s -L "https://drive.google.com/uc?export=download&id=1Tdn3yqpZ79X7R1Ql0zNlNScB1Dv9Fp76" > /dev/null
|
2 |
+
curl -Lb ./cookie "https://drive.google.com/uc?export=download&confirm=`awk '/download/ {print $NF}' ./cookie`&id=1Tdn3yqpZ79X7R1Ql0zNlNScB1Dv9Fp76" -o yolor_p6.pt
|
3 |
+
rm ./cookie
|
4 |
+
|
5 |
+
curl -c ./cookie -s -L "https://drive.google.com/uc?export=download&id=1UflcHlN5ERPdhahMivQYCbWWw7d2wY7U" > /dev/null
|
6 |
+
curl -Lb ./cookie "https://drive.google.com/uc?export=download&confirm=`awk '/download/ {print $NF}' ./cookie`&id=1UflcHlN5ERPdhahMivQYCbWWw7d2wY7U" -o yolor_w6.pt
|
7 |
+
rm ./cookie
|
test.py
ADDED
@@ -0,0 +1,344 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import glob
|
3 |
+
import json
|
4 |
+
import os
|
5 |
+
from pathlib import Path
|
6 |
+
|
7 |
+
import numpy as np
|
8 |
+
import torch
|
9 |
+
import yaml
|
10 |
+
from tqdm import tqdm
|
11 |
+
|
12 |
+
from utils.google_utils import attempt_load
|
13 |
+
from utils.datasets import create_dataloader
|
14 |
+
from utils.general import coco80_to_coco91_class, check_dataset, check_file, check_img_size, box_iou, \
|
15 |
+
non_max_suppression, scale_coords, xyxy2xywh, xywh2xyxy, clip_coords, set_logging, increment_path
|
16 |
+
from utils.loss import compute_loss
|
17 |
+
from utils.metrics import ap_per_class
|
18 |
+
from utils.plots import plot_images, output_to_target
|
19 |
+
from utils.torch_utils import select_device, time_synchronized
|
20 |
+
|
21 |
+
from models.models import *
|
22 |
+
|
23 |
+
def load_classes(path):
|
24 |
+
# Loads *.names file at 'path'
|
25 |
+
with open(path, 'r') as f:
|
26 |
+
names = f.read().split('\n')
|
27 |
+
return list(filter(None, names)) # filter removes empty strings (such as last line)
|
28 |
+
|
29 |
+
|
30 |
+
def test(data,
|
31 |
+
weights=None,
|
32 |
+
batch_size=16,
|
33 |
+
imgsz=640,
|
34 |
+
conf_thres=0.001,
|
35 |
+
iou_thres=0.6, # for NMS
|
36 |
+
save_json=False,
|
37 |
+
single_cls=False,
|
38 |
+
augment=False,
|
39 |
+
verbose=False,
|
40 |
+
model=None,
|
41 |
+
dataloader=None,
|
42 |
+
save_dir=Path(''), # for saving images
|
43 |
+
save_txt=False, # for auto-labelling
|
44 |
+
save_conf=False,
|
45 |
+
plots=True,
|
46 |
+
log_imgs=0): # number of logged images
|
47 |
+
|
48 |
+
# Initialize/load model and set device
|
49 |
+
training = model is not None
|
50 |
+
if training: # called by train.py
|
51 |
+
device = next(model.parameters()).device # get model device
|
52 |
+
|
53 |
+
else: # called directly
|
54 |
+
set_logging()
|
55 |
+
device = select_device(opt.device, batch_size=batch_size)
|
56 |
+
save_txt = opt.save_txt # save *.txt labels
|
57 |
+
|
58 |
+
# Directories
|
59 |
+
save_dir = Path(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok)) # increment run
|
60 |
+
(save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
|
61 |
+
|
62 |
+
# Load model
|
63 |
+
model = Darknet(opt.cfg).to(device)
|
64 |
+
|
65 |
+
# load model
|
66 |
+
try:
|
67 |
+
ckpt = torch.load(weights[0], map_location=device) # load checkpoint
|
68 |
+
ckpt['model'] = {k: v for k, v in ckpt['model'].items() if model.state_dict()[k].numel() == v.numel()}
|
69 |
+
model.load_state_dict(ckpt['model'], strict=False)
|
70 |
+
except:
|
71 |
+
load_darknet_weights(model, weights[0])
|
72 |
+
imgsz = check_img_size(imgsz, s=64) # check img_size
|
73 |
+
|
74 |
+
# Half
|
75 |
+
half = device.type != 'cpu' # half precision only supported on CUDA
|
76 |
+
if half:
|
77 |
+
model.half()
|
78 |
+
|
79 |
+
# Configure
|
80 |
+
model.eval()
|
81 |
+
is_coco = data.endswith('coco.yaml') # is COCO dataset
|
82 |
+
with open(data) as f:
|
83 |
+
data = yaml.load(f, Loader=yaml.FullLoader) # model dict
|
84 |
+
check_dataset(data) # check
|
85 |
+
nc = 1 if single_cls else int(data['nc']) # number of classes
|
86 |
+
iouv = torch.linspace(0.5, 0.95, 10).to(device) # iou vector for mAP@0.5:0.95
|
87 |
+
niou = iouv.numel()
|
88 |
+
|
89 |
+
# Logging
|
90 |
+
log_imgs, wandb = min(log_imgs, 100), None # ceil
|
91 |
+
try:
|
92 |
+
import wandb # Weights & Biases
|
93 |
+
except ImportError:
|
94 |
+
log_imgs = 0
|
95 |
+
|
96 |
+
# Dataloader
|
97 |
+
if not training:
|
98 |
+
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
|
99 |
+
_ = model(img.half() if half else img) if device.type != 'cpu' else None # run once
|
100 |
+
path = data['test'] if opt.task == 'test' else data['val'] # path to val/test images
|
101 |
+
dataloader = create_dataloader(path, imgsz, batch_size, 64, opt, pad=0.5, rect=True)[0]
|
102 |
+
|
103 |
+
seen = 0
|
104 |
+
try:
|
105 |
+
names = model.names if hasattr(model, 'names') else model.module.names
|
106 |
+
except:
|
107 |
+
names = load_classes(opt.names)
|
108 |
+
coco91class = coco80_to_coco91_class()
|
109 |
+
s = ('%20s' + '%12s' * 6) % ('Class', 'Images', 'Targets', 'P', 'R', 'mAP@.5', 'mAP@.5:.95')
|
110 |
+
p, r, f1, mp, mr, map50, map, t0, t1 = 0., 0., 0., 0., 0., 0., 0., 0., 0.
|
111 |
+
loss = torch.zeros(3, device=device)
|
112 |
+
jdict, stats, ap, ap_class, wandb_images = [], [], [], [], []
|
113 |
+
for batch_i, (img, targets, paths, shapes) in enumerate(tqdm(dataloader, desc=s)):
|
114 |
+
img = img.to(device, non_blocking=True)
|
115 |
+
img = img.half() if half else img.float() # uint8 to fp16/32
|
116 |
+
img /= 255.0 # 0 - 255 to 0.0 - 1.0
|
117 |
+
targets = targets.to(device)
|
118 |
+
nb, _, height, width = img.shape # batch size, channels, height, width
|
119 |
+
whwh = torch.Tensor([width, height, width, height]).to(device)
|
120 |
+
|
121 |
+
# Disable gradients
|
122 |
+
with torch.no_grad():
|
123 |
+
# Run model
|
124 |
+
t = time_synchronized()
|
125 |
+
inf_out, train_out = model(img, augment=augment) # inference and training outputs
|
126 |
+
t0 += time_synchronized() - t
|
127 |
+
|
128 |
+
# Compute loss
|
129 |
+
if training: # if model has loss hyperparameters
|
130 |
+
loss += compute_loss([x.float() for x in train_out], targets, model)[1][:3] # box, obj, cls
|
131 |
+
|
132 |
+
# Run NMS
|
133 |
+
t = time_synchronized()
|
134 |
+
output = non_max_suppression(inf_out, conf_thres=conf_thres, iou_thres=iou_thres)
|
135 |
+
t1 += time_synchronized() - t
|
136 |
+
|
137 |
+
# Statistics per image
|
138 |
+
for si, pred in enumerate(output):
|
139 |
+
labels = targets[targets[:, 0] == si, 1:]
|
140 |
+
nl = len(labels)
|
141 |
+
tcls = labels[:, 0].tolist() if nl else [] # target class
|
142 |
+
seen += 1
|
143 |
+
|
144 |
+
if len(pred) == 0:
|
145 |
+
if nl:
|
146 |
+
stats.append((torch.zeros(0, niou, dtype=torch.bool), torch.Tensor(), torch.Tensor(), tcls))
|
147 |
+
continue
|
148 |
+
|
149 |
+
# Append to text file
|
150 |
+
path = Path(paths[si])
|
151 |
+
if save_txt:
|
152 |
+
gn = torch.tensor(shapes[si][0])[[1, 0, 1, 0]] # normalization gain whwh
|
153 |
+
x = pred.clone()
|
154 |
+
x[:, :4] = scale_coords(img[si].shape[1:], x[:, :4], shapes[si][0], shapes[si][1]) # to original
|
155 |
+
for *xyxy, conf, cls in x:
|
156 |
+
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
|
157 |
+
line = (cls, *xywh, conf) if save_conf else (cls, *xywh) # label format
|
158 |
+
with open(save_dir / 'labels' / (path.stem + '.txt'), 'a') as f:
|
159 |
+
f.write(('%g ' * len(line)).rstrip() % line + '\n')
|
160 |
+
|
161 |
+
# W&B logging
|
162 |
+
if plots and len(wandb_images) < log_imgs:
|
163 |
+
box_data = [{"position": {"minX": xyxy[0], "minY": xyxy[1], "maxX": xyxy[2], "maxY": xyxy[3]},
|
164 |
+
"class_id": int(cls),
|
165 |
+
"box_caption": "%s %.3f" % (names[cls], conf),
|
166 |
+
"scores": {"class_score": conf},
|
167 |
+
"domain": "pixel"} for *xyxy, conf, cls in pred.tolist()]
|
168 |
+
boxes = {"predictions": {"box_data": box_data, "class_labels": names}}
|
169 |
+
wandb_images.append(wandb.Image(img[si], boxes=boxes, caption=path.name))
|
170 |
+
|
171 |
+
# Clip boxes to image bounds
|
172 |
+
clip_coords(pred, (height, width))
|
173 |
+
|
174 |
+
# Append to pycocotools JSON dictionary
|
175 |
+
if save_json:
|
176 |
+
# [{"image_id": 42, "category_id": 18, "bbox": [258.15, 41.29, 348.26, 243.78], "score": 0.236}, ...
|
177 |
+
image_id = int(path.stem) if path.stem.isnumeric() else path.stem
|
178 |
+
box = pred[:, :4].clone() # xyxy
|
179 |
+
scale_coords(img[si].shape[1:], box, shapes[si][0], shapes[si][1]) # to original shape
|
180 |
+
box = xyxy2xywh(box) # xywh
|
181 |
+
box[:, :2] -= box[:, 2:] / 2 # xy center to top-left corner
|
182 |
+
for p, b in zip(pred.tolist(), box.tolist()):
|
183 |
+
jdict.append({'image_id': image_id,
|
184 |
+
'category_id': coco91class[int(p[5])] if is_coco else int(p[5]),
|
185 |
+
'bbox': [round(x, 3) for x in b],
|
186 |
+
'score': round(p[4], 5)})
|
187 |
+
|
188 |
+
# Assign all predictions as incorrect
|
189 |
+
correct = torch.zeros(pred.shape[0], niou, dtype=torch.bool, device=device)
|
190 |
+
if nl:
|
191 |
+
detected = [] # target indices
|
192 |
+
tcls_tensor = labels[:, 0]
|
193 |
+
|
194 |
+
# target boxes
|
195 |
+
tbox = xywh2xyxy(labels[:, 1:5]) * whwh
|
196 |
+
|
197 |
+
# Per target class
|
198 |
+
for cls in torch.unique(tcls_tensor):
|
199 |
+
ti = (cls == tcls_tensor).nonzero(as_tuple=False).view(-1) # prediction indices
|
200 |
+
pi = (cls == pred[:, 5]).nonzero(as_tuple=False).view(-1) # target indices
|
201 |
+
|
202 |
+
# Search for detections
|
203 |
+
if pi.shape[0]:
|
204 |
+
# Prediction to target ious
|
205 |
+
ious, i = box_iou(pred[pi, :4], tbox[ti]).max(1) # best ious, indices
|
206 |
+
|
207 |
+
# Append detections
|
208 |
+
detected_set = set()
|
209 |
+
for j in (ious > iouv[0]).nonzero(as_tuple=False):
|
210 |
+
d = ti[i[j]] # detected target
|
211 |
+
if d.item() not in detected_set:
|
212 |
+
detected_set.add(d.item())
|
213 |
+
detected.append(d)
|
214 |
+
correct[pi[j]] = ious[j] > iouv # iou_thres is 1xn
|
215 |
+
if len(detected) == nl: # all targets already located in image
|
216 |
+
break
|
217 |
+
|
218 |
+
# Append statistics (correct, conf, pcls, tcls)
|
219 |
+
stats.append((correct.cpu(), pred[:, 4].cpu(), pred[:, 5].cpu(), tcls))
|
220 |
+
|
221 |
+
# Plot images
|
222 |
+
if plots and batch_i < 3:
|
223 |
+
f = save_dir / f'test_batch{batch_i}_labels.jpg' # filename
|
224 |
+
plot_images(img, targets, paths, f, names) # labels
|
225 |
+
f = save_dir / f'test_batch{batch_i}_pred.jpg'
|
226 |
+
plot_images(img, output_to_target(output, width, height), paths, f, names) # predictions
|
227 |
+
|
228 |
+
# Compute statistics
|
229 |
+
stats = [np.concatenate(x, 0) for x in zip(*stats)] # to numpy
|
230 |
+
if len(stats) and stats[0].any():
|
231 |
+
p, r, ap, f1, ap_class = ap_per_class(*stats, plot=plots, fname=save_dir / 'precision-recall_curve.png')
|
232 |
+
p, r, ap50, ap = p[:, 0], r[:, 0], ap[:, 0], ap.mean(1) # [P, R, AP@0.5, AP@0.5:0.95]
|
233 |
+
mp, mr, map50, map = p.mean(), r.mean(), ap50.mean(), ap.mean()
|
234 |
+
nt = np.bincount(stats[3].astype(np.int64), minlength=nc) # number of targets per class
|
235 |
+
else:
|
236 |
+
nt = torch.zeros(1)
|
237 |
+
|
238 |
+
# W&B logging
|
239 |
+
if plots and wandb:
|
240 |
+
wandb.log({"Images": wandb_images})
|
241 |
+
wandb.log({"Validation": [wandb.Image(str(x), caption=x.name) for x in sorted(save_dir.glob('test*.jpg'))]})
|
242 |
+
|
243 |
+
# Print results
|
244 |
+
pf = '%20s' + '%12.3g' * 6 # print format
|
245 |
+
print(pf % ('all', seen, nt.sum(), mp, mr, map50, map))
|
246 |
+
|
247 |
+
# Print results per class
|
248 |
+
if verbose and nc > 1 and len(stats):
|
249 |
+
for i, c in enumerate(ap_class):
|
250 |
+
print(pf % (names[c], seen, nt[c], p[i], r[i], ap50[i], ap[i]))
|
251 |
+
|
252 |
+
# Print speeds
|
253 |
+
t = tuple(x / seen * 1E3 for x in (t0, t1, t0 + t1)) + (imgsz, imgsz, batch_size) # tuple
|
254 |
+
if not training:
|
255 |
+
print('Speed: %.1f/%.1f/%.1f ms inference/NMS/total per %gx%g image at batch-size %g' % t)
|
256 |
+
|
257 |
+
# Save JSON
|
258 |
+
if save_json and len(jdict):
|
259 |
+
w = Path(weights[0] if isinstance(weights, list) else weights).stem if weights is not None else '' # weights
|
260 |
+
anno_json = glob.glob('../coco/annotations/instances_val*.json')[0] # annotations json
|
261 |
+
pred_json = str(save_dir / f"{w}_predictions.json") # predictions json
|
262 |
+
print('\nEvaluating pycocotools mAP... saving %s...' % pred_json)
|
263 |
+
with open(pred_json, 'w') as f:
|
264 |
+
json.dump(jdict, f)
|
265 |
+
|
266 |
+
try: # https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocoEvalDemo.ipynb
|
267 |
+
from pycocotools.coco import COCO
|
268 |
+
from pycocotools.cocoeval import COCOeval
|
269 |
+
|
270 |
+
anno = COCO(anno_json) # init annotations api
|
271 |
+
pred = anno.loadRes(pred_json) # init predictions api
|
272 |
+
eval = COCOeval(anno, pred, 'bbox')
|
273 |
+
if is_coco:
|
274 |
+
eval.params.imgIds = [int(Path(x).stem) for x in dataloader.dataset.img_files] # image IDs to evaluate
|
275 |
+
eval.evaluate()
|
276 |
+
eval.accumulate()
|
277 |
+
eval.summarize()
|
278 |
+
map, map50 = eval.stats[:2] # update results (mAP@0.5:0.95, mAP@0.5)
|
279 |
+
except Exception as e:
|
280 |
+
print('ERROR: pycocotools unable to run: %s' % e)
|
281 |
+
|
282 |
+
# Return results
|
283 |
+
if not training:
|
284 |
+
print('Results saved to %s' % save_dir)
|
285 |
+
model.float() # for training
|
286 |
+
maps = np.zeros(nc) + map
|
287 |
+
for i, c in enumerate(ap_class):
|
288 |
+
maps[c] = ap[i]
|
289 |
+
return (mp, mr, map50, map, *(loss.cpu() / len(dataloader)).tolist()), maps, t
|
290 |
+
|
291 |
+
|
292 |
+
if __name__ == '__main__':
|
293 |
+
parser = argparse.ArgumentParser(prog='test.py')
|
294 |
+
parser.add_argument('--weights', nargs='+', type=str, default='yolor_p6.pt', help='model.pt path(s)')
|
295 |
+
parser.add_argument('--data', type=str, default='data/coco.yaml', help='*.data path')
|
296 |
+
parser.add_argument('--batch-size', type=int, default=32, help='size of each image batch')
|
297 |
+
parser.add_argument('--img-size', type=int, default=1280, help='inference size (pixels)')
|
298 |
+
parser.add_argument('--conf-thres', type=float, default=0.001, help='object confidence threshold')
|
299 |
+
parser.add_argument('--iou-thres', type=float, default=0.65, help='IOU threshold for NMS')
|
300 |
+
parser.add_argument('--task', default='val', help="'val', 'test', 'study'")
|
301 |
+
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
|
302 |
+
parser.add_argument('--single-cls', action='store_true', help='treat as single-class dataset')
|
303 |
+
parser.add_argument('--augment', action='store_true', help='augmented inference')
|
304 |
+
parser.add_argument('--verbose', action='store_true', help='report mAP by class')
|
305 |
+
parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
|
306 |
+
parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
|
307 |
+
parser.add_argument('--save-json', action='store_true', help='save a cocoapi-compatible JSON results file')
|
308 |
+
parser.add_argument('--project', default='runs/test', help='save to project/name')
|
309 |
+
parser.add_argument('--name', default='exp', help='save to project/name')
|
310 |
+
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
|
311 |
+
parser.add_argument('--cfg', type=str, default='cfg/yolor_p6.cfg', help='*.cfg path')
|
312 |
+
parser.add_argument('--names', type=str, default='data/coco.names', help='*.cfg path')
|
313 |
+
opt = parser.parse_args()
|
314 |
+
opt.save_json |= opt.data.endswith('coco.yaml')
|
315 |
+
opt.data = check_file(opt.data) # check file
|
316 |
+
print(opt)
|
317 |
+
|
318 |
+
if opt.task in ['val', 'test']: # run normally
|
319 |
+
test(opt.data,
|
320 |
+
opt.weights,
|
321 |
+
opt.batch_size,
|
322 |
+
opt.img_size,
|
323 |
+
opt.conf_thres,
|
324 |
+
opt.iou_thres,
|
325 |
+
opt.save_json,
|
326 |
+
opt.single_cls,
|
327 |
+
opt.augment,
|
328 |
+
opt.verbose,
|
329 |
+
save_txt=opt.save_txt,
|
330 |
+
save_conf=opt.save_conf,
|
331 |
+
)
|
332 |
+
|
333 |
+
elif opt.task == 'study': # run over a range of settings and save/plot
|
334 |
+
for weights in ['yolor_p6.pt', 'yolor_w6.pt']:
|
335 |
+
f = 'study_%s_%s.txt' % (Path(opt.data).stem, Path(weights).stem) # filename to save to
|
336 |
+
x = list(range(320, 800, 64)) # x axis
|
337 |
+
y = [] # y axis
|
338 |
+
for i in x: # img-size
|
339 |
+
print('\nRunning %s point %s...' % (f, i))
|
340 |
+
r, _, t = test(opt.data, weights, opt.batch_size, i, opt.conf_thres, opt.iou_thres, opt.save_json)
|
341 |
+
y.append(r + t) # results and times
|
342 |
+
np.savetxt(f, y, fmt='%10.4g') # save
|
343 |
+
os.system('zip -r study.zip study_*.txt')
|
344 |
+
# utils.general.plot_study_txt(f, x) # plot
|
train.py
ADDED
@@ -0,0 +1,619 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import logging
|
3 |
+
import math
|
4 |
+
import os
|
5 |
+
import random
|
6 |
+
import time
|
7 |
+
from pathlib import Path
|
8 |
+
from warnings import warn
|
9 |
+
|
10 |
+
import numpy as np
|
11 |
+
import torch.distributed as dist
|
12 |
+
import torch.nn as nn
|
13 |
+
import torch.nn.functional as F
|
14 |
+
import torch.optim as optim
|
15 |
+
import torch.optim.lr_scheduler as lr_scheduler
|
16 |
+
import torch.utils.data
|
17 |
+
import yaml
|
18 |
+
from torch.cuda import amp
|
19 |
+
from torch.nn.parallel import DistributedDataParallel as DDP
|
20 |
+
from torch.utils.tensorboard import SummaryWriter
|
21 |
+
from tqdm import tqdm
|
22 |
+
|
23 |
+
import test # import test.py to get mAP after each epoch
|
24 |
+
#from models.yolo import Model
|
25 |
+
from models.models import *
|
26 |
+
from utils.autoanchor import check_anchors
|
27 |
+
from utils.datasets import create_dataloader
|
28 |
+
from utils.general import labels_to_class_weights, increment_path, labels_to_image_weights, init_seeds, \
|
29 |
+
fitness, fitness_p, fitness_r, fitness_ap50, fitness_ap, fitness_f, strip_optimizer, get_latest_run,\
|
30 |
+
check_dataset, check_file, check_git_status, check_img_size, print_mutation, set_logging
|
31 |
+
from utils.google_utils import attempt_download
|
32 |
+
from utils.loss import compute_loss
|
33 |
+
from utils.plots import plot_images, plot_labels, plot_results, plot_evolution
|
34 |
+
from utils.torch_utils import ModelEMA, select_device, intersect_dicts, torch_distributed_zero_first
|
35 |
+
|
36 |
+
logger = logging.getLogger(__name__)
|
37 |
+
|
38 |
+
try:
|
39 |
+
import wandb
|
40 |
+
except ImportError:
|
41 |
+
wandb = None
|
42 |
+
logger.info("Install Weights & Biases for experiment logging via 'pip install wandb' (recommended)")
|
43 |
+
|
44 |
+
def train(hyp, opt, device, tb_writer=None, wandb=None):
|
45 |
+
logger.info(f'Hyperparameters {hyp}')
|
46 |
+
save_dir, epochs, batch_size, total_batch_size, weights, rank = \
|
47 |
+
Path(opt.save_dir), opt.epochs, opt.batch_size, opt.total_batch_size, opt.weights, opt.global_rank
|
48 |
+
|
49 |
+
# Directories
|
50 |
+
wdir = save_dir / 'weights'
|
51 |
+
wdir.mkdir(parents=True, exist_ok=True) # make dir
|
52 |
+
last = wdir / 'last.pt'
|
53 |
+
best = wdir / 'best.pt'
|
54 |
+
results_file = save_dir / 'results.txt'
|
55 |
+
|
56 |
+
# Save run settings
|
57 |
+
with open(save_dir / 'hyp.yaml', 'w') as f:
|
58 |
+
yaml.dump(hyp, f, sort_keys=False)
|
59 |
+
with open(save_dir / 'opt.yaml', 'w') as f:
|
60 |
+
yaml.dump(vars(opt), f, sort_keys=False)
|
61 |
+
|
62 |
+
# Configure
|
63 |
+
plots = not opt.evolve # create plots
|
64 |
+
cuda = device.type != 'cpu'
|
65 |
+
init_seeds(2 + rank)
|
66 |
+
with open(opt.data) as f:
|
67 |
+
data_dict = yaml.load(f, Loader=yaml.FullLoader) # data dict
|
68 |
+
with torch_distributed_zero_first(rank):
|
69 |
+
check_dataset(data_dict) # check
|
70 |
+
train_path = data_dict['train']
|
71 |
+
test_path = data_dict['val']
|
72 |
+
nc, names = (1, ['item']) if opt.single_cls else (int(data_dict['nc']), data_dict['names']) # number classes, names
|
73 |
+
assert len(names) == nc, '%g names found for nc=%g dataset in %s' % (len(names), nc, opt.data) # check
|
74 |
+
|
75 |
+
# Model
|
76 |
+
pretrained = weights.endswith('.pt')
|
77 |
+
if pretrained:
|
78 |
+
with torch_distributed_zero_first(rank):
|
79 |
+
attempt_download(weights) # download if not found locally
|
80 |
+
ckpt = torch.load(weights, map_location=device) # load checkpoint
|
81 |
+
model = Darknet(opt.cfg).to(device) # create
|
82 |
+
state_dict = {k: v for k, v in ckpt['model'].items() if model.state_dict()[k].numel() == v.numel()}
|
83 |
+
model.load_state_dict(state_dict, strict=False)
|
84 |
+
print('Transferred %g/%g items from %s' % (len(state_dict), len(model.state_dict()), weights)) # report
|
85 |
+
else:
|
86 |
+
model = Darknet(opt.cfg).to(device) # create
|
87 |
+
|
88 |
+
# Optimizer
|
89 |
+
nbs = 64 # nominal batch size
|
90 |
+
accumulate = max(round(nbs / total_batch_size), 1) # accumulate loss before optimizing
|
91 |
+
hyp['weight_decay'] *= total_batch_size * accumulate / nbs # scale weight_decay
|
92 |
+
|
93 |
+
pg0, pg1, pg2 = [], [], [] # optimizer parameter groups
|
94 |
+
for k, v in dict(model.named_parameters()).items():
|
95 |
+
if '.bias' in k:
|
96 |
+
pg2.append(v) # biases
|
97 |
+
elif 'Conv2d.weight' in k:
|
98 |
+
pg1.append(v) # apply weight_decay
|
99 |
+
elif 'm.weight' in k:
|
100 |
+
pg1.append(v) # apply weight_decay
|
101 |
+
elif 'w.weight' in k:
|
102 |
+
pg1.append(v) # apply weight_decay
|
103 |
+
else:
|
104 |
+
pg0.append(v) # all else
|
105 |
+
|
106 |
+
if opt.adam:
|
107 |
+
optimizer = optim.Adam(pg0, lr=hyp['lr0'], betas=(hyp['momentum'], 0.999)) # adjust beta1 to momentum
|
108 |
+
else:
|
109 |
+
optimizer = optim.SGD(pg0, lr=hyp['lr0'], momentum=hyp['momentum'], nesterov=True)
|
110 |
+
|
111 |
+
optimizer.add_param_group({'params': pg1, 'weight_decay': hyp['weight_decay']}) # add pg1 with weight_decay
|
112 |
+
optimizer.add_param_group({'params': pg2}) # add pg2 (biases)
|
113 |
+
logger.info('Optimizer groups: %g .bias, %g conv.weight, %g other' % (len(pg2), len(pg1), len(pg0)))
|
114 |
+
del pg0, pg1, pg2
|
115 |
+
|
116 |
+
# Scheduler https://arxiv.org/pdf/1812.01187.pdf
|
117 |
+
# https://pytorch.org/docs/stable/_modules/torch/optim/lr_scheduler.html#OneCycleLR
|
118 |
+
lf = lambda x: ((1 + math.cos(x * math.pi / epochs)) / 2) * (1 - hyp['lrf']) + hyp['lrf'] # cosine
|
119 |
+
scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf)
|
120 |
+
# plot_lr_scheduler(optimizer, scheduler, epochs)
|
121 |
+
|
122 |
+
# Logging
|
123 |
+
if wandb and wandb.run is None:
|
124 |
+
opt.hyp = hyp # add hyperparameters
|
125 |
+
wandb_run = wandb.init(config=opt, resume="allow",
|
126 |
+
project='YOLOR' if opt.project == 'runs/train' else Path(opt.project).stem,
|
127 |
+
name=save_dir.stem,
|
128 |
+
id=ckpt.get('wandb_id') if 'ckpt' in locals() else None)
|
129 |
+
|
130 |
+
# Resume
|
131 |
+
start_epoch, best_fitness = 0, 0.0
|
132 |
+
best_fitness_p, best_fitness_r, best_fitness_ap50, best_fitness_ap, best_fitness_f = 0.0, 0.0, 0.0, 0.0, 0.0
|
133 |
+
if pretrained:
|
134 |
+
# Optimizer
|
135 |
+
if ckpt['optimizer'] is not None:
|
136 |
+
optimizer.load_state_dict(ckpt['optimizer'])
|
137 |
+
best_fitness = ckpt['best_fitness']
|
138 |
+
best_fitness_p = ckpt['best_fitness_p']
|
139 |
+
best_fitness_r = ckpt['best_fitness_r']
|
140 |
+
best_fitness_ap50 = ckpt['best_fitness_ap50']
|
141 |
+
best_fitness_ap = ckpt['best_fitness_ap']
|
142 |
+
best_fitness_f = ckpt['best_fitness_f']
|
143 |
+
|
144 |
+
# Results
|
145 |
+
if ckpt.get('training_results') is not None:
|
146 |
+
with open(results_file, 'w') as file:
|
147 |
+
file.write(ckpt['training_results']) # write results.txt
|
148 |
+
|
149 |
+
# Epochs
|
150 |
+
start_epoch = ckpt['epoch'] + 1
|
151 |
+
if opt.resume:
|
152 |
+
assert start_epoch > 0, '%s training to %g epochs is finished, nothing to resume.' % (weights, epochs)
|
153 |
+
if epochs < start_epoch:
|
154 |
+
logger.info('%s has been trained for %g epochs. Fine-tuning for %g additional epochs.' %
|
155 |
+
(weights, ckpt['epoch'], epochs))
|
156 |
+
epochs += ckpt['epoch'] # finetune additional epochs
|
157 |
+
|
158 |
+
del ckpt, state_dict
|
159 |
+
|
160 |
+
# Image sizes
|
161 |
+
gs = 64 #int(max(model.stride)) # grid size (max stride)
|
162 |
+
imgsz, imgsz_test = [check_img_size(x, gs) for x in opt.img_size] # verify imgsz are gs-multiples
|
163 |
+
|
164 |
+
# DP mode
|
165 |
+
if cuda and rank == -1 and torch.cuda.device_count() > 1:
|
166 |
+
model = torch.nn.DataParallel(model)
|
167 |
+
|
168 |
+
# SyncBatchNorm
|
169 |
+
if opt.sync_bn and cuda and rank != -1:
|
170 |
+
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device)
|
171 |
+
logger.info('Using SyncBatchNorm()')
|
172 |
+
|
173 |
+
# EMA
|
174 |
+
ema = ModelEMA(model) if rank in [-1, 0] else None
|
175 |
+
|
176 |
+
# DDP mode
|
177 |
+
if cuda and rank != -1:
|
178 |
+
model = DDP(model, device_ids=[opt.local_rank], output_device=opt.local_rank)
|
179 |
+
|
180 |
+
# Trainloader
|
181 |
+
dataloader, dataset = create_dataloader(train_path, imgsz, batch_size, gs, opt,
|
182 |
+
hyp=hyp, augment=True, cache=opt.cache_images, rect=opt.rect,
|
183 |
+
rank=rank, world_size=opt.world_size, workers=opt.workers)
|
184 |
+
mlc = np.concatenate(dataset.labels, 0)[:, 0].max() # max label class
|
185 |
+
nb = len(dataloader) # number of batches
|
186 |
+
assert mlc < nc, 'Label class %g exceeds nc=%g in %s. Possible class labels are 0-%g' % (mlc, nc, opt.data, nc - 1)
|
187 |
+
|
188 |
+
# Process 0
|
189 |
+
if rank in [-1, 0]:
|
190 |
+
ema.updates = start_epoch * nb // accumulate # set EMA updates
|
191 |
+
testloader = create_dataloader(test_path, imgsz_test, batch_size*2, gs, opt,
|
192 |
+
hyp=hyp, cache=opt.cache_images and not opt.notest, rect=True,
|
193 |
+
rank=-1, world_size=opt.world_size, workers=opt.workers)[0] # testloader
|
194 |
+
|
195 |
+
if not opt.resume:
|
196 |
+
labels = np.concatenate(dataset.labels, 0)
|
197 |
+
c = torch.tensor(labels[:, 0]) # classes
|
198 |
+
# cf = torch.bincount(c.long(), minlength=nc) + 1. # frequency
|
199 |
+
# model._initialize_biases(cf.to(device))
|
200 |
+
if plots:
|
201 |
+
plot_labels(labels, save_dir=save_dir)
|
202 |
+
if tb_writer:
|
203 |
+
tb_writer.add_histogram('classes', c, 0)
|
204 |
+
if wandb:
|
205 |
+
wandb.log({"Labels": [wandb.Image(str(x), caption=x.name) for x in save_dir.glob('*labels*.png')]})
|
206 |
+
|
207 |
+
# Anchors
|
208 |
+
# if not opt.noautoanchor:
|
209 |
+
# check_anchors(dataset, model=model, thr=hyp['anchor_t'], imgsz=imgsz)
|
210 |
+
|
211 |
+
# Model parameters
|
212 |
+
hyp['cls'] *= nc / 80. # scale coco-tuned hyp['cls'] to current dataset
|
213 |
+
model.nc = nc # attach number of classes to model
|
214 |
+
model.hyp = hyp # attach hyperparameters to model
|
215 |
+
model.gr = 1.0 # iou loss ratio (obj_loss = 1.0 or iou)
|
216 |
+
model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) # attach class weights
|
217 |
+
model.names = names
|
218 |
+
|
219 |
+
# Start training
|
220 |
+
t0 = time.time()
|
221 |
+
nw = max(round(hyp['warmup_epochs'] * nb), 1000) # number of warmup iterations, max(3 epochs, 1k iterations)
|
222 |
+
# nw = min(nw, (epochs - start_epoch) / 2 * nb) # limit warmup to < 1/2 of training
|
223 |
+
maps = np.zeros(nc) # mAP per class
|
224 |
+
results = (0, 0, 0, 0, 0, 0, 0) # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
|
225 |
+
scheduler.last_epoch = start_epoch - 1 # do not move
|
226 |
+
scaler = amp.GradScaler(enabled=cuda)
|
227 |
+
logger.info('Image sizes %g train, %g test\n'
|
228 |
+
'Using %g dataloader workers\nLogging results to %s\n'
|
229 |
+
'Starting training for %g epochs...' % (imgsz, imgsz_test, dataloader.num_workers, save_dir, epochs))
|
230 |
+
|
231 |
+
torch.save(model, wdir / 'init.pt')
|
232 |
+
|
233 |
+
for epoch in range(start_epoch, epochs): # epoch ------------------------------------------------------------------
|
234 |
+
model.train()
|
235 |
+
|
236 |
+
# Update image weights (optional)
|
237 |
+
if opt.image_weights:
|
238 |
+
# Generate indices
|
239 |
+
if rank in [-1, 0]:
|
240 |
+
cw = model.class_weights.cpu().numpy() * (1 - maps) ** 2 # class weights
|
241 |
+
iw = labels_to_image_weights(dataset.labels, nc=nc, class_weights=cw) # image weights
|
242 |
+
dataset.indices = random.choices(range(dataset.n), weights=iw, k=dataset.n) # rand weighted idx
|
243 |
+
# Broadcast if DDP
|
244 |
+
if rank != -1:
|
245 |
+
indices = (torch.tensor(dataset.indices) if rank == 0 else torch.zeros(dataset.n)).int()
|
246 |
+
dist.broadcast(indices, 0)
|
247 |
+
if rank != 0:
|
248 |
+
dataset.indices = indices.cpu().numpy()
|
249 |
+
|
250 |
+
# Update mosaic border
|
251 |
+
# b = int(random.uniform(0.25 * imgsz, 0.75 * imgsz + gs) // gs * gs)
|
252 |
+
# dataset.mosaic_border = [b - imgsz, -b] # height, width borders
|
253 |
+
|
254 |
+
mloss = torch.zeros(4, device=device) # mean losses
|
255 |
+
if rank != -1:
|
256 |
+
dataloader.sampler.set_epoch(epoch)
|
257 |
+
pbar = enumerate(dataloader)
|
258 |
+
logger.info(('\n' + '%10s' * 8) % ('Epoch', 'gpu_mem', 'box', 'obj', 'cls', 'total', 'targets', 'img_size'))
|
259 |
+
if rank in [-1, 0]:
|
260 |
+
pbar = tqdm(pbar, total=nb) # progress bar
|
261 |
+
optimizer.zero_grad()
|
262 |
+
for i, (imgs, targets, paths, _) in pbar: # batch -------------------------------------------------------------
|
263 |
+
ni = i + nb * epoch # number integrated batches (since train start)
|
264 |
+
imgs = imgs.to(device, non_blocking=True).float() / 255.0 # uint8 to float32, 0-255 to 0.0-1.0
|
265 |
+
|
266 |
+
# Warmup
|
267 |
+
if ni <= nw:
|
268 |
+
xi = [0, nw] # x interp
|
269 |
+
# model.gr = np.interp(ni, xi, [0.0, 1.0]) # iou loss ratio (obj_loss = 1.0 or iou)
|
270 |
+
accumulate = max(1, np.interp(ni, xi, [1, nbs / total_batch_size]).round())
|
271 |
+
for j, x in enumerate(optimizer.param_groups):
|
272 |
+
# bias lr falls from 0.1 to lr0, all other lrs rise from 0.0 to lr0
|
273 |
+
x['lr'] = np.interp(ni, xi, [hyp['warmup_bias_lr'] if j == 2 else 0.0, x['initial_lr'] * lf(epoch)])
|
274 |
+
if 'momentum' in x:
|
275 |
+
x['momentum'] = np.interp(ni, xi, [hyp['warmup_momentum'], hyp['momentum']])
|
276 |
+
|
277 |
+
# Multi-scale
|
278 |
+
if opt.multi_scale:
|
279 |
+
sz = random.randrange(imgsz * 0.5, imgsz * 1.5 + gs) // gs * gs # size
|
280 |
+
sf = sz / max(imgs.shape[2:]) # scale factor
|
281 |
+
if sf != 1:
|
282 |
+
ns = [math.ceil(x * sf / gs) * gs for x in imgs.shape[2:]] # new shape (stretched to gs-multiple)
|
283 |
+
imgs = F.interpolate(imgs, size=ns, mode='bilinear', align_corners=False)
|
284 |
+
|
285 |
+
# Forward
|
286 |
+
with amp.autocast(enabled=cuda):
|
287 |
+
pred = model(imgs) # forward
|
288 |
+
loss, loss_items = compute_loss(pred, targets.to(device), model) # loss scaled by batch_size
|
289 |
+
if rank != -1:
|
290 |
+
loss *= opt.world_size # gradient averaged between devices in DDP mode
|
291 |
+
|
292 |
+
# Backward
|
293 |
+
scaler.scale(loss).backward()
|
294 |
+
|
295 |
+
# Optimize
|
296 |
+
if ni % accumulate == 0:
|
297 |
+
scaler.step(optimizer) # optimizer.step
|
298 |
+
scaler.update()
|
299 |
+
optimizer.zero_grad()
|
300 |
+
if ema:
|
301 |
+
ema.update(model)
|
302 |
+
|
303 |
+
# Print
|
304 |
+
if rank in [-1, 0]:
|
305 |
+
mloss = (mloss * i + loss_items) / (i + 1) # update mean losses
|
306 |
+
mem = '%.3gG' % (torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0) # (GB)
|
307 |
+
s = ('%10s' * 2 + '%10.4g' * 6) % (
|
308 |
+
'%g/%g' % (epoch, epochs - 1), mem, *mloss, targets.shape[0], imgs.shape[-1])
|
309 |
+
pbar.set_description(s)
|
310 |
+
|
311 |
+
# Plot
|
312 |
+
if plots and ni < 3:
|
313 |
+
f = save_dir / f'train_batch{ni}.jpg' # filename
|
314 |
+
plot_images(images=imgs, targets=targets, paths=paths, fname=f)
|
315 |
+
# if tb_writer:
|
316 |
+
# tb_writer.add_image(f, result, dataformats='HWC', global_step=epoch)
|
317 |
+
# tb_writer.add_graph(model, imgs) # add model to tensorboard
|
318 |
+
elif plots and ni == 3 and wandb:
|
319 |
+
wandb.log({"Mosaics": [wandb.Image(str(x), caption=x.name) for x in save_dir.glob('train*.jpg')]})
|
320 |
+
|
321 |
+
# end batch ------------------------------------------------------------------------------------------------
|
322 |
+
# end epoch ----------------------------------------------------------------------------------------------------
|
323 |
+
|
324 |
+
# Scheduler
|
325 |
+
lr = [x['lr'] for x in optimizer.param_groups] # for tensorboard
|
326 |
+
scheduler.step()
|
327 |
+
|
328 |
+
# DDP process 0 or single-GPU
|
329 |
+
if rank in [-1, 0]:
|
330 |
+
# mAP
|
331 |
+
if ema:
|
332 |
+
ema.update_attr(model)
|
333 |
+
final_epoch = epoch + 1 == epochs
|
334 |
+
if not opt.notest or final_epoch: # Calculate mAP
|
335 |
+
if epoch >= 3:
|
336 |
+
results, maps, times = test.test(opt.data,
|
337 |
+
batch_size=batch_size*2,
|
338 |
+
imgsz=imgsz_test,
|
339 |
+
model=ema.ema.module if hasattr(ema.ema, 'module') else ema.ema,
|
340 |
+
single_cls=opt.single_cls,
|
341 |
+
dataloader=testloader,
|
342 |
+
save_dir=save_dir,
|
343 |
+
plots=plots and final_epoch,
|
344 |
+
log_imgs=opt.log_imgs if wandb else 0)
|
345 |
+
|
346 |
+
# Write
|
347 |
+
with open(results_file, 'a') as f:
|
348 |
+
f.write(s + '%10.4g' * 7 % results + '\n') # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
|
349 |
+
if len(opt.name) and opt.bucket:
|
350 |
+
os.system('gsutil cp %s gs://%s/results/results%s.txt' % (results_file, opt.bucket, opt.name))
|
351 |
+
|
352 |
+
# Log
|
353 |
+
tags = ['train/box_loss', 'train/obj_loss', 'train/cls_loss', # train loss
|
354 |
+
'metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/mAP_0.5:0.95',
|
355 |
+
'val/box_loss', 'val/obj_loss', 'val/cls_loss', # val loss
|
356 |
+
'x/lr0', 'x/lr1', 'x/lr2'] # params
|
357 |
+
for x, tag in zip(list(mloss[:-1]) + list(results) + lr, tags):
|
358 |
+
if tb_writer:
|
359 |
+
tb_writer.add_scalar(tag, x, epoch) # tensorboard
|
360 |
+
if wandb:
|
361 |
+
wandb.log({tag: x}) # W&B
|
362 |
+
|
363 |
+
# Update best mAP
|
364 |
+
fi = fitness(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
365 |
+
fi_p = fitness_p(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
366 |
+
fi_r = fitness_r(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
367 |
+
fi_ap50 = fitness_ap50(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
368 |
+
fi_ap = fitness_ap(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
369 |
+
if (fi_p > 0.0) or (fi_r > 0.0):
|
370 |
+
fi_f = fitness_f(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
371 |
+
else:
|
372 |
+
fi_f = 0.0
|
373 |
+
if fi > best_fitness:
|
374 |
+
best_fitness = fi
|
375 |
+
if fi_p > best_fitness_p:
|
376 |
+
best_fitness_p = fi_p
|
377 |
+
if fi_r > best_fitness_r:
|
378 |
+
best_fitness_r = fi_r
|
379 |
+
if fi_ap50 > best_fitness_ap50:
|
380 |
+
best_fitness_ap50 = fi_ap50
|
381 |
+
if fi_ap > best_fitness_ap:
|
382 |
+
best_fitness_ap = fi_ap
|
383 |
+
if fi_f > best_fitness_f:
|
384 |
+
best_fitness_f = fi_f
|
385 |
+
|
386 |
+
# Save model
|
387 |
+
save = (not opt.nosave) or (final_epoch and not opt.evolve)
|
388 |
+
if save:
|
389 |
+
with open(results_file, 'r') as f: # create checkpoint
|
390 |
+
ckpt = {'epoch': epoch,
|
391 |
+
'best_fitness': best_fitness,
|
392 |
+
'best_fitness_p': best_fitness_p,
|
393 |
+
'best_fitness_r': best_fitness_r,
|
394 |
+
'best_fitness_ap50': best_fitness_ap50,
|
395 |
+
'best_fitness_ap': best_fitness_ap,
|
396 |
+
'best_fitness_f': best_fitness_f,
|
397 |
+
'training_results': f.read(),
|
398 |
+
'model': ema.ema.module.state_dict() if hasattr(ema, 'module') else ema.ema.state_dict(),
|
399 |
+
'optimizer': None if final_epoch else optimizer.state_dict(),
|
400 |
+
'wandb_id': wandb_run.id if wandb else None}
|
401 |
+
|
402 |
+
# Save last, best and delete
|
403 |
+
torch.save(ckpt, last)
|
404 |
+
if best_fitness == fi:
|
405 |
+
torch.save(ckpt, best)
|
406 |
+
if (best_fitness == fi) and (epoch >= 200):
|
407 |
+
torch.save(ckpt, wdir / 'best_{:03d}.pt'.format(epoch))
|
408 |
+
if best_fitness == fi:
|
409 |
+
torch.save(ckpt, wdir / 'best_overall.pt')
|
410 |
+
if best_fitness_p == fi_p:
|
411 |
+
torch.save(ckpt, wdir / 'best_p.pt')
|
412 |
+
if best_fitness_r == fi_r:
|
413 |
+
torch.save(ckpt, wdir / 'best_r.pt')
|
414 |
+
if best_fitness_ap50 == fi_ap50:
|
415 |
+
torch.save(ckpt, wdir / 'best_ap50.pt')
|
416 |
+
if best_fitness_ap == fi_ap:
|
417 |
+
torch.save(ckpt, wdir / 'best_ap.pt')
|
418 |
+
if best_fitness_f == fi_f:
|
419 |
+
torch.save(ckpt, wdir / 'best_f.pt')
|
420 |
+
if epoch == 0:
|
421 |
+
torch.save(ckpt, wdir / 'epoch_{:03d}.pt'.format(epoch))
|
422 |
+
if ((epoch+1) % 25) == 0:
|
423 |
+
torch.save(ckpt, wdir / 'epoch_{:03d}.pt'.format(epoch))
|
424 |
+
if epoch >= (epochs-5):
|
425 |
+
torch.save(ckpt, wdir / 'last_{:03d}.pt'.format(epoch))
|
426 |
+
elif epoch >= 420:
|
427 |
+
torch.save(ckpt, wdir / 'last_{:03d}.pt'.format(epoch))
|
428 |
+
del ckpt
|
429 |
+
# end epoch ----------------------------------------------------------------------------------------------------
|
430 |
+
# end training
|
431 |
+
|
432 |
+
if rank in [-1, 0]:
|
433 |
+
# Strip optimizers
|
434 |
+
n = opt.name if opt.name.isnumeric() else ''
|
435 |
+
fresults, flast, fbest = save_dir / f'results{n}.txt', wdir / f'last{n}.pt', wdir / f'best{n}.pt'
|
436 |
+
for f1, f2 in zip([wdir / 'last.pt', wdir / 'best.pt', results_file], [flast, fbest, fresults]):
|
437 |
+
if f1.exists():
|
438 |
+
os.rename(f1, f2) # rename
|
439 |
+
if str(f2).endswith('.pt'): # is *.pt
|
440 |
+
strip_optimizer(f2) # strip optimizer
|
441 |
+
os.system('gsutil cp %s gs://%s/weights' % (f2, opt.bucket)) if opt.bucket else None # upload
|
442 |
+
# Finish
|
443 |
+
if plots:
|
444 |
+
plot_results(save_dir=save_dir) # save as results.png
|
445 |
+
if wandb:
|
446 |
+
wandb.log({"Results": [wandb.Image(str(save_dir / x), caption=x) for x in
|
447 |
+
['results.png', 'precision-recall_curve.png']]})
|
448 |
+
logger.info('%g epochs completed in %.3f hours.\n' % (epoch - start_epoch + 1, (time.time() - t0) / 3600))
|
449 |
+
else:
|
450 |
+
dist.destroy_process_group()
|
451 |
+
|
452 |
+
wandb.run.finish() if wandb and wandb.run else None
|
453 |
+
torch.cuda.empty_cache()
|
454 |
+
return results
|
455 |
+
|
456 |
+
|
457 |
+
if __name__ == '__main__':
|
458 |
+
parser = argparse.ArgumentParser()
|
459 |
+
parser.add_argument('--weights', type=str, default='yolor_p6.pt', help='initial weights path')
|
460 |
+
parser.add_argument('--cfg', type=str, default='', help='model.yaml path')
|
461 |
+
parser.add_argument('--data', type=str, default='data/coco.yaml', help='data.yaml path')
|
462 |
+
parser.add_argument('--hyp', type=str, default='data/hyp.scratch.1280.yaml', help='hyperparameters path')
|
463 |
+
parser.add_argument('--epochs', type=int, default=300)
|
464 |
+
parser.add_argument('--batch-size', type=int, default=8, help='total batch size for all GPUs')
|
465 |
+
parser.add_argument('--img-size', nargs='+', type=int, default=[1280, 1280], help='[train, test] image sizes')
|
466 |
+
parser.add_argument('--rect', action='store_true', help='rectangular training')
|
467 |
+
parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')
|
468 |
+
parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')
|
469 |
+
parser.add_argument('--notest', action='store_true', help='only test final epoch')
|
470 |
+
parser.add_argument('--noautoanchor', action='store_true', help='disable autoanchor check')
|
471 |
+
parser.add_argument('--evolve', action='store_true', help='evolve hyperparameters')
|
472 |
+
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
|
473 |
+
parser.add_argument('--cache-images', action='store_true', help='cache images for faster training')
|
474 |
+
parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')
|
475 |
+
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
|
476 |
+
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
|
477 |
+
parser.add_argument('--single-cls', action='store_true', help='train as single-class dataset')
|
478 |
+
parser.add_argument('--adam', action='store_true', help='use torch.optim.Adam() optimizer')
|
479 |
+
parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')
|
480 |
+
parser.add_argument('--local_rank', type=int, default=-1, help='DDP parameter, do not modify')
|
481 |
+
parser.add_argument('--log-imgs', type=int, default=16, help='number of images for W&B logging, max 100')
|
482 |
+
parser.add_argument('--workers', type=int, default=8, help='maximum number of dataloader workers')
|
483 |
+
parser.add_argument('--project', default='runs/train', help='save to project/name')
|
484 |
+
parser.add_argument('--name', default='exp', help='save to project/name')
|
485 |
+
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
|
486 |
+
opt = parser.parse_args()
|
487 |
+
|
488 |
+
# Set DDP variables
|
489 |
+
opt.total_batch_size = opt.batch_size
|
490 |
+
opt.world_size = int(os.environ['WORLD_SIZE']) if 'WORLD_SIZE' in os.environ else 1
|
491 |
+
opt.global_rank = int(os.environ['RANK']) if 'RANK' in os.environ else -1
|
492 |
+
set_logging(opt.global_rank)
|
493 |
+
if opt.global_rank in [-1, 0]:
|
494 |
+
check_git_status()
|
495 |
+
|
496 |
+
# Resume
|
497 |
+
if opt.resume: # resume an interrupted run
|
498 |
+
ckpt = opt.resume if isinstance(opt.resume, str) else get_latest_run() # specified or most recent path
|
499 |
+
assert os.path.isfile(ckpt), 'ERROR: --resume checkpoint does not exist'
|
500 |
+
with open(Path(ckpt).parent.parent / 'opt.yaml') as f:
|
501 |
+
opt = argparse.Namespace(**yaml.load(f, Loader=yaml.FullLoader)) # replace
|
502 |
+
opt.cfg, opt.weights, opt.resume = '', ckpt, True
|
503 |
+
logger.info('Resuming training from %s' % ckpt)
|
504 |
+
else:
|
505 |
+
# opt.hyp = opt.hyp or ('hyp.finetune.yaml' if opt.weights else 'hyp.scratch.yaml')
|
506 |
+
opt.data, opt.cfg, opt.hyp = check_file(opt.data), check_file(opt.cfg), check_file(opt.hyp) # check files
|
507 |
+
assert len(opt.cfg) or len(opt.weights), 'either --cfg or --weights must be specified'
|
508 |
+
opt.img_size.extend([opt.img_size[-1]] * (2 - len(opt.img_size))) # extend to 2 sizes (train, test)
|
509 |
+
opt.name = 'evolve' if opt.evolve else opt.name
|
510 |
+
opt.save_dir = increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok | opt.evolve) # increment run
|
511 |
+
|
512 |
+
# DDP mode
|
513 |
+
device = select_device(opt.device, batch_size=opt.batch_size)
|
514 |
+
if opt.local_rank != -1:
|
515 |
+
assert torch.cuda.device_count() > opt.local_rank
|
516 |
+
torch.cuda.set_device(opt.local_rank)
|
517 |
+
device = torch.device('cuda', opt.local_rank)
|
518 |
+
dist.init_process_group(backend='nccl', init_method='env://') # distributed backend
|
519 |
+
assert opt.batch_size % opt.world_size == 0, '--batch-size must be multiple of CUDA device count'
|
520 |
+
opt.batch_size = opt.total_batch_size // opt.world_size
|
521 |
+
|
522 |
+
# Hyperparameters
|
523 |
+
with open(opt.hyp) as f:
|
524 |
+
hyp = yaml.load(f, Loader=yaml.FullLoader) # load hyps
|
525 |
+
if 'box' not in hyp:
|
526 |
+
warn('Compatibility: %s missing "box" which was renamed from "giou" in %s' %
|
527 |
+
(opt.hyp, 'https://github.com/ultralytics/yolov5/pull/1120'))
|
528 |
+
hyp['box'] = hyp.pop('giou')
|
529 |
+
|
530 |
+
# Train
|
531 |
+
logger.info(opt)
|
532 |
+
if not opt.evolve:
|
533 |
+
tb_writer = None # init loggers
|
534 |
+
if opt.global_rank in [-1, 0]:
|
535 |
+
logger.info(f'Start Tensorboard with "tensorboard --logdir {opt.project}", view at http://localhost:6006/')
|
536 |
+
tb_writer = SummaryWriter(opt.save_dir) # Tensorboard
|
537 |
+
train(hyp, opt, device, tb_writer, wandb)
|
538 |
+
|
539 |
+
# Evolve hyperparameters (optional)
|
540 |
+
else:
|
541 |
+
# Hyperparameter evolution metadata (mutation scale 0-1, lower_limit, upper_limit)
|
542 |
+
meta = {'lr0': (1, 1e-5, 1e-1), # initial learning rate (SGD=1E-2, Adam=1E-3)
|
543 |
+
'lrf': (1, 0.01, 1.0), # final OneCycleLR learning rate (lr0 * lrf)
|
544 |
+
'momentum': (0.3, 0.6, 0.98), # SGD momentum/Adam beta1
|
545 |
+
'weight_decay': (1, 0.0, 0.001), # optimizer weight decay
|
546 |
+
'warmup_epochs': (1, 0.0, 5.0), # warmup epochs (fractions ok)
|
547 |
+
'warmup_momentum': (1, 0.0, 0.95), # warmup initial momentum
|
548 |
+
'warmup_bias_lr': (1, 0.0, 0.2), # warmup initial bias lr
|
549 |
+
'box': (1, 0.02, 0.2), # box loss gain
|
550 |
+
'cls': (1, 0.2, 4.0), # cls loss gain
|
551 |
+
'cls_pw': (1, 0.5, 2.0), # cls BCELoss positive_weight
|
552 |
+
'obj': (1, 0.2, 4.0), # obj loss gain (scale with pixels)
|
553 |
+
'obj_pw': (1, 0.5, 2.0), # obj BCELoss positive_weight
|
554 |
+
'iou_t': (0, 0.1, 0.7), # IoU training threshold
|
555 |
+
'anchor_t': (1, 2.0, 8.0), # anchor-multiple threshold
|
556 |
+
'anchors': (2, 2.0, 10.0), # anchors per output grid (0 to ignore)
|
557 |
+
'fl_gamma': (0, 0.0, 2.0), # focal loss gamma (efficientDet default gamma=1.5)
|
558 |
+
'hsv_h': (1, 0.0, 0.1), # image HSV-Hue augmentation (fraction)
|
559 |
+
'hsv_s': (1, 0.0, 0.9), # image HSV-Saturation augmentation (fraction)
|
560 |
+
'hsv_v': (1, 0.0, 0.9), # image HSV-Value augmentation (fraction)
|
561 |
+
'degrees': (1, 0.0, 45.0), # image rotation (+/- deg)
|
562 |
+
'translate': (1, 0.0, 0.9), # image translation (+/- fraction)
|
563 |
+
'scale': (1, 0.0, 0.9), # image scale (+/- gain)
|
564 |
+
'shear': (1, 0.0, 10.0), # image shear (+/- deg)
|
565 |
+
'perspective': (0, 0.0, 0.001), # image perspective (+/- fraction), range 0-0.001
|
566 |
+
'flipud': (1, 0.0, 1.0), # image flip up-down (probability)
|
567 |
+
'fliplr': (0, 0.0, 1.0), # image flip left-right (probability)
|
568 |
+
'mosaic': (1, 0.0, 1.0), # image mixup (probability)
|
569 |
+
'mixup': (1, 0.0, 1.0)} # image mixup (probability)
|
570 |
+
|
571 |
+
assert opt.local_rank == -1, 'DDP mode not implemented for --evolve'
|
572 |
+
opt.notest, opt.nosave = True, True # only test/save final epoch
|
573 |
+
# ei = [isinstance(x, (int, float)) for x in hyp.values()] # evolvable indices
|
574 |
+
yaml_file = Path(opt.save_dir) / 'hyp_evolved.yaml' # save best result here
|
575 |
+
if opt.bucket:
|
576 |
+
os.system('gsutil cp gs://%s/evolve.txt .' % opt.bucket) # download evolve.txt if exists
|
577 |
+
|
578 |
+
for _ in range(300): # generations to evolve
|
579 |
+
if Path('evolve.txt').exists(): # if evolve.txt exists: select best hyps and mutate
|
580 |
+
# Select parent(s)
|
581 |
+
parent = 'single' # parent selection method: 'single' or 'weighted'
|
582 |
+
x = np.loadtxt('evolve.txt', ndmin=2)
|
583 |
+
n = min(5, len(x)) # number of previous results to consider
|
584 |
+
x = x[np.argsort(-fitness(x))][:n] # top n mutations
|
585 |
+
w = fitness(x) - fitness(x).min() # weights
|
586 |
+
if parent == 'single' or len(x) == 1:
|
587 |
+
# x = x[random.randint(0, n - 1)] # random selection
|
588 |
+
x = x[random.choices(range(n), weights=w)[0]] # weighted selection
|
589 |
+
elif parent == 'weighted':
|
590 |
+
x = (x * w.reshape(n, 1)).sum(0) / w.sum() # weighted combination
|
591 |
+
|
592 |
+
# Mutate
|
593 |
+
mp, s = 0.8, 0.2 # mutation probability, sigma
|
594 |
+
npr = np.random
|
595 |
+
npr.seed(int(time.time()))
|
596 |
+
g = np.array([x[0] for x in meta.values()]) # gains 0-1
|
597 |
+
ng = len(meta)
|
598 |
+
v = np.ones(ng)
|
599 |
+
while all(v == 1): # mutate until a change occurs (prevent duplicates)
|
600 |
+
v = (g * (npr.random(ng) < mp) * npr.randn(ng) * npr.random() * s + 1).clip(0.3, 3.0)
|
601 |
+
for i, k in enumerate(hyp.keys()): # plt.hist(v.ravel(), 300)
|
602 |
+
hyp[k] = float(x[i + 7] * v[i]) # mutate
|
603 |
+
|
604 |
+
# Constrain to limits
|
605 |
+
for k, v in meta.items():
|
606 |
+
hyp[k] = max(hyp[k], v[1]) # lower limit
|
607 |
+
hyp[k] = min(hyp[k], v[2]) # upper limit
|
608 |
+
hyp[k] = round(hyp[k], 5) # significant digits
|
609 |
+
|
610 |
+
# Train mutation
|
611 |
+
results = train(hyp.copy(), opt, device, wandb=wandb)
|
612 |
+
|
613 |
+
# Write mutation results
|
614 |
+
print_mutation(hyp.copy(), results, yaml_file, opt.bucket)
|
615 |
+
|
616 |
+
# Plot results
|
617 |
+
plot_evolution(yaml_file)
|
618 |
+
print(f'Hyperparameter evolution complete. Best results saved as: {yaml_file}\n'
|
619 |
+
f'Command to train a new model with these hyperparameters: $ python train.py --hyp {yaml_file}')
|
tune.py
ADDED
@@ -0,0 +1,619 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import logging
|
3 |
+
import math
|
4 |
+
import os
|
5 |
+
import random
|
6 |
+
import time
|
7 |
+
from pathlib import Path
|
8 |
+
from warnings import warn
|
9 |
+
|
10 |
+
import numpy as np
|
11 |
+
import torch.distributed as dist
|
12 |
+
import torch.nn as nn
|
13 |
+
import torch.nn.functional as F
|
14 |
+
import torch.optim as optim
|
15 |
+
import torch.optim.lr_scheduler as lr_scheduler
|
16 |
+
import torch.utils.data
|
17 |
+
import yaml
|
18 |
+
from torch.cuda import amp
|
19 |
+
from torch.nn.parallel import DistributedDataParallel as DDP
|
20 |
+
from torch.utils.tensorboard import SummaryWriter
|
21 |
+
from tqdm import tqdm
|
22 |
+
|
23 |
+
import test # import test.py to get mAP after each epoch
|
24 |
+
#from models.yolo import Model
|
25 |
+
from models.models import *
|
26 |
+
from utils.autoanchor import check_anchors
|
27 |
+
from utils.datasets import create_dataloader9 as create_dataloader
|
28 |
+
from utils.general import labels_to_class_weights, increment_path, labels_to_image_weights, init_seeds, \
|
29 |
+
fitness, fitness_p, fitness_r, fitness_ap50, fitness_ap, fitness_f, strip_optimizer, get_latest_run,\
|
30 |
+
check_dataset, check_file, check_git_status, check_img_size, print_mutation, set_logging
|
31 |
+
from utils.google_utils import attempt_download
|
32 |
+
from utils.loss import compute_loss
|
33 |
+
from utils.plots import plot_images, plot_labels, plot_results, plot_evolution
|
34 |
+
from utils.torch_utils import ModelEMA, select_device, intersect_dicts, torch_distributed_zero_first
|
35 |
+
|
36 |
+
logger = logging.getLogger(__name__)
|
37 |
+
|
38 |
+
try:
|
39 |
+
import wandb
|
40 |
+
except ImportError:
|
41 |
+
wandb = None
|
42 |
+
logger.info("Install Weights & Biases for experiment logging via 'pip install wandb' (recommended)")
|
43 |
+
|
44 |
+
def train(hyp, opt, device, tb_writer=None, wandb=None):
|
45 |
+
logger.info(f'Hyperparameters {hyp}')
|
46 |
+
save_dir, epochs, batch_size, total_batch_size, weights, rank = \
|
47 |
+
Path(opt.save_dir), opt.epochs, opt.batch_size, opt.total_batch_size, opt.weights, opt.global_rank
|
48 |
+
|
49 |
+
# Directories
|
50 |
+
wdir = save_dir / 'weights'
|
51 |
+
wdir.mkdir(parents=True, exist_ok=True) # make dir
|
52 |
+
last = wdir / 'last.pt'
|
53 |
+
best = wdir / 'best.pt'
|
54 |
+
results_file = save_dir / 'results.txt'
|
55 |
+
|
56 |
+
# Save run settings
|
57 |
+
with open(save_dir / 'hyp.yaml', 'w') as f:
|
58 |
+
yaml.dump(hyp, f, sort_keys=False)
|
59 |
+
with open(save_dir / 'opt.yaml', 'w') as f:
|
60 |
+
yaml.dump(vars(opt), f, sort_keys=False)
|
61 |
+
|
62 |
+
# Configure
|
63 |
+
plots = not opt.evolve # create plots
|
64 |
+
cuda = device.type != 'cpu'
|
65 |
+
init_seeds(2 + rank)
|
66 |
+
with open(opt.data) as f:
|
67 |
+
data_dict = yaml.load(f, Loader=yaml.FullLoader) # data dict
|
68 |
+
with torch_distributed_zero_first(rank):
|
69 |
+
check_dataset(data_dict) # check
|
70 |
+
train_path = data_dict['train']
|
71 |
+
test_path = data_dict['val']
|
72 |
+
nc, names = (1, ['item']) if opt.single_cls else (int(data_dict['nc']), data_dict['names']) # number classes, names
|
73 |
+
assert len(names) == nc, '%g names found for nc=%g dataset in %s' % (len(names), nc, opt.data) # check
|
74 |
+
|
75 |
+
# Model
|
76 |
+
pretrained = weights.endswith('.pt')
|
77 |
+
if pretrained:
|
78 |
+
with torch_distributed_zero_first(rank):
|
79 |
+
attempt_download(weights) # download if not found locally
|
80 |
+
ckpt = torch.load(weights, map_location=device) # load checkpoint
|
81 |
+
model = Darknet(opt.cfg).to(device) # create
|
82 |
+
state_dict = {k: v for k, v in ckpt['model'].items() if model.state_dict()[k].numel() == v.numel()}
|
83 |
+
model.load_state_dict(state_dict, strict=False)
|
84 |
+
print('Transferred %g/%g items from %s' % (len(state_dict), len(model.state_dict()), weights)) # report
|
85 |
+
else:
|
86 |
+
model = Darknet(opt.cfg).to(device) # create
|
87 |
+
|
88 |
+
# Optimizer
|
89 |
+
nbs = 64 # nominal batch size
|
90 |
+
accumulate = max(round(nbs / total_batch_size), 1) # accumulate loss before optimizing
|
91 |
+
hyp['weight_decay'] *= total_batch_size * accumulate / nbs # scale weight_decay
|
92 |
+
|
93 |
+
pg0, pg1, pg2 = [], [], [] # optimizer parameter groups
|
94 |
+
for k, v in dict(model.named_parameters()).items():
|
95 |
+
if '.bias' in k:
|
96 |
+
pg2.append(v) # biases
|
97 |
+
elif 'Conv2d.weight' in k:
|
98 |
+
pg1.append(v) # apply weight_decay
|
99 |
+
elif 'm.weight' in k:
|
100 |
+
pg1.append(v) # apply weight_decay
|
101 |
+
elif 'w.weight' in k:
|
102 |
+
pg1.append(v) # apply weight_decay
|
103 |
+
else:
|
104 |
+
pg0.append(v) # all else
|
105 |
+
|
106 |
+
if opt.adam:
|
107 |
+
optimizer = optim.Adam(pg0, lr=hyp['lr0'], betas=(hyp['momentum'], 0.999)) # adjust beta1 to momentum
|
108 |
+
else:
|
109 |
+
optimizer = optim.SGD(pg0, lr=hyp['lr0'], momentum=hyp['momentum'], nesterov=True)
|
110 |
+
|
111 |
+
optimizer.add_param_group({'params': pg1, 'weight_decay': hyp['weight_decay']}) # add pg1 with weight_decay
|
112 |
+
optimizer.add_param_group({'params': pg2}) # add pg2 (biases)
|
113 |
+
logger.info('Optimizer groups: %g .bias, %g conv.weight, %g other' % (len(pg2), len(pg1), len(pg0)))
|
114 |
+
del pg0, pg1, pg2
|
115 |
+
|
116 |
+
# Scheduler https://arxiv.org/pdf/1812.01187.pdf
|
117 |
+
# https://pytorch.org/docs/stable/_modules/torch/optim/lr_scheduler.html#OneCycleLR
|
118 |
+
lf = lambda x: ((1 + math.cos(x * math.pi / epochs)) / 2) * (1 - hyp['lrf']) + hyp['lrf'] # cosine
|
119 |
+
scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf)
|
120 |
+
# plot_lr_scheduler(optimizer, scheduler, epochs)
|
121 |
+
|
122 |
+
# Logging
|
123 |
+
if wandb and wandb.run is None:
|
124 |
+
opt.hyp = hyp # add hyperparameters
|
125 |
+
wandb_run = wandb.init(config=opt, resume="allow",
|
126 |
+
project='YOLOR' if opt.project == 'runs/train' else Path(opt.project).stem,
|
127 |
+
name=save_dir.stem,
|
128 |
+
id=ckpt.get('wandb_id') if 'ckpt' in locals() else None)
|
129 |
+
|
130 |
+
# Resume
|
131 |
+
start_epoch, best_fitness = 0, 0.0
|
132 |
+
best_fitness_p, best_fitness_r, best_fitness_ap50, best_fitness_ap, best_fitness_f = 0.0, 0.0, 0.0, 0.0, 0.0
|
133 |
+
if pretrained:
|
134 |
+
# Optimizer
|
135 |
+
if ckpt['optimizer'] is not None:
|
136 |
+
optimizer.load_state_dict(ckpt['optimizer'])
|
137 |
+
best_fitness = ckpt['best_fitness']
|
138 |
+
best_fitness_p = ckpt['best_fitness_p']
|
139 |
+
best_fitness_r = ckpt['best_fitness_r']
|
140 |
+
best_fitness_ap50 = ckpt['best_fitness_ap50']
|
141 |
+
best_fitness_ap = ckpt['best_fitness_ap']
|
142 |
+
best_fitness_f = ckpt['best_fitness_f']
|
143 |
+
|
144 |
+
# Results
|
145 |
+
if ckpt.get('training_results') is not None:
|
146 |
+
with open(results_file, 'w') as file:
|
147 |
+
file.write(ckpt['training_results']) # write results.txt
|
148 |
+
|
149 |
+
# Epochs
|
150 |
+
start_epoch = ckpt['epoch'] + 1
|
151 |
+
if opt.resume:
|
152 |
+
assert start_epoch > 0, '%s training to %g epochs is finished, nothing to resume.' % (weights, epochs)
|
153 |
+
if epochs < start_epoch:
|
154 |
+
logger.info('%s has been trained for %g epochs. Fine-tuning for %g additional epochs.' %
|
155 |
+
(weights, ckpt['epoch'], epochs))
|
156 |
+
epochs += ckpt['epoch'] # finetune additional epochs
|
157 |
+
|
158 |
+
del ckpt, state_dict
|
159 |
+
|
160 |
+
# Image sizes
|
161 |
+
gs = 64 #int(max(model.stride)) # grid size (max stride)
|
162 |
+
imgsz, imgsz_test = [check_img_size(x, gs) for x in opt.img_size] # verify imgsz are gs-multiples
|
163 |
+
|
164 |
+
# DP mode
|
165 |
+
if cuda and rank == -1 and torch.cuda.device_count() > 1:
|
166 |
+
model = torch.nn.DataParallel(model)
|
167 |
+
|
168 |
+
# SyncBatchNorm
|
169 |
+
if opt.sync_bn and cuda and rank != -1:
|
170 |
+
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device)
|
171 |
+
logger.info('Using SyncBatchNorm()')
|
172 |
+
|
173 |
+
# EMA
|
174 |
+
ema = ModelEMA(model) if rank in [-1, 0] else None
|
175 |
+
|
176 |
+
# DDP mode
|
177 |
+
if cuda and rank != -1:
|
178 |
+
model = DDP(model, device_ids=[opt.local_rank], output_device=opt.local_rank)
|
179 |
+
|
180 |
+
# Trainloader
|
181 |
+
dataloader, dataset = create_dataloader(train_path, imgsz, batch_size, gs, opt,
|
182 |
+
hyp=hyp, augment=True, cache=opt.cache_images, rect=opt.rect,
|
183 |
+
rank=rank, world_size=opt.world_size, workers=opt.workers)
|
184 |
+
mlc = np.concatenate(dataset.labels, 0)[:, 0].max() # max label class
|
185 |
+
nb = len(dataloader) # number of batches
|
186 |
+
assert mlc < nc, 'Label class %g exceeds nc=%g in %s. Possible class labels are 0-%g' % (mlc, nc, opt.data, nc - 1)
|
187 |
+
|
188 |
+
# Process 0
|
189 |
+
if rank in [-1, 0]:
|
190 |
+
ema.updates = start_epoch * nb // accumulate # set EMA updates
|
191 |
+
testloader = create_dataloader(test_path, imgsz_test, batch_size*2, gs, opt,
|
192 |
+
hyp=hyp, cache=opt.cache_images and not opt.notest, rect=True,
|
193 |
+
rank=-1, world_size=opt.world_size, workers=opt.workers)[0] # testloader
|
194 |
+
|
195 |
+
if not opt.resume:
|
196 |
+
labels = np.concatenate(dataset.labels, 0)
|
197 |
+
c = torch.tensor(labels[:, 0]) # classes
|
198 |
+
# cf = torch.bincount(c.long(), minlength=nc) + 1. # frequency
|
199 |
+
# model._initialize_biases(cf.to(device))
|
200 |
+
if plots:
|
201 |
+
plot_labels(labels, save_dir=save_dir)
|
202 |
+
if tb_writer:
|
203 |
+
tb_writer.add_histogram('classes', c, 0)
|
204 |
+
if wandb:
|
205 |
+
wandb.log({"Labels": [wandb.Image(str(x), caption=x.name) for x in save_dir.glob('*labels*.png')]})
|
206 |
+
|
207 |
+
# Anchors
|
208 |
+
# if not opt.noautoanchor:
|
209 |
+
# check_anchors(dataset, model=model, thr=hyp['anchor_t'], imgsz=imgsz)
|
210 |
+
|
211 |
+
# Model parameters
|
212 |
+
hyp['cls'] *= nc / 80. # scale coco-tuned hyp['cls'] to current dataset
|
213 |
+
model.nc = nc # attach number of classes to model
|
214 |
+
model.hyp = hyp # attach hyperparameters to model
|
215 |
+
model.gr = 1.0 # iou loss ratio (obj_loss = 1.0 or iou)
|
216 |
+
model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) # attach class weights
|
217 |
+
model.names = names
|
218 |
+
|
219 |
+
# Start training
|
220 |
+
t0 = time.time()
|
221 |
+
nw = max(round(hyp['warmup_epochs'] * nb), 1000) # number of warmup iterations, max(3 epochs, 1k iterations)
|
222 |
+
# nw = min(nw, (epochs - start_epoch) / 2 * nb) # limit warmup to < 1/2 of training
|
223 |
+
maps = np.zeros(nc) # mAP per class
|
224 |
+
results = (0, 0, 0, 0, 0, 0, 0) # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
|
225 |
+
scheduler.last_epoch = start_epoch - 1 # do not move
|
226 |
+
scaler = amp.GradScaler(enabled=cuda)
|
227 |
+
logger.info('Image sizes %g train, %g test\n'
|
228 |
+
'Using %g dataloader workers\nLogging results to %s\n'
|
229 |
+
'Starting training for %g epochs...' % (imgsz, imgsz_test, dataloader.num_workers, save_dir, epochs))
|
230 |
+
|
231 |
+
torch.save(model, wdir / 'init.pt')
|
232 |
+
|
233 |
+
for epoch in range(start_epoch, epochs): # epoch ------------------------------------------------------------------
|
234 |
+
model.train()
|
235 |
+
|
236 |
+
# Update image weights (optional)
|
237 |
+
if opt.image_weights:
|
238 |
+
# Generate indices
|
239 |
+
if rank in [-1, 0]:
|
240 |
+
cw = model.class_weights.cpu().numpy() * (1 - maps) ** 2 # class weights
|
241 |
+
iw = labels_to_image_weights(dataset.labels, nc=nc, class_weights=cw) # image weights
|
242 |
+
dataset.indices = random.choices(range(dataset.n), weights=iw, k=dataset.n) # rand weighted idx
|
243 |
+
# Broadcast if DDP
|
244 |
+
if rank != -1:
|
245 |
+
indices = (torch.tensor(dataset.indices) if rank == 0 else torch.zeros(dataset.n)).int()
|
246 |
+
dist.broadcast(indices, 0)
|
247 |
+
if rank != 0:
|
248 |
+
dataset.indices = indices.cpu().numpy()
|
249 |
+
|
250 |
+
# Update mosaic border
|
251 |
+
# b = int(random.uniform(0.25 * imgsz, 0.75 * imgsz + gs) // gs * gs)
|
252 |
+
# dataset.mosaic_border = [b - imgsz, -b] # height, width borders
|
253 |
+
|
254 |
+
mloss = torch.zeros(4, device=device) # mean losses
|
255 |
+
if rank != -1:
|
256 |
+
dataloader.sampler.set_epoch(epoch)
|
257 |
+
pbar = enumerate(dataloader)
|
258 |
+
logger.info(('\n' + '%10s' * 8) % ('Epoch', 'gpu_mem', 'box', 'obj', 'cls', 'total', 'targets', 'img_size'))
|
259 |
+
if rank in [-1, 0]:
|
260 |
+
pbar = tqdm(pbar, total=nb) # progress bar
|
261 |
+
optimizer.zero_grad()
|
262 |
+
for i, (imgs, targets, paths, _) in pbar: # batch -------------------------------------------------------------
|
263 |
+
ni = i + nb * epoch # number integrated batches (since train start)
|
264 |
+
imgs = imgs.to(device, non_blocking=True).float() / 255.0 # uint8 to float32, 0-255 to 0.0-1.0
|
265 |
+
|
266 |
+
# Warmup
|
267 |
+
if ni <= nw:
|
268 |
+
xi = [0, nw] # x interp
|
269 |
+
# model.gr = np.interp(ni, xi, [0.0, 1.0]) # iou loss ratio (obj_loss = 1.0 or iou)
|
270 |
+
accumulate = max(1, np.interp(ni, xi, [1, nbs / total_batch_size]).round())
|
271 |
+
for j, x in enumerate(optimizer.param_groups):
|
272 |
+
# bias lr falls from 0.1 to lr0, all other lrs rise from 0.0 to lr0
|
273 |
+
x['lr'] = np.interp(ni, xi, [hyp['warmup_bias_lr'] if j == 2 else 0.0, x['initial_lr'] * lf(epoch)])
|
274 |
+
if 'momentum' in x:
|
275 |
+
x['momentum'] = np.interp(ni, xi, [hyp['warmup_momentum'], hyp['momentum']])
|
276 |
+
|
277 |
+
# Multi-scale
|
278 |
+
if opt.multi_scale:
|
279 |
+
sz = random.randrange(imgsz * 0.5, imgsz * 1.5 + gs) // gs * gs # size
|
280 |
+
sf = sz / max(imgs.shape[2:]) # scale factor
|
281 |
+
if sf != 1:
|
282 |
+
ns = [math.ceil(x * sf / gs) * gs for x in imgs.shape[2:]] # new shape (stretched to gs-multiple)
|
283 |
+
imgs = F.interpolate(imgs, size=ns, mode='bilinear', align_corners=False)
|
284 |
+
|
285 |
+
# Forward
|
286 |
+
with amp.autocast(enabled=cuda):
|
287 |
+
pred = model(imgs) # forward
|
288 |
+
loss, loss_items = compute_loss(pred, targets.to(device), model) # loss scaled by batch_size
|
289 |
+
if rank != -1:
|
290 |
+
loss *= opt.world_size # gradient averaged between devices in DDP mode
|
291 |
+
|
292 |
+
# Backward
|
293 |
+
scaler.scale(loss).backward()
|
294 |
+
|
295 |
+
# Optimize
|
296 |
+
if ni % accumulate == 0:
|
297 |
+
scaler.step(optimizer) # optimizer.step
|
298 |
+
scaler.update()
|
299 |
+
optimizer.zero_grad()
|
300 |
+
if ema:
|
301 |
+
ema.update(model)
|
302 |
+
|
303 |
+
# Print
|
304 |
+
if rank in [-1, 0]:
|
305 |
+
mloss = (mloss * i + loss_items) / (i + 1) # update mean losses
|
306 |
+
mem = '%.3gG' % (torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0) # (GB)
|
307 |
+
s = ('%10s' * 2 + '%10.4g' * 6) % (
|
308 |
+
'%g/%g' % (epoch, epochs - 1), mem, *mloss, targets.shape[0], imgs.shape[-1])
|
309 |
+
pbar.set_description(s)
|
310 |
+
|
311 |
+
# Plot
|
312 |
+
if plots and ni < 3:
|
313 |
+
f = save_dir / f'train_batch{ni}.jpg' # filename
|
314 |
+
plot_images(images=imgs, targets=targets, paths=paths, fname=f)
|
315 |
+
# if tb_writer:
|
316 |
+
# tb_writer.add_image(f, result, dataformats='HWC', global_step=epoch)
|
317 |
+
# tb_writer.add_graph(model, imgs) # add model to tensorboard
|
318 |
+
elif plots and ni == 3 and wandb:
|
319 |
+
wandb.log({"Mosaics": [wandb.Image(str(x), caption=x.name) for x in save_dir.glob('train*.jpg')]})
|
320 |
+
|
321 |
+
# end batch ------------------------------------------------------------------------------------------------
|
322 |
+
# end epoch ----------------------------------------------------------------------------------------------------
|
323 |
+
|
324 |
+
# Scheduler
|
325 |
+
lr = [x['lr'] for x in optimizer.param_groups] # for tensorboard
|
326 |
+
scheduler.step()
|
327 |
+
|
328 |
+
# DDP process 0 or single-GPU
|
329 |
+
if rank in [-1, 0]:
|
330 |
+
# mAP
|
331 |
+
if ema:
|
332 |
+
ema.update_attr(model)
|
333 |
+
final_epoch = epoch + 1 == epochs
|
334 |
+
if not opt.notest or final_epoch: # Calculate mAP
|
335 |
+
if epoch >= 3:
|
336 |
+
results, maps, times = test.test(opt.data,
|
337 |
+
batch_size=batch_size*2,
|
338 |
+
imgsz=imgsz_test,
|
339 |
+
model=ema.ema.module if hasattr(ema.ema, 'module') else ema.ema,
|
340 |
+
single_cls=opt.single_cls,
|
341 |
+
dataloader=testloader,
|
342 |
+
save_dir=save_dir,
|
343 |
+
plots=plots and final_epoch,
|
344 |
+
log_imgs=opt.log_imgs if wandb else 0)
|
345 |
+
|
346 |
+
# Write
|
347 |
+
with open(results_file, 'a') as f:
|
348 |
+
f.write(s + '%10.4g' * 7 % results + '\n') # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
|
349 |
+
if len(opt.name) and opt.bucket:
|
350 |
+
os.system('gsutil cp %s gs://%s/results/results%s.txt' % (results_file, opt.bucket, opt.name))
|
351 |
+
|
352 |
+
# Log
|
353 |
+
tags = ['train/box_loss', 'train/obj_loss', 'train/cls_loss', # train loss
|
354 |
+
'metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/mAP_0.5:0.95',
|
355 |
+
'val/box_loss', 'val/obj_loss', 'val/cls_loss', # val loss
|
356 |
+
'x/lr0', 'x/lr1', 'x/lr2'] # params
|
357 |
+
for x, tag in zip(list(mloss[:-1]) + list(results) + lr, tags):
|
358 |
+
if tb_writer:
|
359 |
+
tb_writer.add_scalar(tag, x, epoch) # tensorboard
|
360 |
+
if wandb:
|
361 |
+
wandb.log({tag: x}) # W&B
|
362 |
+
|
363 |
+
# Update best mAP
|
364 |
+
fi = fitness(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
365 |
+
fi_p = fitness_p(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
366 |
+
fi_r = fitness_r(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
367 |
+
fi_ap50 = fitness_ap50(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
368 |
+
fi_ap = fitness_ap(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
369 |
+
if (fi_p > 0.0) or (fi_r > 0.0):
|
370 |
+
fi_f = fitness_f(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
|
371 |
+
else:
|
372 |
+
fi_f = 0.0
|
373 |
+
if fi > best_fitness:
|
374 |
+
best_fitness = fi
|
375 |
+
if fi_p > best_fitness_p:
|
376 |
+
best_fitness_p = fi_p
|
377 |
+
if fi_r > best_fitness_r:
|
378 |
+
best_fitness_r = fi_r
|
379 |
+
if fi_ap50 > best_fitness_ap50:
|
380 |
+
best_fitness_ap50 = fi_ap50
|
381 |
+
if fi_ap > best_fitness_ap:
|
382 |
+
best_fitness_ap = fi_ap
|
383 |
+
if fi_f > best_fitness_f:
|
384 |
+
best_fitness_f = fi_f
|
385 |
+
|
386 |
+
# Save model
|
387 |
+
save = (not opt.nosave) or (final_epoch and not opt.evolve)
|
388 |
+
if save:
|
389 |
+
with open(results_file, 'r') as f: # create checkpoint
|
390 |
+
ckpt = {'epoch': epoch,
|
391 |
+
'best_fitness': best_fitness,
|
392 |
+
'best_fitness_p': best_fitness_p,
|
393 |
+
'best_fitness_r': best_fitness_r,
|
394 |
+
'best_fitness_ap50': best_fitness_ap50,
|
395 |
+
'best_fitness_ap': best_fitness_ap,
|
396 |
+
'best_fitness_f': best_fitness_f,
|
397 |
+
'training_results': f.read(),
|
398 |
+
'model': ema.ema.module.state_dict() if hasattr(ema, 'module') else ema.ema.state_dict(),
|
399 |
+
'optimizer': None if final_epoch else optimizer.state_dict(),
|
400 |
+
'wandb_id': wandb_run.id if wandb else None}
|
401 |
+
|
402 |
+
# Save last, best and delete
|
403 |
+
torch.save(ckpt, last)
|
404 |
+
if best_fitness == fi:
|
405 |
+
torch.save(ckpt, best)
|
406 |
+
if (best_fitness == fi) and (epoch >= 200):
|
407 |
+
torch.save(ckpt, wdir / 'best_{:03d}.pt'.format(epoch))
|
408 |
+
if best_fitness == fi:
|
409 |
+
torch.save(ckpt, wdir / 'best_overall.pt')
|
410 |
+
if best_fitness_p == fi_p:
|
411 |
+
torch.save(ckpt, wdir / 'best_p.pt')
|
412 |
+
if best_fitness_r == fi_r:
|
413 |
+
torch.save(ckpt, wdir / 'best_r.pt')
|
414 |
+
if best_fitness_ap50 == fi_ap50:
|
415 |
+
torch.save(ckpt, wdir / 'best_ap50.pt')
|
416 |
+
if best_fitness_ap == fi_ap:
|
417 |
+
torch.save(ckpt, wdir / 'best_ap.pt')
|
418 |
+
if best_fitness_f == fi_f:
|
419 |
+
torch.save(ckpt, wdir / 'best_f.pt')
|
420 |
+
if epoch == 0:
|
421 |
+
torch.save(ckpt, wdir / 'epoch_{:03d}.pt'.format(epoch))
|
422 |
+
if ((epoch+1) % 25) == 0:
|
423 |
+
torch.save(ckpt, wdir / 'epoch_{:03d}.pt'.format(epoch))
|
424 |
+
if epoch >= (epochs-5):
|
425 |
+
torch.save(ckpt, wdir / 'last_{:03d}.pt'.format(epoch))
|
426 |
+
elif epoch >= 420:
|
427 |
+
torch.save(ckpt, wdir / 'last_{:03d}.pt'.format(epoch))
|
428 |
+
del ckpt
|
429 |
+
# end epoch ----------------------------------------------------------------------------------------------------
|
430 |
+
# end training
|
431 |
+
|
432 |
+
if rank in [-1, 0]:
|
433 |
+
# Strip optimizers
|
434 |
+
n = opt.name if opt.name.isnumeric() else ''
|
435 |
+
fresults, flast, fbest = save_dir / f'results{n}.txt', wdir / f'last{n}.pt', wdir / f'best{n}.pt'
|
436 |
+
for f1, f2 in zip([wdir / 'last.pt', wdir / 'best.pt', results_file], [flast, fbest, fresults]):
|
437 |
+
if f1.exists():
|
438 |
+
os.rename(f1, f2) # rename
|
439 |
+
if str(f2).endswith('.pt'): # is *.pt
|
440 |
+
strip_optimizer(f2) # strip optimizer
|
441 |
+
os.system('gsutil cp %s gs://%s/weights' % (f2, opt.bucket)) if opt.bucket else None # upload
|
442 |
+
# Finish
|
443 |
+
if plots:
|
444 |
+
plot_results(save_dir=save_dir) # save as results.png
|
445 |
+
if wandb:
|
446 |
+
wandb.log({"Results": [wandb.Image(str(save_dir / x), caption=x) for x in
|
447 |
+
['results.png', 'precision-recall_curve.png']]})
|
448 |
+
logger.info('%g epochs completed in %.3f hours.\n' % (epoch - start_epoch + 1, (time.time() - t0) / 3600))
|
449 |
+
else:
|
450 |
+
dist.destroy_process_group()
|
451 |
+
|
452 |
+
wandb.run.finish() if wandb and wandb.run else None
|
453 |
+
torch.cuda.empty_cache()
|
454 |
+
return results
|
455 |
+
|
456 |
+
|
457 |
+
if __name__ == '__main__':
|
458 |
+
parser = argparse.ArgumentParser()
|
459 |
+
parser.add_argument('--weights', type=str, default='yolor_p6.pt', help='initial weights path')
|
460 |
+
parser.add_argument('--cfg', type=str, default='', help='model.yaml path')
|
461 |
+
parser.add_argument('--data', type=str, default='data/coco.yaml', help='data.yaml path')
|
462 |
+
parser.add_argument('--hyp', type=str, default='data/hyp.scratch.1280.yaml', help='hyperparameters path')
|
463 |
+
parser.add_argument('--epochs', type=int, default=300)
|
464 |
+
parser.add_argument('--batch-size', type=int, default=8, help='total batch size for all GPUs')
|
465 |
+
parser.add_argument('--img-size', nargs='+', type=int, default=[1280, 1280], help='[train, test] image sizes')
|
466 |
+
parser.add_argument('--rect', action='store_true', help='rectangular training')
|
467 |
+
parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')
|
468 |
+
parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')
|
469 |
+
parser.add_argument('--notest', action='store_true', help='only test final epoch')
|
470 |
+
parser.add_argument('--noautoanchor', action='store_true', help='disable autoanchor check')
|
471 |
+
parser.add_argument('--evolve', action='store_true', help='evolve hyperparameters')
|
472 |
+
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
|
473 |
+
parser.add_argument('--cache-images', action='store_true', help='cache images for faster training')
|
474 |
+
parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')
|
475 |
+
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
|
476 |
+
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
|
477 |
+
parser.add_argument('--single-cls', action='store_true', help='train as single-class dataset')
|
478 |
+
parser.add_argument('--adam', action='store_true', help='use torch.optim.Adam() optimizer')
|
479 |
+
parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')
|
480 |
+
parser.add_argument('--local_rank', type=int, default=-1, help='DDP parameter, do not modify')
|
481 |
+
parser.add_argument('--log-imgs', type=int, default=16, help='number of images for W&B logging, max 100')
|
482 |
+
parser.add_argument('--workers', type=int, default=8, help='maximum number of dataloader workers')
|
483 |
+
parser.add_argument('--project', default='runs/train', help='save to project/name')
|
484 |
+
parser.add_argument('--name', default='exp', help='save to project/name')
|
485 |
+
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
|
486 |
+
opt = parser.parse_args()
|
487 |
+
|
488 |
+
# Set DDP variables
|
489 |
+
opt.total_batch_size = opt.batch_size
|
490 |
+
opt.world_size = int(os.environ['WORLD_SIZE']) if 'WORLD_SIZE' in os.environ else 1
|
491 |
+
opt.global_rank = int(os.environ['RANK']) if 'RANK' in os.environ else -1
|
492 |
+
set_logging(opt.global_rank)
|
493 |
+
if opt.global_rank in [-1, 0]:
|
494 |
+
check_git_status()
|
495 |
+
|
496 |
+
# Resume
|
497 |
+
if opt.resume: # resume an interrupted run
|
498 |
+
ckpt = opt.resume if isinstance(opt.resume, str) else get_latest_run() # specified or most recent path
|
499 |
+
assert os.path.isfile(ckpt), 'ERROR: --resume checkpoint does not exist'
|
500 |
+
with open(Path(ckpt).parent.parent / 'opt.yaml') as f:
|
501 |
+
opt = argparse.Namespace(**yaml.load(f, Loader=yaml.FullLoader)) # replace
|
502 |
+
opt.cfg, opt.weights, opt.resume = '', ckpt, True
|
503 |
+
logger.info('Resuming training from %s' % ckpt)
|
504 |
+
else:
|
505 |
+
# opt.hyp = opt.hyp or ('hyp.finetune.yaml' if opt.weights else 'hyp.scratch.yaml')
|
506 |
+
opt.data, opt.cfg, opt.hyp = check_file(opt.data), check_file(opt.cfg), check_file(opt.hyp) # check files
|
507 |
+
assert len(opt.cfg) or len(opt.weights), 'either --cfg or --weights must be specified'
|
508 |
+
opt.img_size.extend([opt.img_size[-1]] * (2 - len(opt.img_size))) # extend to 2 sizes (train, test)
|
509 |
+
opt.name = 'evolve' if opt.evolve else opt.name
|
510 |
+
opt.save_dir = increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok | opt.evolve) # increment run
|
511 |
+
|
512 |
+
# DDP mode
|
513 |
+
device = select_device(opt.device, batch_size=opt.batch_size)
|
514 |
+
if opt.local_rank != -1:
|
515 |
+
assert torch.cuda.device_count() > opt.local_rank
|
516 |
+
torch.cuda.set_device(opt.local_rank)
|
517 |
+
device = torch.device('cuda', opt.local_rank)
|
518 |
+
dist.init_process_group(backend='nccl', init_method='env://') # distributed backend
|
519 |
+
assert opt.batch_size % opt.world_size == 0, '--batch-size must be multiple of CUDA device count'
|
520 |
+
opt.batch_size = opt.total_batch_size // opt.world_size
|
521 |
+
|
522 |
+
# Hyperparameters
|
523 |
+
with open(opt.hyp) as f:
|
524 |
+
hyp = yaml.load(f, Loader=yaml.FullLoader) # load hyps
|
525 |
+
if 'box' not in hyp:
|
526 |
+
warn('Compatibility: %s missing "box" which was renamed from "giou" in %s' %
|
527 |
+
(opt.hyp, 'https://github.com/ultralytics/yolov5/pull/1120'))
|
528 |
+
hyp['box'] = hyp.pop('giou')
|
529 |
+
|
530 |
+
# Train
|
531 |
+
logger.info(opt)
|
532 |
+
if not opt.evolve:
|
533 |
+
tb_writer = None # init loggers
|
534 |
+
if opt.global_rank in [-1, 0]:
|
535 |
+
logger.info(f'Start Tensorboard with "tensorboard --logdir {opt.project}", view at http://localhost:6006/')
|
536 |
+
tb_writer = SummaryWriter(opt.save_dir) # Tensorboard
|
537 |
+
train(hyp, opt, device, tb_writer, wandb)
|
538 |
+
|
539 |
+
# Evolve hyperparameters (optional)
|
540 |
+
else:
|
541 |
+
# Hyperparameter evolution metadata (mutation scale 0-1, lower_limit, upper_limit)
|
542 |
+
meta = {'lr0': (1, 1e-5, 1e-1), # initial learning rate (SGD=1E-2, Adam=1E-3)
|
543 |
+
'lrf': (1, 0.01, 1.0), # final OneCycleLR learning rate (lr0 * lrf)
|
544 |
+
'momentum': (0.3, 0.6, 0.98), # SGD momentum/Adam beta1
|
545 |
+
'weight_decay': (1, 0.0, 0.001), # optimizer weight decay
|
546 |
+
'warmup_epochs': (1, 0.0, 5.0), # warmup epochs (fractions ok)
|
547 |
+
'warmup_momentum': (1, 0.0, 0.95), # warmup initial momentum
|
548 |
+
'warmup_bias_lr': (1, 0.0, 0.2), # warmup initial bias lr
|
549 |
+
'box': (1, 0.02, 0.2), # box loss gain
|
550 |
+
'cls': (1, 0.2, 4.0), # cls loss gain
|
551 |
+
'cls_pw': (1, 0.5, 2.0), # cls BCELoss positive_weight
|
552 |
+
'obj': (1, 0.2, 4.0), # obj loss gain (scale with pixels)
|
553 |
+
'obj_pw': (1, 0.5, 2.0), # obj BCELoss positive_weight
|
554 |
+
'iou_t': (0, 0.1, 0.7), # IoU training threshold
|
555 |
+
'anchor_t': (1, 2.0, 8.0), # anchor-multiple threshold
|
556 |
+
'anchors': (2, 2.0, 10.0), # anchors per output grid (0 to ignore)
|
557 |
+
'fl_gamma': (0, 0.0, 2.0), # focal loss gamma (efficientDet default gamma=1.5)
|
558 |
+
'hsv_h': (1, 0.0, 0.1), # image HSV-Hue augmentation (fraction)
|
559 |
+
'hsv_s': (1, 0.0, 0.9), # image HSV-Saturation augmentation (fraction)
|
560 |
+
'hsv_v': (1, 0.0, 0.9), # image HSV-Value augmentation (fraction)
|
561 |
+
'degrees': (1, 0.0, 45.0), # image rotation (+/- deg)
|
562 |
+
'translate': (1, 0.0, 0.9), # image translation (+/- fraction)
|
563 |
+
'scale': (1, 0.0, 0.9), # image scale (+/- gain)
|
564 |
+
'shear': (1, 0.0, 10.0), # image shear (+/- deg)
|
565 |
+
'perspective': (0, 0.0, 0.001), # image perspective (+/- fraction), range 0-0.001
|
566 |
+
'flipud': (1, 0.0, 1.0), # image flip up-down (probability)
|
567 |
+
'fliplr': (0, 0.0, 1.0), # image flip left-right (probability)
|
568 |
+
'mosaic': (1, 0.0, 1.0), # image mixup (probability)
|
569 |
+
'mixup': (1, 0.0, 1.0)} # image mixup (probability)
|
570 |
+
|
571 |
+
assert opt.local_rank == -1, 'DDP mode not implemented for --evolve'
|
572 |
+
opt.notest, opt.nosave = True, True # only test/save final epoch
|
573 |
+
# ei = [isinstance(x, (int, float)) for x in hyp.values()] # evolvable indices
|
574 |
+
yaml_file = Path(opt.save_dir) / 'hyp_evolved.yaml' # save best result here
|
575 |
+
if opt.bucket:
|
576 |
+
os.system('gsutil cp gs://%s/evolve.txt .' % opt.bucket) # download evolve.txt if exists
|
577 |
+
|
578 |
+
for _ in range(300): # generations to evolve
|
579 |
+
if Path('evolve.txt').exists(): # if evolve.txt exists: select best hyps and mutate
|
580 |
+
# Select parent(s)
|
581 |
+
parent = 'single' # parent selection method: 'single' or 'weighted'
|
582 |
+
x = np.loadtxt('evolve.txt', ndmin=2)
|
583 |
+
n = min(5, len(x)) # number of previous results to consider
|
584 |
+
x = x[np.argsort(-fitness(x))][:n] # top n mutations
|
585 |
+
w = fitness(x) - fitness(x).min() # weights
|
586 |
+
if parent == 'single' or len(x) == 1:
|
587 |
+
# x = x[random.randint(0, n - 1)] # random selection
|
588 |
+
x = x[random.choices(range(n), weights=w)[0]] # weighted selection
|
589 |
+
elif parent == 'weighted':
|
590 |
+
x = (x * w.reshape(n, 1)).sum(0) / w.sum() # weighted combination
|
591 |
+
|
592 |
+
# Mutate
|
593 |
+
mp, s = 0.8, 0.2 # mutation probability, sigma
|
594 |
+
npr = np.random
|
595 |
+
npr.seed(int(time.time()))
|
596 |
+
g = np.array([x[0] for x in meta.values()]) # gains 0-1
|
597 |
+
ng = len(meta)
|
598 |
+
v = np.ones(ng)
|
599 |
+
while all(v == 1): # mutate until a change occurs (prevent duplicates)
|
600 |
+
v = (g * (npr.random(ng) < mp) * npr.randn(ng) * npr.random() * s + 1).clip(0.3, 3.0)
|
601 |
+
for i, k in enumerate(hyp.keys()): # plt.hist(v.ravel(), 300)
|
602 |
+
hyp[k] = float(x[i + 7] * v[i]) # mutate
|
603 |
+
|
604 |
+
# Constrain to limits
|
605 |
+
for k, v in meta.items():
|
606 |
+
hyp[k] = max(hyp[k], v[1]) # lower limit
|
607 |
+
hyp[k] = min(hyp[k], v[2]) # upper limit
|
608 |
+
hyp[k] = round(hyp[k], 5) # significant digits
|
609 |
+
|
610 |
+
# Train mutation
|
611 |
+
results = train(hyp.copy(), opt, device, wandb=wandb)
|
612 |
+
|
613 |
+
# Write mutation results
|
614 |
+
print_mutation(hyp.copy(), results, yaml_file, opt.bucket)
|
615 |
+
|
616 |
+
# Plot results
|
617 |
+
plot_evolution(yaml_file)
|
618 |
+
print(f'Hyperparameter evolution complete. Best results saved as: {yaml_file}\n'
|
619 |
+
f'Command to train a new model with these hyperparameters: $ python train.py --hyp {yaml_file}')
|
utils/__init__.py
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
|
utils/__pycache__/__init__.cpython-37.pyc
ADDED
Binary file (152 Bytes). View file
|
|
utils/__pycache__/__init__.cpython-38.pyc
ADDED
Binary file (156 Bytes). View file
|
|
utils/__pycache__/datasets.cpython-37.pyc
ADDED
Binary file (37.9 kB). View file
|
|
utils/__pycache__/datasets.cpython-38.pyc
ADDED
Binary file (35.8 kB). View file
|
|
utils/__pycache__/general.cpython-37.pyc
ADDED
Binary file (14 kB). View file
|
|
utils/__pycache__/google_utils.cpython-37.pyc
ADDED
Binary file (2.94 kB). View file
|
|
utils/__pycache__/google_utils.cpython-38.pyc
ADDED
Binary file (2.96 kB). View file
|
|
utils/__pycache__/layers.cpython-37.pyc
ADDED
Binary file (24.3 kB). View file
|
|
utils/__pycache__/metrics.cpython-37.pyc
ADDED
Binary file (4.04 kB). View file
|
|
utils/__pycache__/parse_config.cpython-37.pyc
ADDED
Binary file (2.73 kB). View file
|
|
utils/__pycache__/plots.cpython-37.pyc
ADDED
Binary file (13.9 kB). View file
|
|
utils/__pycache__/torch_utils.cpython-37.pyc
ADDED
Binary file (9.17 kB). View file
|
|