Keiser41 commited on
Commit
22d8ab7
1 Parent(s): 04d0629

Upload 98 files

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +28 -0
  2. .gitignore +24 -0
  3. LICENSE +201 -0
  4. README.md +262 -0
  5. __pycache__/distributed.cpython-310.pyc +0 -0
  6. __pycache__/models.cpython-310.pyc +0 -0
  7. __pycache__/models.cpython-38.pyc +0 -0
  8. __pycache__/utils.cpython-310.pyc +0 -0
  9. __pycache__/utils.cpython-38.pyc +0 -0
  10. __pycache__/vgg_model.cpython-310.pyc +0 -0
  11. __pycache__/vgg_model.cpython-38.pyc +0 -0
  12. assets/PFFB.png +0 -0
  13. assets/Pipeline.png +3 -0
  14. assets/network.png +3 -0
  15. data/__pycache__/data_loader.cpython-310.pyc +0 -0
  16. data/__pycache__/tps_transformation.cpython-310.pyc +0 -0
  17. data/data_loader.py +97 -0
  18. data/data_loader_sketch.py +120 -0
  19. data/prepare_data.py +84 -0
  20. data/prepare_data_sketch.py +84 -0
  21. data/thinplate/__init__.py +9 -0
  22. data/thinplate/__pycache__/__init__.cpython-310.pyc +0 -0
  23. data/thinplate/__pycache__/numpy.cpython-310.pyc +0 -0
  24. data/thinplate/__pycache__/pytorch.cpython-310.pyc +0 -0
  25. data/thinplate/numpy.py +115 -0
  26. data/thinplate/pytorch.py +126 -0
  27. data/thinplate/tests/__init__.py +0 -0
  28. data/thinplate/tests/test_tps_numpy.py +85 -0
  29. data/thinplate/tests/test_tps_pytorch.py +43 -0
  30. data/tps_transformation.py +44 -0
  31. discriminator.py +31 -0
  32. distributed.py +126 -0
  33. experiments/Color2Manga_gray/074000_gray.pt +3 -0
  34. experiments/Color2Manga_sketch/116000_sketch.pt +3 -0
  35. experiments/Discriminator/074000_d.pt +3 -0
  36. experiments/Discriminator/116000_d.pt +3 -0
  37. experiments/VGG19/vgg19-dcbb9e9d.pth +3 -0
  38. extractor/Open-Sans-Bold.ttf +0 -0
  39. extractor/__pycache__/manga_panel_extractor.cpython-310.pyc +0 -0
  40. extractor/__pycache__/manga_panel_extractor.cpython-38.pyc +0 -0
  41. extractor/manga_panel_extractor.py +174 -0
  42. inference.py +229 -0
  43. models.py +223 -0
  44. real_manga/class1/Color 1659315.jpg +3 -0
  45. real_manga/class1/Color 3223141571376159.jpg +3 -0
  46. real_manga/class1/Color 3486521.jpg +3 -0
  47. real_manga/class1/Color 5102676.jpg +3 -0
  48. real_manga/class1/Color 5570824.jpg +3 -0
  49. real_manga/class1/Color 5674950.jpg +3 -0
  50. real_manga/class1/Color 5828407151952509.jpg +3 -0
.gitattributes CHANGED
@@ -33,3 +33,31 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ assets/network.png filter=lfs diff=lfs merge=lfs -text
37
+ assets/Pipeline.png filter=lfs diff=lfs merge=lfs -text
38
+ real_manga/class1/Color[[:space:]]1659315.jpg filter=lfs diff=lfs merge=lfs -text
39
+ real_manga/class1/Color[[:space:]]3223141571376159.jpg filter=lfs diff=lfs merge=lfs -text
40
+ real_manga/class1/Color[[:space:]]3486521.jpg filter=lfs diff=lfs merge=lfs -text
41
+ real_manga/class1/Color[[:space:]]5102676.jpg filter=lfs diff=lfs merge=lfs -text
42
+ real_manga/class1/Color[[:space:]]5570824.jpg filter=lfs diff=lfs merge=lfs -text
43
+ real_manga/class1/Color[[:space:]]5674950.jpg filter=lfs diff=lfs merge=lfs -text
44
+ real_manga/class1/Color[[:space:]]5828407151952509.jpg filter=lfs diff=lfs merge=lfs -text
45
+ real_manga/class1/Color[[:space:]]5851155317235124.jpg filter=lfs diff=lfs merge=lfs -text
46
+ real_manga/class1/Color[[:space:]]6429789966786911.jpg filter=lfs diff=lfs merge=lfs -text
47
+ real_manga/class1/Color[[:space:]]6813581942189493.jpg filter=lfs diff=lfs merge=lfs -text
48
+ real_manga/class1/Color[[:space:]]8096755.jpg filter=lfs diff=lfs merge=lfs -text
49
+ test_datasets/gray_test/001_in.png filter=lfs diff=lfs merge=lfs -text
50
+ test_datasets/gray_test/002_in_ref_b.jpeg filter=lfs diff=lfs merge=lfs -text
51
+ test_datasets/gray_test/004_in.png filter=lfs diff=lfs merge=lfs -text
52
+ test_datasets/gray_test/005_in.png filter=lfs diff=lfs merge=lfs -text
53
+ test_datasets/gray_test/006_in.png filter=lfs diff=lfs merge=lfs -text
54
+ test_datasets/gray_test/006_ref.png filter=lfs diff=lfs merge=lfs -text
55
+ test_datasets/gray_test/out/001_in_color_a.png filter=lfs diff=lfs merge=lfs -text
56
+ test_datasets/gray_test/out/001_in_color_b.png filter=lfs diff=lfs merge=lfs -text
57
+ test_datasets/gray_test/out/002_in_color_a.png filter=lfs diff=lfs merge=lfs -text
58
+ test_datasets/gray_test/out/002_in_color_b.png filter=lfs diff=lfs merge=lfs -text
59
+ test_datasets/gray_test/out/003_in_color_a.png filter=lfs diff=lfs merge=lfs -text
60
+ test_datasets/gray_test/out/003_in_color_b.png filter=lfs diff=lfs merge=lfs -text
61
+ test_datasets/gray_test/out/004_in_color.png filter=lfs diff=lfs merge=lfs -text
62
+ test_datasets/gray_test/out/005_in_color.png filter=lfs diff=lfs merge=lfs -text
63
+ test_datasets/gray_test/out/006_in_color.png filter=lfs diff=lfs merge=lfs -text
.gitignore ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ### Example user template template
2
+ ### Example user template
3
+
4
+ # IntelliJ project files
5
+ .idea
6
+ *.iml
7
+ out
8
+ gen
9
+
10
+ # Debug file
11
+ datacheck.py
12
+ test_gray2color.py
13
+ val.py
14
+
15
+ experiments/
16
+ misc/
17
+ results/
18
+ test_datasets/*
19
+ !/test_datasets/gray_test
20
+ !/test_datasets/gray_test/out
21
+ !/test_datasets/sketch_test
22
+ !/test_datasets/sketch_test/out
23
+ train_datasets/
24
+ training_logs/
LICENSE ADDED
@@ -0,0 +1,201 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Apache License
2
+ Version 2.0, January 2004
3
+ http://www.apache.org/licenses/
4
+
5
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
6
+
7
+ 1. Definitions.
8
+
9
+ "License" shall mean the terms and conditions for use, reproduction,
10
+ and distribution as defined by Sections 1 through 9 of this document.
11
+
12
+ "Licensor" shall mean the copyright owner or entity authorized by
13
+ the copyright owner that is granting the License.
14
+
15
+ "Legal Entity" shall mean the union of the acting entity and all
16
+ other entities that control, are controlled by, or are under common
17
+ control with that entity. For the purposes of this definition,
18
+ "control" means (i) the power, direct or indirect, to cause the
19
+ direction or management of such entity, whether by contract or
20
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
21
+ outstanding shares, or (iii) beneficial ownership of such entity.
22
+
23
+ "You" (or "Your") shall mean an individual or Legal Entity
24
+ exercising permissions granted by this License.
25
+
26
+ "Source" form shall mean the preferred form for making modifications,
27
+ including but not limited to software source code, documentation
28
+ source, and configuration files.
29
+
30
+ "Object" form shall mean any form resulting from mechanical
31
+ transformation or translation of a Source form, including but
32
+ not limited to compiled object code, generated documentation,
33
+ and conversions to other media types.
34
+
35
+ "Work" shall mean the work of authorship, whether in Source or
36
+ Object form, made available under the License, as indicated by a
37
+ copyright notice that is included in or attached to the work
38
+ (an example is provided in the Appendix below).
39
+
40
+ "Derivative Works" shall mean any work, whether in Source or Object
41
+ form, that is based on (or derived from) the Work and for which the
42
+ editorial revisions, annotations, elaborations, or other modifications
43
+ represent, as a whole, an original work of authorship. For the purposes
44
+ of this License, Derivative Works shall not include works that remain
45
+ separable from, or merely link (or bind by name) to the interfaces of,
46
+ the Work and Derivative Works thereof.
47
+
48
+ "Contribution" shall mean any work of authorship, including
49
+ the original version of the Work and any modifications or additions
50
+ to that Work or Derivative Works thereof, that is intentionally
51
+ submitted to Licensor for inclusion in the Work by the copyright owner
52
+ or by an individual or Legal Entity authorized to submit on behalf of
53
+ the copyright owner. For the purposes of this definition, "submitted"
54
+ means any form of electronic, verbal, or written communication sent
55
+ to the Licensor or its representatives, including but not limited to
56
+ communication on electronic mailing lists, source code control systems,
57
+ and issue tracking systems that are managed by, or on behalf of, the
58
+ Licensor for the purpose of discussing and improving the Work, but
59
+ excluding communication that is conspicuously marked or otherwise
60
+ designated in writing by the copyright owner as "Not a Contribution."
61
+
62
+ "Contributor" shall mean Licensor and any individual or Legal Entity
63
+ on behalf of whom a Contribution has been received by Licensor and
64
+ subsequently incorporated within the Work.
65
+
66
+ 2. Grant of Copyright License. Subject to the terms and conditions of
67
+ this License, each Contributor hereby grants to You a perpetual,
68
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
69
+ copyright license to reproduce, prepare Derivative Works of,
70
+ publicly display, publicly perform, sublicense, and distribute the
71
+ Work and such Derivative Works in Source or Object form.
72
+
73
+ 3. Grant of Patent License. Subject to the terms and conditions of
74
+ this License, each Contributor hereby grants to You a perpetual,
75
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
76
+ (except as stated in this section) patent license to make, have made,
77
+ use, offer to sell, sell, import, and otherwise transfer the Work,
78
+ where such license applies only to those patent claims licensable
79
+ by such Contributor that are necessarily infringed by their
80
+ Contribution(s) alone or by combination of their Contribution(s)
81
+ with the Work to which such Contribution(s) was submitted. If You
82
+ institute patent litigation against any entity (including a
83
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
84
+ or a Contribution incorporated within the Work constitutes direct
85
+ or contributory patent infringement, then any patent licenses
86
+ granted to You under this License for that Work shall terminate
87
+ as of the date such litigation is filed.
88
+
89
+ 4. Redistribution. You may reproduce and distribute copies of the
90
+ Work or Derivative Works thereof in any medium, with or without
91
+ modifications, and in Source or Object form, provided that You
92
+ meet the following conditions:
93
+
94
+ (a) You must give any other recipients of the Work or
95
+ Derivative Works a copy of this License; and
96
+
97
+ (b) You must cause any modified files to carry prominent notices
98
+ stating that You changed the files; and
99
+
100
+ (c) You must retain, in the Source form of any Derivative Works
101
+ that You distribute, all copyright, patent, trademark, and
102
+ attribution notices from the Source form of the Work,
103
+ excluding those notices that do not pertain to any part of
104
+ the Derivative Works; and
105
+
106
+ (d) If the Work includes a "NOTICE" text file as part of its
107
+ distribution, then any Derivative Works that You distribute must
108
+ include a readable copy of the attribution notices contained
109
+ within such NOTICE file, excluding those notices that do not
110
+ pertain to any part of the Derivative Works, in at least one
111
+ of the following places: within a NOTICE text file distributed
112
+ as part of the Derivative Works; within the Source form or
113
+ documentation, if provided along with the Derivative Works; or,
114
+ within a display generated by the Derivative Works, if and
115
+ wherever such third-party notices normally appear. The contents
116
+ of the NOTICE file are for informational purposes only and
117
+ do not modify the License. You may add Your own attribution
118
+ notices within Derivative Works that You distribute, alongside
119
+ or as an addendum to the NOTICE text from the Work, provided
120
+ that such additional attribution notices cannot be construed
121
+ as modifying the License.
122
+
123
+ You may add Your own copyright statement to Your modifications and
124
+ may provide additional or different license terms and conditions
125
+ for use, reproduction, or distribution of Your modifications, or
126
+ for any such Derivative Works as a whole, provided Your use,
127
+ reproduction, and distribution of the Work otherwise complies with
128
+ the conditions stated in this License.
129
+
130
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
131
+ any Contribution intentionally submitted for inclusion in the Work
132
+ by You to the Licensor shall be under the terms and conditions of
133
+ this License, without any additional terms or conditions.
134
+ Notwithstanding the above, nothing herein shall supersede or modify
135
+ the terms of any separate license agreement you may have executed
136
+ with Licensor regarding such Contributions.
137
+
138
+ 6. Trademarks. This License does not grant permission to use the trade
139
+ names, trademarks, service marks, or product names of the Licensor,
140
+ except as required for reasonable and customary use in describing the
141
+ origin of the Work and reproducing the content of the NOTICE file.
142
+
143
+ 7. Disclaimer of Warranty. Unless required by applicable law or
144
+ agreed to in writing, Licensor provides the Work (and each
145
+ Contributor provides its Contributions) on an "AS IS" BASIS,
146
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
147
+ implied, including, without limitation, any warranties or conditions
148
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
149
+ PARTICULAR PURPOSE. You are solely responsible for determining the
150
+ appropriateness of using or redistributing the Work and assume any
151
+ risks associated with Your exercise of permissions under this License.
152
+
153
+ 8. Limitation of Liability. In no event and under no legal theory,
154
+ whether in tort (including negligence), contract, or otherwise,
155
+ unless required by applicable law (such as deliberate and grossly
156
+ negligent acts) or agreed to in writing, shall any Contributor be
157
+ liable to You for damages, including any direct, indirect, special,
158
+ incidental, or consequential damages of any character arising as a
159
+ result of this License or out of the use or inability to use the
160
+ Work (including but not limited to damages for loss of goodwill,
161
+ work stoppage, computer failure or malfunction, or any and all
162
+ other commercial damages or losses), even if such Contributor
163
+ has been advised of the possibility of such damages.
164
+
165
+ 9. Accepting Warranty or Additional Liability. While redistributing
166
+ the Work or Derivative Works thereof, You may choose to offer,
167
+ and charge a fee for, acceptance of support, warranty, indemnity,
168
+ or other liability obligations and/or rights consistent with this
169
+ License. However, in accepting such obligations, You may act only
170
+ on Your own behalf and on Your sole responsibility, not on behalf
171
+ of any other Contributor, and only if You agree to indemnify,
172
+ defend, and hold each Contributor harmless for any liability
173
+ incurred by, or claims asserted against, such Contributor by reason
174
+ of your accepting any such warranty or additional liability.
175
+
176
+ END OF TERMS AND CONDITIONS
177
+
178
+ APPENDIX: How to apply the Apache License to your work.
179
+
180
+ To apply the Apache License to your work, attach the following
181
+ boilerplate notice, with the fields enclosed by brackets "[]"
182
+ replaced with your own identifying information. (Don't include
183
+ the brackets!) The text should be enclosed in the appropriate
184
+ comment syntax for the file format. We also recommend that a
185
+ file or class name and description of purpose be included on the
186
+ same "printed page" as the copyright notice for easier
187
+ identification within third-party archives.
188
+
189
+ Copyright [yyyy] [name of copyright owner]
190
+
191
+ Licensed under the Apache License, Version 2.0 (the "License");
192
+ you may not use this file except in compliance with the License.
193
+ You may obtain a copy of the License at
194
+
195
+ http://www.apache.org/licenses/LICENSE-2.0
196
+
197
+ Unless required by applicable law or agreed to in writing, software
198
+ distributed under the License is distributed on an "AS IS" BASIS,
199
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
200
+ See the License for the specific language governing permissions and
201
+ limitations under the License.
README.md ADDED
@@ -0,0 +1,262 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Reference-Image-Embed-Manga-Colorization
2
+
3
+ An amazing manga colorization project
4
+
5
+ You can colorize gray manga or character sketches using any reference image you want, this model will faithfully retain the color features and transfer them to your manga. This is useful when you wish the color of the character's hair or clothes to be consistent.
6
+
7
+ If the project is helpful, please leave a ⭐ this repo. best luck, my friend 😊 <br>
8
+
9
+ ## Overview
10
+ <p align="left">
11
+ <img src="./assets/network.png">
12
+ </p>
13
+
14
+ It's basically a cGAN(Conditional Generative Adversarial Network) architecture.
15
+
16
+ ### Generator
17
+
18
+ Generator is divided into two parts.
19
+
20
+ `Color Embedding Layer` consists of part of pretrained VGG19 net and an MLP(Multilayer Perceptron), which is used to extract `color embedding` from reference image(for training, its preprocessed Ground Truth Image).
21
+
22
+ Another part is a U-net-like network. The encoder layer extracts `content embedding` from gray input image(only contains L-channel information), and the decoder layer reconstructs the image with `color embedding` through PFFB(Progressive Feature Formalization Block) and outputs the ab_channel information.
23
+
24
+ <p align="left">
25
+ <img src="./assets/PFFB.png">
26
+ </p>
27
+
28
+ The figure shows how PFFB works.
29
+
30
+ It generates a filter by applying color embedding, and then convolving with content features. The figure is from this [paper](https://arxiv.org/abs/2106.08017) and check it for more details.
31
+
32
+ ### Discriminator
33
+
34
+ Discriminator is a PatchGAN, referring to [pix2pix](https://arxiv.org/abs/1611.07004v3). The difference is that there are two conditions used for input. One is the gray image waiting for colorization, and one is the reference image providing color information.
35
+
36
+ ### Loss
37
+
38
+ There are three losses in total, `L1 loss`, `perceptual loss` produced by pretrained vgg19, and `adversarial loss` produced by discriminator. The ratio is `1: 0.1: 0.01`.
39
+
40
+ ### Pipeline
41
+
42
+ <p align="left">
43
+ <img src="./assets/Pipeline.png">
44
+ </p>
45
+
46
+ - a. Segment panels from input manga image, `Manga-Panel-Extractor` is from [here](https://github.com/pvnieo/Manga-Panel-Extractor).
47
+ - b. Select a reference image for each panel, and generator will colorize each panel.
48
+ - c. Concatenate all colorized panels into original format.
49
+
50
+ ## Results
51
+ ### Gray model
52
+
53
+ | Original | Reference | Colorization |
54
+ |:----------:|:-----------:|:----------:|
55
+ | <img src="test_datasets/gray_test/001_in.png" width="400"> | <img src="test_datasets/gray_test/001_ref_a.png" width="200"> | <img src="test_datasets/gray_test/out/001_in_color_a.png" width="400"> |
56
+ | <img src="test_datasets/gray_test/001_in.png" width="400"> | <img src="test_datasets/gray_test/001_ref_b.png" width="200"> | <img src="test_datasets/gray_test/out/001_in_color_b.png" width="400"> |
57
+ | <img src="test_datasets/gray_test/002_in.jpeg" width="400"> | <img src="test_datasets/gray_test/002_in_ref_a.jpg" width="200"> | <img src="test_datasets/gray_test/out/002_in_color_a.png" width="400"> |
58
+ | <img src="test_datasets/gray_test/002_in.jpeg" width="400"> | <img src="test_datasets/gray_test/002_in_ref_b.jpeg" width="200"> | <img src="test_datasets/gray_test/out/002_in_color_b.png" width="400"> |
59
+ | <img src="test_datasets/gray_test/003_in.jpeg" width="400"> | <img src="test_datasets/gray_test/003_in_ref_a.jpg" width="200"> | <img src="test_datasets/gray_test/out/003_in_color_a.png" width="400"> |
60
+ | <img src="test_datasets/gray_test/003_in.jpeg" width="400"> | <img src="test_datasets/gray_test/003_in_ref_b.jpg" width="200"> | <img src="test_datasets/gray_test/out/003_in_color_b.png" width="400"> |
61
+ | <img src="test_datasets/gray_test/004_in.png" width="400"> |<img src="test_datasets/gray_test/004_ref_1.jpg" width="100"><img src="test_datasets/gray_test/004_ref_2.jpg" width="100">| <img src="test_datasets/gray_test/out/004_in_color.png" width="400">|
62
+ | <img src="test_datasets/gray_test/005_in.png" width="400"> | <img src="test_datasets/gray_test/005_ref_1.jpeg" width="100"><img src="test_datasets/gray_test/005_ref_2.jpg" width="100"><img src="test_datasets/gray_test/005_ref_3.jpeg" width="100"> | <img src="test_datasets/gray_test/out/005_in_color.png" width="400"> |
63
+ | <img src="test_datasets/gray_test/006_in.png" width="400"> | <img src="test_datasets/gray_test/006_ref.png" width="200"> | <img src="test_datasets/gray_test/out/006_in_color.png" width="400"> |
64
+
65
+ ### sketch model
66
+
67
+ | Original | Reference | Colorization |
68
+ | ------------------------------------------------------------ | ------------------------------------------------------------ | ------------------------------------------------------------ |
69
+ | <img src="test_datasets/sketch_test/001_in.jpg" width="400"> | <img src="test_datasets/sketch_test/001_ref_a.jpg" width="200"> | <img src="test_datasets/sketch_test/out/001_in_color_a.png" width="400"> |
70
+ | <img src="test_datasets/sketch_test/001_in.jpg" width="400"> | <img src="test_datasets/sketch_test/001_ref_b.jpg" width="200"> | <img src="test_datasets/sketch_test/out/001_in_color_b.png" width="400"> |
71
+
72
+
73
+
74
+ ## Dependencies and Installation
75
+
76
+ 1. Clone this GitHub repo.
77
+ ```
78
+ git clone https://github.com/linSensiGit/Example_Based_Manga_Colorization---cGAN.git
79
+
80
+ cd Example_Based_Manga_Colorization---cGAN
81
+ ```
82
+
83
+ 2. Create Environment
84
+ - Python >= 3.6 (Recommend to use [Anaconda](https://www.anaconda.com/download/#linux))
85
+
86
+ - [PyTorch >= 1.5.0](https://pytorch.org/) (Default GPU mode)
87
+
88
+ ```
89
+ # My environment for reference
90
+ - Python = 3.9.15
91
+ - PyTorch = 1.13.0
92
+ - Torchvision = 0.14.0
93
+ - Cuda = 11.7
94
+ - GPU = RTX 3060ti
95
+ ```
96
+
97
+ 3. Install Dependencies
98
+
99
+ ```
100
+ pip3 install -r requirement.txt
101
+ ```
102
+
103
+ ## Get Started
104
+
105
+ Once you've set up the environment, several things need to be done before colorization.
106
+
107
+ ### Prepare pretrained models
108
+
109
+ 1. Download generator. I have trained two generators, for [gray manga](https://drive.google.com/file/d/11RQGvBKySEtRcBdYD8O5ZLb54jB7SAgN/view?usp=drive_link) colorization and [sketch](https://drive.google.com/file/d/1I4XwOYIGAoQwMOicknZl0s6AWcwpARmR/view?usp=drive_link) colorization. Choose what you need.
110
+
111
+ 2. Download [VGG model](https://drive.google.com/file/d/1S7t3mD-tznEUrMmq5bRsLZk4fkN24QSV/view?usp=drive_link) , it's part of generator.
112
+
113
+ 3. Download discriminator, for training [gray manga](https://drive.google.com/file/d/1DHHE9um_xOm0brTpbHb_R7K7J4mn37FS/view?usp=drive_link) colorization and [sketch](https://drive.google.com/file/d/1WgIPYY4b4GcpHW9EWFrFoTxL9SlilQbN/view?usp=drive_link) colorization. (optional)
114
+
115
+ 4. Put the pretrained model in the correct directory:
116
+
117
+ ```
118
+ Colorful-Manga-GAN
119
+ |- experiments
120
+ |- Color2Manga_gray
121
+ |- xxx000_gray.pt
122
+ |- Color2Manga_sketch
123
+ |- xxx000_sketch.pt
124
+ |- Discriminator
125
+ |- xxx000_d.pt
126
+ |- VGG19
127
+ |- vgg19-dcbb9e9d.pth
128
+ ```
129
+
130
+ ### Quick test
131
+
132
+ I have collected some test datasets which contain manga pages and corresponding reference images. You can check it in the path `./test_datasets`. When you use the file `inference.py` to test, you may need to edit the input file path or pretrained weights path in this file.
133
+
134
+ ```
135
+ python inference.py
136
+
137
+ # If you don't want to segment your manga
138
+ python inference.py -ne
139
+ ```
140
+ Initially, `Manga-Panel-Extractor` will segment the manga page into panels.
141
+
142
+ Then follow the instructions in the console and you will get the colorized image.
143
+
144
+ ## Train your Own Model
145
+ ### Prepare Datasets
146
+
147
+ There are three datasets I used to train the model.
148
+
149
+ For gray model, [Anime Face Dataset](https://www.kaggle.com/datasets/scribbless/another-anime-face-dataset) and Tagged [Anime Illustrations Dataset](https://www.kaggle.com/datasets/mylesoneill/tagged-anime-illustrations) are used. And I only use `danbooru-images` folder in the second Dataset.
150
+
151
+ For sketch model, [Anime Sketch Colorization Pair Dataset](https://www.kaggle.com/datasets/ktaebum/anime-sketch-colorization-pair) is used.
152
+
153
+ All the datasets are from [Kaggle](https://www.kaggle.com/).
154
+
155
+ Follow instructions are based on my dataset, but feel free to use your own dataset if you like.
156
+
157
+ ### Preprocess training data
158
+
159
+ ```
160
+ cd data
161
+ python prepare_data.py
162
+ ```
163
+
164
+ If you are using ` Anime Sketch Colorization Pair` dataset :
165
+
166
+ ```
167
+ python prepare_data_sketch.py
168
+ ```
169
+
170
+ Several arguments needed to be assigned :
171
+
172
+ ```
173
+ usage: prepare_data.py [-h] [--out OUT] [--size SIZE] [--n_worker N_WORKER]
174
+ [--resample RESAMPLE]
175
+ path
176
+ positional arguments:
177
+ path the path of datasets
178
+ optional arguments:
179
+ -h, --help show this help message and exit
180
+ --out OUT the path to save generated lmdb
181
+ --size SIZE compressed image size (128, 256, 512, 1024) alternative
182
+ --n_worker N_WORKER The number of threads, depends on your CPU
183
+ --resample RESAMPLE
184
+ ```
185
+
186
+ For instance, you can run the command like this:
187
+
188
+ ```
189
+ python prepare_data.py --out ../train_datasets/Sketch_train_lmdb --n_worker 20 --size 256 E:/Dataset/animefaces256cleaner
190
+ ```
191
+
192
+ ### Training
193
+
194
+ There are four scripts in total for training
195
+
196
+ `train.py` —— train only generator
197
+
198
+ `train_disc` —— train only discriminator
199
+
200
+ `train_all_gray.py`—— train both generator and discriminator, under the usual dataset
201
+
202
+ `train_all_sketch.py`—— train both generator and discriminator, under sketch pair dataset specific
203
+
204
+
205
+
206
+ All of these scripts share similar commands to drive:
207
+
208
+ ```
209
+ usage: train_all_gray.py [-h] [--datasets DATASETS] [--iter ITER]
210
+ [--batch BATCH] [--size SIZE] [--ckpt CKPT]
211
+ [--ckpt_disc CKPT_DISC] [--lr LR] [--lr_disc LR_DISC]
212
+ [--experiment_name EXPERIMENT_NAME] [--wandb]
213
+ [--local_rank LOCAL_RANK]
214
+ optional arguments:
215
+ -h, --help show this help message and exit
216
+ --datasets DATASETS the path of training dataset
217
+ --iter ITER number of iteration in total
218
+ --batch BATCH batch size
219
+ --size SIZE size of image in dataset, usually 256
220
+ --ckpt CKPT path of pretrained generator
221
+ --ckpt_disc CKPT_DISC path of pretrained discriminator
222
+ --lr LR learning rate of generator
223
+ --lr_disc LR_DISC learning rate of discriminator
224
+ --experiment_name EXPERIMENT_NAME used to save training_logs and trained model
225
+ --wandb
226
+ --local_rank LOCAL_RANK
227
+ ```
228
+
229
+ There may be a slight difference, you could check the code for more details.
230
+
231
+
232
+
233
+ For instance, you can run the command like this:
234
+
235
+ ```
236
+ python train_all_gray.py --batch 8 --experiment_name Color2Manga_sketch --ckpt experiments/Color2Manga_sketch/078000.pt --datasets ./train_datasets/Sketch_train_lmdb --ckpt_disc experiments/Discriminator/078000_d.pt
237
+ ```
238
+
239
+ ## Work in Progress
240
+ - [ ] Add SR model instead of directly interpolate upscaling
241
+ - [ ] Optimize the generator network(adding L-channel information to output which is essential for colorize sketch)
242
+ - [ ] Better developed manga-panel-extractor(current segmentation is not precise enough)
243
+ - [ ] Develop a front UI and add color hint so that users could adjust the color of a specific area
244
+
245
+ ## 😁Contact
246
+
247
+ If you have any questions, please feel free to contact me via `shizifeng0615@outlook.com`
248
+
249
+ ## 🙌 Acknowledgement
250
+ Based on https://github.com/zhaohengyuan1/Color2Embed
251
+
252
+ Thx https://github.com/pvnieo/Manga-Panel-Extractor
253
+
254
+ ## Reference
255
+
256
+ [1] Zhao, Hengyuan et al. “Color2Embed: Fast Exemplar-Based Image Colorization using Color Embeddings.” (2021).
257
+
258
+ [2] Isola, Phillip et al. “Image-to-Image Translation with Conditional Adversarial Networks.” *2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)* (2016): 5967-5976.
259
+
260
+ [3] Furusawa, Chie et al. “Comicolorization: semi-automatic manga colorization.” *SIGGRAPH Asia 2017 Technical Briefs* (2017): n. pag.
261
+
262
+ [4] Satoshi Iizuka, Edgar Simo-Serra, and Hiroshi Ishikawa. "Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic Image Colorization with Simultaneous Classification". ACM Transaction on Graphics (Proc. of SIGGRAPH), 35(4):110, 2016.
__pycache__/distributed.cpython-310.pyc ADDED
Binary file (2.98 kB). View file
 
__pycache__/models.cpython-310.pyc ADDED
Binary file (6.13 kB). View file
 
__pycache__/models.cpython-38.pyc ADDED
Binary file (6.22 kB). View file
 
__pycache__/utils.cpython-310.pyc ADDED
Binary file (2.72 kB). View file
 
__pycache__/utils.cpython-38.pyc ADDED
Binary file (2.72 kB). View file
 
__pycache__/vgg_model.cpython-310.pyc ADDED
Binary file (3.97 kB). View file
 
__pycache__/vgg_model.cpython-38.pyc ADDED
Binary file (3.96 kB). View file
 
assets/PFFB.png ADDED
assets/Pipeline.png ADDED

Git LFS Details

  • SHA256: 86e46eb233e565ec37601e8af382f3c1fa87e939adce1aa84648485c97a93840
  • Pointer size: 132 Bytes
  • Size of remote file: 1.96 MB
assets/network.png ADDED

Git LFS Details

  • SHA256: 136ce29bf0db8f123947506ac32db2f13fefd4ad2b7b5f29db0623558ee66fc9
  • Pointer size: 132 Bytes
  • Size of remote file: 1.2 MB
data/__pycache__/data_loader.cpython-310.pyc ADDED
Binary file (3.04 kB). View file
 
data/__pycache__/tps_transformation.cpython-310.pyc ADDED
Binary file (1.1 kB). View file
 
data/data_loader.py ADDED
@@ -0,0 +1,97 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from io import BytesIO
2
+
3
+ import numpy as np
4
+ import lmdb
5
+ from PIL import Image
6
+ from skimage import color
7
+ import torch
8
+ from torch.utils.data import Dataset
9
+ from data.tps_transformation import tps_transform
10
+
11
+ def RGB2Lab(inputs):
12
+ return color.rgb2lab(inputs)
13
+
14
+ def Normalize(inputs):
15
+ # output l [-50,50] ab[-128,128]
16
+ l = inputs[:, :, 0:1]
17
+ ab = inputs[:, :, 1:3]
18
+ l = l - 50
19
+ # ab = ab
20
+ lab = np.concatenate((l, ab), 2)
21
+
22
+ return lab.astype('float32')
23
+
24
+ def selfnormalize(inputs):
25
+ d = torch.max(inputs) - torch.min(inputs)
26
+ out = (inputs) / d
27
+ return out
28
+
29
+ def to_gray(inputs):
30
+ img_gray = np.clip((np.concatenate((inputs[:,:,:1], inputs[:,:,:1], inputs[:,:,:1]), 2)+50)/100*255, 0, 255).astype('uint8')
31
+
32
+ return img_gray
33
+
34
+ def numpy2tensor(inputs):
35
+ out = torch.from_numpy(inputs.transpose(2,0,1))
36
+ return out
37
+
38
+ class MultiResolutionDataset(Dataset):
39
+ def __init__(self, path, transform, resolution=256):
40
+ self.env = lmdb.open(
41
+ path,
42
+ max_readers=32,
43
+ readonly=True,
44
+ lock=False,
45
+ readahead=False,
46
+ meminit=False,
47
+ )
48
+
49
+ if not self.env:
50
+ raise IOError('Cannot open lmdb dataset', path)
51
+
52
+ with self.env.begin(write=False) as txn:
53
+ self.length = int(txn.get('length'.encode('utf-8')).decode('utf-8'))
54
+
55
+ self.resolution = resolution
56
+ self.transform = transform
57
+
58
+ def __len__(self):
59
+ return self.length
60
+
61
+ def __getitem__(self, index):
62
+ with self.env.begin(write=False) as txn:
63
+ key = f'{self.resolution}-{str(index).zfill(5)}'.encode('utf-8')
64
+ img_bytes = txn.get(key)
65
+
66
+ buffer = BytesIO(img_bytes)
67
+ img = Image.open(buffer)
68
+ img_src = np.array(img) # [0,255] uint8
69
+
70
+ # ima_a = img_src
71
+ # ima_a = ima_a.astype('uint8')
72
+ # ima_a = Image.fromarray(ima_a)
73
+ # ima_a.show()
74
+
75
+ ## add gaussian noise
76
+ noise = np.random.uniform(-5, 5, np.shape(img_src))
77
+ img_ref = np.clip(np.array(img_src) + noise, 0, 255)
78
+
79
+
80
+ img_ref = tps_transform(img_ref) # [0,255] uint8
81
+ img_ref = np.clip(img_ref, 0, 255)
82
+ img_ref = img_ref.astype('uint8')
83
+ img_ref = Image.fromarray(img_ref)
84
+ img_ref = np.array(self.transform(img_ref)) # [0,255] uint8
85
+
86
+ img_lab = Normalize(RGB2Lab(img_src)) # l [-50,50] ab [-128, 128]
87
+
88
+ img = img_src.astype('float32') # [0,255] float32 RGB
89
+ img_ref = img_ref.astype('float32') # [0,255] float32 RGB
90
+
91
+ img = numpy2tensor(img)
92
+ img_ref = numpy2tensor(img_ref) # [B, 3, 256, 256]
93
+ img_lab = numpy2tensor(img_lab)
94
+
95
+ return img, img_ref, img_lab
96
+
97
+
data/data_loader_sketch.py ADDED
@@ -0,0 +1,120 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from io import BytesIO
2
+
3
+ import numpy as np
4
+ import lmdb
5
+ from PIL import Image
6
+ from skimage import color
7
+ import torch
8
+ from torch.utils.data import Dataset
9
+ from data.tps_transformation import tps_transform
10
+
11
+ def RGB2Lab(inputs):
12
+ return color.rgb2lab(inputs)
13
+
14
+ def Normalize(inputs):
15
+ # output l [-50,50] ab[-128,128]
16
+ l = inputs[:, :, 0:1]
17
+ ab = inputs[:, :, 1:3]
18
+ l = l - 50
19
+ # ab = ab
20
+ lab = np.concatenate((l, ab), 2)
21
+
22
+ return lab.astype('float32')
23
+
24
+ def selfnormalize(inputs):
25
+ d = torch.max(inputs) - torch.min(inputs)
26
+ out = (inputs) / d
27
+ return out
28
+
29
+ def to_gray(inputs):
30
+ img_gray = np.clip((np.concatenate((inputs[:,:,:1], inputs[:,:,:1], inputs[:,:,:1]), 2)+50)/100*255, 0, 255).astype('uint8')
31
+
32
+ return img_gray
33
+
34
+ def numpy2tensor(inputs):
35
+ out = torch.from_numpy(inputs.transpose(2,0,1))
36
+ return out
37
+
38
+ class MultiResolutionDataset(Dataset):
39
+ def __init__(self, path, transform, resolution=256):
40
+ self.env = lmdb.open(
41
+ path,
42
+ max_readers=32,
43
+ readonly=True,
44
+ lock=False,
45
+ readahead=False,
46
+ meminit=False,
47
+ )
48
+
49
+ if not self.env:
50
+ raise IOError('Cannot open lmdb dataset', path)
51
+
52
+ with self.env.begin(write=False) as txn:
53
+ self.length = int(txn.get('length'.encode('utf-8')).decode('utf-8'))
54
+
55
+ self.resolution = resolution
56
+ self.transform = transform
57
+
58
+ def __len__(self):
59
+ return self.length
60
+
61
+ def __getitem__(self, index):
62
+ with self.env.begin(write=False) as txn:
63
+ key = f'{self.resolution}-{str(index).zfill(5)}'.encode('utf-8')
64
+ img_bytes = txn.get(key)
65
+
66
+ buffer = BytesIO(img_bytes)
67
+ img = Image.open(buffer)
68
+ img_src = np.array(img) # [0,255] uint8
69
+
70
+ # ima_a = img_src
71
+ # ima_a = ima_a.astype('uint8')
72
+ # ima_a = Image.fromarray(ima_a)
73
+ # ima_a.show()
74
+
75
+ # get the left color image
76
+ img_ref = img_src[:, :256]
77
+ ## add gaussian noise
78
+ noise = np.random.uniform(-5, 5, np.shape(img_ref))
79
+ img_ref = np.clip(np.array(img_ref) + noise, 0, 255)
80
+
81
+
82
+ img_ref = tps_transform(img_ref) # [0,255] uint8
83
+ img_ref = np.clip(img_ref, 0, 255)
84
+ img_ref = img_ref.astype('uint8')
85
+ img_ref = Image.fromarray(img_ref)
86
+ img_ref = np.array(self.transform(img_ref)) # [0,255] uint8
87
+
88
+ img_lab = img_src[:, :256]
89
+ img_lab = Normalize(RGB2Lab(img_lab)) # l [-50,50] ab [-128, 128]
90
+
91
+ img_lab_sketch = img_src[:, 256:]
92
+ img_lab_sketch = Normalize(RGB2Lab(img_lab_sketch)) # l [-50,50] ab [-128, 128]
93
+
94
+ img = img_src[:, :256].astype('float32') # [0,255] float32 RGB
95
+ img_ref = img_ref.astype('float32') # [0,255] float32 RGB
96
+
97
+ # ima_a = img
98
+ # ima_a = ima_a.astype('uint8')
99
+ # ima_a = Image.fromarray(ima_a)
100
+ # ima_a.show()
101
+ #
102
+ # ima_a = img_ref
103
+ # ima_a = ima_a.astype('uint8')
104
+ # ima_a = Image.fromarray(ima_a)
105
+ # ima_a.show()
106
+ #
107
+ # ima_a = img_lab
108
+ # ima_a = ima_a.astype('uint8')
109
+ # ima_a = Image.fromarray(ima_a)
110
+ # ima_a.show()
111
+
112
+
113
+ img = numpy2tensor(img)
114
+ img_ref = numpy2tensor(img_ref) # [B, 3, 256, 256]
115
+ img_lab = numpy2tensor(img_lab)
116
+ img_lab_sketch = numpy2tensor(img_lab_sketch)
117
+
118
+ return img, img_ref, img_lab, img_lab_sketch
119
+
120
+
data/prepare_data.py ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import argparse
2
+ from io import BytesIO
3
+ import multiprocessing
4
+ from functools import partial
5
+
6
+ from PIL import Image
7
+ import lmdb
8
+ from tqdm import tqdm
9
+ from torchvision import datasets
10
+ from torchvision.transforms import functional as trans_fn
11
+
12
+
13
+ def resize_and_convert(img, size, resample, quality=100):
14
+ img = trans_fn.resize(img, size, resample)
15
+ img = trans_fn.center_crop(img, size)
16
+ buffer = BytesIO()
17
+ img.save(buffer, format='jpeg', quality=quality)
18
+ val = buffer.getvalue()
19
+
20
+ return val
21
+
22
+
23
+ def resize_multiple(img, sizes=(128, 256, 512, 1024), resample=Image.LANCZOS, quality=100):
24
+ imgs = []
25
+
26
+ for size in sizes:
27
+ imgs.append(resize_and_convert(img, size, resample, quality))
28
+
29
+ return imgs
30
+
31
+
32
+ def resize_worker(img_file, sizes, resample):
33
+ i, file = img_file
34
+ img = Image.open(file)
35
+ img = img.convert('RGB')
36
+ out = resize_multiple(img, sizes=sizes, resample=resample)
37
+
38
+ return i, out
39
+
40
+
41
+ def prepare(env, dataset, n_worker, sizes=(128, 256, 512, 1024), resample=Image.LANCZOS):
42
+ resize_fn = partial(resize_worker, sizes=sizes, resample=resample)
43
+
44
+ files = sorted(dataset.imgs, key=lambda x: x[0])
45
+ # print(files)
46
+ # eixt()
47
+ files = [(i, file) for i, (file, label) in enumerate(files)]
48
+ total = 0
49
+
50
+ with multiprocessing.Pool(n_worker) as pool:
51
+ for i, imgs in tqdm(pool.imap_unordered(resize_fn, files)):
52
+ for size, img in zip(sizes, imgs):
53
+ key = f'{size}-{str(i).zfill(5)}'.encode('utf-8')
54
+
55
+ with env.begin(write=True) as txn:
56
+ txn.put(key, img)
57
+
58
+ total += 1
59
+
60
+ with env.begin(write=True) as txn:
61
+ txn.put('length'.encode('utf-8'), str(total).encode('utf-8'))
62
+
63
+
64
+ if __name__ == '__main__':
65
+ parser = argparse.ArgumentParser()
66
+ parser.add_argument('--out', type=str)
67
+ parser.add_argument('--size', type=str, default='128,256,512,1024')
68
+ parser.add_argument('--n_worker', type=int, default=8)
69
+ parser.add_argument('--resample', type=str, default='lanczos')
70
+ parser.add_argument('path', type=str)
71
+
72
+ args = parser.parse_args()
73
+
74
+ resample_map = {'lanczos': Image.LANCZOS, 'bilinear': Image.BILINEAR}
75
+ resample = resample_map[args.resample]
76
+
77
+ sizes = [int(s.strip()) for s in args.size.split(',')]
78
+
79
+ print(f'Make dataset of image sizes:', ', '.join(str(s) for s in sizes))
80
+
81
+ imgset = datasets.ImageFolder(args.path)
82
+
83
+ with lmdb.open(args.out, map_size=6 * 1024 * 1024 * 1024, readahead=False) as env:
84
+ prepare(env, imgset, args.n_worker, sizes=sizes, resample=resample)
data/prepare_data_sketch.py ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import argparse
2
+ from io import BytesIO
3
+ import multiprocessing
4
+ from functools import partial
5
+
6
+ from PIL import Image
7
+ import lmdb
8
+ from tqdm import tqdm
9
+ from torchvision import datasets
10
+ from torchvision.transforms import functional as trans_fn
11
+
12
+
13
+ def resize_and_convert(img, size, resample, quality=100):
14
+ img = trans_fn.resize(img, size=[256, 512], interpolation=resample)
15
+ img = trans_fn.center_crop(img, output_size=[256, 512])
16
+ buffer = BytesIO()
17
+ img.save(buffer, format='jpeg', quality=quality)
18
+ val = buffer.getvalue()
19
+
20
+ return val
21
+
22
+
23
+ def resize_multiple(img, sizes=(128, 256, 512, 1024), resample=Image.LANCZOS, quality=100):
24
+ imgs = []
25
+
26
+ for size in sizes:
27
+ imgs.append(resize_and_convert(img, size, resample, quality))
28
+
29
+ return imgs
30
+
31
+
32
+ def resize_worker(img_file, sizes, resample):
33
+ i, file = img_file
34
+ img = Image.open(file)
35
+ img = img.convert('RGB')
36
+ out = resize_multiple(img, sizes=sizes, resample=resample)
37
+
38
+ return i, out
39
+
40
+
41
+ def prepare(env, dataset, n_worker, sizes=(128, 256, 512, 1024), resample=Image.LANCZOS):
42
+ resize_fn = partial(resize_worker, sizes=sizes, resample=resample)
43
+
44
+ files = sorted(dataset.imgs, key=lambda x: x[0])
45
+ # print(files)
46
+ # eixt()
47
+ files = [(i, file) for i, (file, label) in enumerate(files)]
48
+ total = 0
49
+
50
+ with multiprocessing.Pool(n_worker) as pool:
51
+ for i, imgs in tqdm(pool.imap_unordered(resize_fn, files)):
52
+ for size, img in zip(sizes, imgs):
53
+ key = f'{size}-{str(i).zfill(5)}'.encode('utf-8')
54
+
55
+ with env.begin(write=True) as txn:
56
+ txn.put(key, img)
57
+
58
+ total += 1
59
+
60
+ with env.begin(write=True) as txn:
61
+ txn.put('length'.encode('utf-8'), str(total).encode('utf-8'))
62
+
63
+
64
+ if __name__ == '__main__':
65
+ parser = argparse.ArgumentParser()
66
+ parser.add_argument('--out', type=str)
67
+ parser.add_argument('--size', type=str, default='128,256,512,1024')
68
+ parser.add_argument('--n_worker', type=int, default=8)
69
+ parser.add_argument('--resample', type=str, default='lanczos')
70
+ parser.add_argument('path', type=str)
71
+
72
+ args = parser.parse_args()
73
+
74
+ resample_map = {'lanczos': Image.LANCZOS, 'bilinear': Image.BILINEAR}
75
+ resample = resample_map[args.resample]
76
+
77
+ sizes = [int(s.strip()) for s in args.size.split(',')]
78
+
79
+ print(f'Make dataset of image sizes:', ', '.join(str(s) for s in sizes))
80
+
81
+ imgset = datasets.ImageFolder(args.path)
82
+
83
+ with lmdb.open(args.out, map_size=6 * 1024 * 1024 * 1024, readahead=False) as env:
84
+ prepare(env, imgset, args.n_worker, sizes=sizes, resample=resample)
data/thinplate/__init__.py ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ from data.thinplate.numpy import *
2
+
3
+ try:
4
+ import torch
5
+ import data.thinplate.pytorch as torch
6
+ except ImportError:
7
+ pass
8
+
9
+ __version__ = '1.0.0'
data/thinplate/__pycache__/__init__.cpython-310.pyc ADDED
Binary file (432 Bytes). View file
 
data/thinplate/__pycache__/numpy.cpython-310.pyc ADDED
Binary file (3.59 kB). View file
 
data/thinplate/__pycache__/pytorch.cpython-310.pyc ADDED
Binary file (4 kB). View file
 
data/thinplate/numpy.py ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2018 Christoph Heindl.
2
+ #
3
+ # Licensed under MIT License
4
+ # ============================================================
5
+
6
+ import numpy as np
7
+
8
+ class TPS:
9
+ @staticmethod
10
+ def fit(c, lambd=0., reduced=False):
11
+ n = c.shape[0]
12
+
13
+ U = TPS.u(TPS.d(c, c))
14
+ K = U + np.eye(n, dtype=np.float32)*lambd
15
+
16
+ P = np.ones((n, 3), dtype=np.float32)
17
+ P[:, 1:] = c[:, :2]
18
+
19
+ v = np.zeros(n+3, dtype=np.float32)
20
+ v[:n] = c[:, -1]
21
+
22
+ A = np.zeros((n+3, n+3), dtype=np.float32)
23
+ A[:n, :n] = K
24
+ A[:n, -3:] = P
25
+ A[-3:, :n] = P.T
26
+
27
+ theta = np.linalg.solve(A, v) # p has structure w,a
28
+ return theta[1:] if reduced else theta
29
+
30
+ @staticmethod
31
+ def d(a, b):
32
+ return np.sqrt(np.square(a[:, None, :2] - b[None, :, :2]).sum(-1))
33
+
34
+ @staticmethod
35
+ def u(r):
36
+ return r**2 * np.log(r + 1e-6)
37
+
38
+ @staticmethod
39
+ def z(x, c, theta):
40
+ x = np.atleast_2d(x)
41
+ U = TPS.u(TPS.d(x, c))
42
+ w, a = theta[:-3], theta[-3:]
43
+ reduced = theta.shape[0] == c.shape[0] + 2
44
+ if reduced:
45
+ w = np.concatenate((-np.sum(w, keepdims=True), w))
46
+ b = np.dot(U, w)
47
+ return a[0] + a[1]*x[:, 0] + a[2]*x[:, 1] + b
48
+
49
+ def uniform_grid(shape):
50
+ '''Uniform grid coordinates.
51
+
52
+ Params
53
+ ------
54
+ shape : tuple
55
+ HxW defining the number of height and width dimension of the grid
56
+
57
+ Returns
58
+ -------
59
+ points: HxWx2 tensor
60
+ Grid coordinates over [0,1] normalized image range.
61
+ '''
62
+
63
+ H,W = shape[:2]
64
+ c = np.empty((H, W, 2))
65
+ c[..., 0] = np.linspace(0, 1, W, dtype=np.float32)
66
+ c[..., 1] = np.expand_dims(np.linspace(0, 1, H, dtype=np.float32), -1)
67
+
68
+ return c
69
+
70
+ def tps_theta_from_points(c_src, c_dst, reduced=False):
71
+ delta = c_src - c_dst
72
+
73
+ cx = np.column_stack((c_dst, delta[:, 0]))
74
+ cy = np.column_stack((c_dst, delta[:, 1]))
75
+
76
+ theta_dx = TPS.fit(cx, reduced=reduced)
77
+ theta_dy = TPS.fit(cy, reduced=reduced)
78
+
79
+ return np.stack((theta_dx, theta_dy), -1)
80
+
81
+
82
+ def tps_grid(theta, c_dst, dshape):
83
+ ugrid = uniform_grid(dshape)
84
+
85
+ reduced = c_dst.shape[0] + 2 == theta.shape[0]
86
+
87
+ dx = TPS.z(ugrid.reshape((-1, 2)), c_dst, theta[:, 0]).reshape(dshape[:2])
88
+ dy = TPS.z(ugrid.reshape((-1, 2)), c_dst, theta[:, 1]).reshape(dshape[:2])
89
+ dgrid = np.stack((dx, dy), -1)
90
+
91
+ grid = dgrid + ugrid
92
+
93
+ return grid # H'xW'x2 grid[i,j] in range [0..1]
94
+
95
+ def tps_grid_to_remap(grid, sshape):
96
+ '''Convert a dense grid to OpenCV's remap compatible maps.
97
+
98
+ Params
99
+ ------
100
+ grid : HxWx2 array
101
+ Normalized flow field coordinates as computed by compute_densegrid.
102
+ sshape : tuple
103
+ Height and width of source image in pixels.
104
+
105
+
106
+ Returns
107
+ -------
108
+ mapx : HxW array
109
+ mapy : HxW array
110
+ '''
111
+
112
+ mx = (grid[:, :, 0] * sshape[1]).astype(np.float32)
113
+ my = (grid[:, :, 1] * sshape[0]).astype(np.float32)
114
+
115
+ return mx, my
data/thinplate/pytorch.py ADDED
@@ -0,0 +1,126 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2018 Christoph Heindl.
2
+ #
3
+ # Licensed under MIT License
4
+ # ============================================================
5
+
6
+ import torch
7
+
8
+ def tps(theta, ctrl, grid):
9
+ '''Evaluate the thin-plate-spline (TPS) surface at xy locations arranged in a grid.
10
+ The TPS surface is a minimum bend interpolation surface defined by a set of control points.
11
+ The function value for a x,y location is given by
12
+
13
+ TPS(x,y) := theta[-3] + theta[-2]*x + theta[-1]*y + \sum_t=0,T theta[t] U(x,y,ctrl[t])
14
+
15
+ This method computes the TPS value for multiple batches over multiple grid locations for 2
16
+ surfaces in one go.
17
+
18
+ Params
19
+ ------
20
+ theta: Nx(T+3)x2 tensor, or Nx(T+2)x2 tensor
21
+ Batch size N, T+3 or T+2 (reduced form) model parameters for T control points in dx and dy.
22
+ ctrl: NxTx2 tensor or Tx2 tensor
23
+ T control points in normalized image coordinates [0..1]
24
+ grid: NxHxWx3 tensor
25
+ Grid locations to evaluate with homogeneous 1 in first coordinate.
26
+
27
+ Returns
28
+ -------
29
+ z: NxHxWx2 tensor
30
+ Function values at each grid location in dx and dy.
31
+ '''
32
+
33
+ N, H, W, _ = grid.size()
34
+
35
+ if ctrl.dim() == 2:
36
+ ctrl = ctrl.expand(N, *ctrl.size())
37
+
38
+ T = ctrl.shape[1]
39
+
40
+ diff = grid[...,1:].unsqueeze(-2) - ctrl.unsqueeze(1).unsqueeze(1)
41
+ D = torch.sqrt((diff**2).sum(-1))
42
+ U = (D**2) * torch.log(D + 1e-6)
43
+
44
+ w, a = theta[:, :-3, :], theta[:, -3:, :]
45
+
46
+ reduced = T + 2 == theta.shape[1]
47
+ if reduced:
48
+ w = torch.cat((-w.sum(dim=1, keepdim=True), w), dim=1)
49
+
50
+ # U is NxHxWxT
51
+ b = torch.bmm(U.view(N, -1, T), w).view(N,H,W,2)
52
+ # b is NxHxWx2
53
+ z = torch.bmm(grid.view(N,-1,3), a).view(N,H,W,2) + b
54
+
55
+ return z
56
+
57
+ def tps_grid(theta, ctrl, size):
58
+ '''Compute a thin-plate-spline grid from parameters for sampling.
59
+
60
+ Params
61
+ ------
62
+ theta: Nx(T+3)x2 tensor
63
+ Batch size N, T+3 model parameters for T control points in dx and dy.
64
+ ctrl: NxTx2 tensor, or Tx2 tensor
65
+ T control points in normalized image coordinates [0..1]
66
+ size: tuple
67
+ Output grid size as NxCxHxW. C unused. This defines the output image
68
+ size when sampling.
69
+
70
+ Returns
71
+ -------
72
+ grid : NxHxWx2 tensor
73
+ Grid suitable for sampling in pytorch containing source image
74
+ locations for each output pixel.
75
+ '''
76
+ N, _, H, W = size
77
+
78
+ grid = theta.new(N, H, W, 3)
79
+ grid[:, :, :, 0] = 1.
80
+ grid[:, :, :, 1] = torch.linspace(0, 1, W)
81
+ grid[:, :, :, 2] = torch.linspace(0, 1, H).unsqueeze(-1)
82
+
83
+ z = tps(theta, ctrl, grid)
84
+ return (grid[...,1:] + z)*2-1 # [-1,1] range required by F.sample_grid
85
+
86
+ def tps_sparse(theta, ctrl, xy):
87
+ if xy.dim() == 2:
88
+ xy = xy.expand(theta.shape[0], *xy.size())
89
+
90
+ N, M = xy.shape[:2]
91
+ grid = xy.new(N, M, 3)
92
+ grid[..., 0] = 1.
93
+ grid[..., 1:] = xy
94
+
95
+ z = tps(theta, ctrl, grid.view(N,M,1,3))
96
+ return xy + z.view(N, M, 2)
97
+
98
+ def uniform_grid(shape):
99
+ '''Uniformly places control points aranged in grid accross normalized image coordinates.
100
+
101
+ Params
102
+ ------
103
+ shape : tuple
104
+ HxW defining the number of control points in height and width dimension
105
+
106
+ Returns
107
+ -------
108
+ points: HxWx2 tensor
109
+ Control points over [0,1] normalized image range.
110
+ '''
111
+ H,W = shape[:2]
112
+ c = torch.zeros(H, W, 2)
113
+ c[..., 0] = torch.linspace(0, 1, W)
114
+ c[..., 1] = torch.linspace(0, 1, H).unsqueeze(-1)
115
+ return c
116
+
117
+ if __name__ == '__main__':
118
+ c = torch.tensor([
119
+ [0., 0],
120
+ [1., 0],
121
+ [1., 1],
122
+ [0, 1],
123
+ ]).unsqueeze(0)
124
+ theta = torch.zeros(1, 4+3, 2)
125
+ size= (1,1,6,3)
126
+ print(tps_grid(theta, c, size).shape)
data/thinplate/tests/__init__.py ADDED
File without changes
data/thinplate/tests/test_tps_numpy.py ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ import numpy as np
3
+ from numpy.testing import assert_allclose
4
+ import thinplate as tps
5
+
6
+ def test_numpy_fit():
7
+ c = np.array([
8
+ [0., 0, 0.0],
9
+ [1., 0, 0.0],
10
+ [1., 1, 0.0],
11
+ [0, 1, 0.0],
12
+ ])
13
+
14
+ theta = tps.TPS.fit(c)
15
+ assert_allclose(theta, 0)
16
+ assert_allclose(tps.TPS.z(c, c, theta), c[:, 2])
17
+
18
+ c = np.array([
19
+ [0., 0, 1.0],
20
+ [1., 0, 1.0],
21
+ [1., 1, 1.0],
22
+ [0, 1, 1.0],
23
+ ])
24
+
25
+ theta = tps.TPS.fit(c)
26
+ assert_allclose(theta[:-3], 0)
27
+ assert_allclose(theta[-3:], [1, 0, 0])
28
+ assert_allclose(tps.TPS.z(c, c, theta), c[:, 2], atol=1e-3)
29
+
30
+ # reduced form
31
+ theta = tps.TPS.fit(c, reduced=True)
32
+ assert len(theta) == c.shape[0] + 2
33
+ assert_allclose(theta[:-3], 0)
34
+ assert_allclose(theta[-3:], [1, 0, 0])
35
+ assert_allclose(tps.TPS.z(c, c, theta), c[:, 2], atol=1e-3)
36
+
37
+ c = np.array([
38
+ [0., 0, -.5],
39
+ [1., 0, 0.5],
40
+ [1., 1, 0.2],
41
+ [0, 1, 0.8],
42
+ ])
43
+
44
+ theta = tps.TPS.fit(c)
45
+ assert_allclose(tps.TPS.z(c, c, theta), c[:, 2], atol=1e-3)
46
+
47
+ def test_numpy_densegrid():
48
+
49
+ # enlarges a small rectangle to full view
50
+
51
+ import cv2
52
+
53
+ img = np.zeros((40, 40), dtype=np.uint8)
54
+ img[10:21, 10:21] = 255
55
+
56
+ c_dst = np.array([
57
+ [0., 0],
58
+ [1., 0],
59
+ [1, 1],
60
+ [0, 1],
61
+ ])
62
+
63
+
64
+ c_src = np.array([
65
+ [10., 10],
66
+ [20., 10],
67
+ [20, 20],
68
+ [10, 20],
69
+ ]) / 40.
70
+
71
+ theta = tps.tps_theta_from_points(c_src, c_dst)
72
+ theta_r = tps.tps_theta_from_points(c_src, c_dst, reduced=True)
73
+
74
+ grid = tps.tps_grid(theta, c_dst, (20,20))
75
+ grid_r = tps.tps_grid(theta_r, c_dst, (20,20))
76
+
77
+ mapx, mapy = tps.tps_grid_to_remap(grid, img.shape)
78
+ warped = cv2.remap(img, mapx, mapy, cv2.INTER_CUBIC)
79
+
80
+ assert img.min() == 0.
81
+ assert img.max() == 255.
82
+ assert warped.shape == (20,20)
83
+ assert warped.min() == 255.
84
+ assert warped.max() == 255.
85
+ assert np.linalg.norm(grid.reshape(-1,2) - grid_r.reshape(-1,2)) < 1e-3
data/thinplate/tests/test_tps_pytorch.py ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+ import torch.optim as optim
3
+ import torch.nn.functional as F
4
+
5
+ import numpy as np
6
+ import thinplate as tps
7
+
8
+ from numpy.testing import assert_allclose
9
+
10
+ def test_pytorch_grid():
11
+
12
+ c_dst = np.array([
13
+ [0., 0],
14
+ [1., 0],
15
+ [1, 1],
16
+ [0, 1],
17
+ ], dtype=np.float32)
18
+
19
+
20
+ c_src = np.array([
21
+ [10., 10],
22
+ [20., 10],
23
+ [20, 20],
24
+ [10, 20],
25
+ ], dtype=np.float32) / 40.
26
+
27
+ theta = tps.tps_theta_from_points(c_src, c_dst)
28
+ theta_r = tps.tps_theta_from_points(c_src, c_dst, reduced=True)
29
+
30
+ np_grid = tps.tps_grid(theta, c_dst, (20,20))
31
+ np_grid_r = tps.tps_grid(theta_r, c_dst, (20,20))
32
+
33
+ pth_theta = torch.tensor(theta).unsqueeze(0)
34
+ pth_grid = tps.torch.tps_grid(pth_theta, torch.tensor(c_dst), (1, 1, 20, 20)).squeeze().numpy()
35
+ pth_grid = (pth_grid + 1) / 2 # convert [-1,1] range to [0,1]
36
+
37
+ pth_theta_r = torch.tensor(theta_r).unsqueeze(0)
38
+ pth_grid_r = tps.torch.tps_grid(pth_theta_r, torch.tensor(c_dst), (1, 1, 20, 20)).squeeze().numpy()
39
+ pth_grid_r = (pth_grid_r + 1) / 2 # convert [-1,1] range to [0,1]
40
+
41
+ assert_allclose(np_grid, pth_grid)
42
+ assert_allclose(np_grid_r, pth_grid_r)
43
+ assert_allclose(np_grid_r, np_grid)
data/tps_transformation.py ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import numpy as np
2
+ import data.thinplate as tps
3
+ import cv2
4
+ import random
5
+ import math
6
+
7
+ # Reference : https://github.com/cheind/py-thin-plate-spline
8
+
9
+ def tps_transform(img, dshape=None):
10
+
11
+ while True:
12
+ point1 = round(random.uniform(0.3, 0.7), 2)
13
+ point2 = round(random.uniform(0.3, 0.7), 2)
14
+ range_1 = round(random.uniform(-0.25, 0.25), 2)
15
+ range_2 = round(random.uniform(-0.25, 0.25), 2)
16
+ if math.isclose(point1 + range_1, point2 + range_2):
17
+ continue
18
+ else:
19
+ break
20
+
21
+ c_src = np.array([
22
+ [0.0, 0.0],
23
+ [1., 0],
24
+ [1, 1],
25
+ [0, 1],
26
+ [point1, point1],
27
+ [point2, point2],
28
+ ])
29
+
30
+ c_dst = np.array([
31
+ [0., 0],
32
+ [1., 0],
33
+ [1, 1],
34
+ [0, 1],
35
+ [point1 + range_1, point1 + range_1],
36
+ [point2 + range_2, point2 + range_2],
37
+ ])
38
+
39
+ dshape = dshape or img.shape
40
+ theta = tps.tps_theta_from_points(c_src, c_dst, reduced=True)
41
+ grid = tps.tps_grid(theta, c_dst, dshape)
42
+ mapx, mapy = tps.tps_grid_to_remap(grid, img.shape)
43
+ return cv2.remap(img, mapx, mapy, cv2.INTER_CUBIC)
44
+
discriminator.py ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch.nn as nn
2
+ import torch.nn.functional as F
3
+ import torch
4
+
5
+
6
+ class Discriminator(nn.Module):
7
+ def __init__(self, in_channels=3):
8
+ super(Discriminator, self).__init__()
9
+
10
+ def discriminator_block(in_filters, out_filters, normalization=True):
11
+ """Returns downsampling layers of each discriminator block"""
12
+ layers = [nn.Conv2d(in_filters, out_filters, 4, stride=2, padding=1)]
13
+ if normalization:
14
+ layers.append(nn.InstanceNorm2d(out_filters))
15
+ layers.append(nn.LeakyReLU(0.2, inplace=True))
16
+ return layers
17
+
18
+ self.model = nn.Sequential(
19
+ *discriminator_block(in_channels * 3, 64, normalization=False),
20
+ *discriminator_block(64, 128),
21
+ *discriminator_block(128, 256),
22
+ *discriminator_block(256, 512),
23
+ nn.ZeroPad2d((1, 0, 1, 0)),
24
+ nn.Conv2d(512, 1, 4, padding=1, bias=False)
25
+ )
26
+
27
+ def forward(self, img_out, img_l, img_ref ):
28
+ # Concatenate image and condition image by channels to produce input
29
+ img_input = torch.cat((img_out, img_l, img_ref), 1)
30
+ return self.model(img_input)
31
+
distributed.py ADDED
@@ -0,0 +1,126 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import math
2
+ import pickle
3
+
4
+ import torch
5
+ from torch import distributed as dist
6
+ from torch.utils.data.sampler import Sampler
7
+
8
+
9
+ def get_rank():
10
+ if not dist.is_available():
11
+ return 0
12
+
13
+ if not dist.is_initialized():
14
+ return 0
15
+
16
+ return dist.get_rank()
17
+
18
+
19
+ def synchronize():
20
+ if not dist.is_available():
21
+ return
22
+
23
+ if not dist.is_initialized():
24
+ return
25
+
26
+ world_size = dist.get_world_size()
27
+
28
+ if world_size == 1:
29
+ return
30
+
31
+ dist.barrier()
32
+
33
+
34
+ def get_world_size():
35
+ if not dist.is_available():
36
+ return 1
37
+
38
+ if not dist.is_initialized():
39
+ return 1
40
+
41
+ return dist.get_world_size()
42
+
43
+
44
+ def reduce_sum(tensor):
45
+ if not dist.is_available():
46
+ return tensor
47
+
48
+ if not dist.is_initialized():
49
+ return tensor
50
+
51
+ tensor = tensor.clone()
52
+ dist.all_reduce(tensor, op=dist.ReduceOp.SUM)
53
+
54
+ return tensor
55
+
56
+
57
+ def gather_grad(params):
58
+ world_size = get_world_size()
59
+
60
+ if world_size == 1:
61
+ return
62
+
63
+ for param in params:
64
+ if param.grad is not None:
65
+ dist.all_reduce(param.grad.data, op=dist.ReduceOp.SUM)
66
+ param.grad.data.div_(world_size)
67
+
68
+
69
+ def all_gather(data):
70
+ world_size = get_world_size()
71
+
72
+ if world_size == 1:
73
+ return [data]
74
+
75
+ buffer = pickle.dumps(data)
76
+ storage = torch.ByteStorage.from_buffer(buffer)
77
+ tensor = torch.ByteTensor(storage).to('cuda')
78
+
79
+ local_size = torch.IntTensor([tensor.numel()]).to('cuda')
80
+ size_list = [torch.IntTensor([0]).to('cuda') for _ in range(world_size)]
81
+ dist.all_gather(size_list, local_size)
82
+ size_list = [int(size.item()) for size in size_list]
83
+ max_size = max(size_list)
84
+
85
+ tensor_list = []
86
+ for _ in size_list:
87
+ tensor_list.append(torch.ByteTensor(size=(max_size,)).to('cuda'))
88
+
89
+ if local_size != max_size:
90
+ padding = torch.ByteTensor(size=(max_size - local_size,)).to('cuda')
91
+ tensor = torch.cat((tensor, padding), 0)
92
+
93
+ dist.all_gather(tensor_list, tensor)
94
+
95
+ data_list = []
96
+
97
+ for size, tensor in zip(size_list, tensor_list):
98
+ buffer = tensor.cpu().numpy().tobytes()[:size]
99
+ data_list.append(pickle.loads(buffer))
100
+
101
+ return data_list
102
+
103
+
104
+ def reduce_loss_dict(loss_dict):
105
+ world_size = get_world_size()
106
+
107
+ if world_size < 2:
108
+ return loss_dict
109
+
110
+ with torch.no_grad():
111
+ keys = []
112
+ losses = []
113
+
114
+ for k in sorted(loss_dict.keys()):
115
+ keys.append(k)
116
+ losses.append(loss_dict[k])
117
+
118
+ losses = torch.stack(losses, 0)
119
+ dist.reduce(losses, dst=0)
120
+
121
+ if dist.get_rank() == 0:
122
+ losses /= world_size
123
+
124
+ reduced_losses = {k: v for k, v in zip(keys, losses)}
125
+
126
+ return reduced_losses
experiments/Color2Manga_gray/074000_gray.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e2f4785d00d4463ecb5f02d79f79f9da747a57179b5b016408e65da0e4f62572
3
+ size 1091510163
experiments/Color2Manga_sketch/116000_sketch.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:52505452ec908ffd1ae4499a205a55263a8cd7d7bdf4623b59edccf8e8636d33
3
+ size 1091510163
experiments/Discriminator/074000_d.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e622a55cb8bae33377e85963eaa496d7e9bd9e1f4449b853d41235729cc7d40f
3
+ size 33261919
experiments/Discriminator/116000_d.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:169ae73ef7c788ec3921c918d0a9ebdecc4115492b177dfd98660b7816d6ce5a
3
+ size 33261983
experiments/VGG19/vgg19-dcbb9e9d.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dcbb9e9dad569fff7a846263a77324fc34978fea2bfb039c012d710e1776ae44
3
+ size 574673361
extractor/Open-Sans-Bold.ttf ADDED
Binary file (225 kB). View file
 
extractor/__pycache__/manga_panel_extractor.cpython-310.pyc ADDED
Binary file (5.89 kB). View file
 
extractor/__pycache__/manga_panel_extractor.cpython-38.pyc ADDED
Binary file (5.89 kB). View file
 
extractor/manga_panel_extractor.py ADDED
@@ -0,0 +1,174 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # stdlib
2
+ import argparse
3
+ from argparse import RawTextHelpFormatter
4
+ import os
5
+ from os.path import splitext, basename, exists, join
6
+ from os import makedirs
7
+ # 3p
8
+ from tqdm import tqdm
9
+ import numpy as np
10
+ from skimage import measure
11
+ from PIL import Image
12
+ from PIL import ImageFont
13
+ from PIL import ImageDraw
14
+ import cv2
15
+ # project
16
+ from utils import get_files, load_image
17
+ from skimage import io
18
+
19
+
20
+ class PanelExtractor:
21
+ def __init__(self, min_pct_panel=2, max_pct_panel=90, paper_th=0.35):
22
+ assert min_pct_panel < max_pct_panel, "Minimum percentage must be smaller than maximum percentage"
23
+ self.min_panel = min_pct_panel / 100
24
+ self.max_panel = max_pct_panel / 100
25
+ self.paper_th = paper_th
26
+
27
+ def _generate_panel_blocks(self, img):
28
+ img = img if len(img.shape) == 2 else img[:, :, 0]
29
+ blur = cv2.GaussianBlur(img, (5, 5), 0)
30
+ thresh = cv2.threshold(blur, 230, 255, cv2.THRESH_BINARY)[1]
31
+ cv2.rectangle(thresh, (0, 0), tuple(img.shape[::-1]), (0, 0, 0), 10)
32
+ num_labels, labels, stats, centroids = cv2.connectedComponentsWithStats(thresh, 4, cv2.CV_32S)
33
+ ind = np.argsort(stats[:, 4], )[::-1][1]
34
+ panel_block_mask = ((labels == ind) * 255).astype("uint8")
35
+ # Image.fromarray(panel_block_mask).show()
36
+ return panel_block_mask
37
+
38
+ def generate_panels(self, img):
39
+ block_mask = self._generate_panel_blocks(img)
40
+ cv2.rectangle(block_mask, (0, 0), tuple(block_mask.shape[::-1]), (255, 255, 255), 10)
41
+ # Image.fromarray(block_mask).show()
42
+
43
+ # detect contours
44
+ contours, hierarchy = cv2.findContours(block_mask, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
45
+ panels = []
46
+ masks = []
47
+ panel_masks = []
48
+ # print(len(contours))
49
+
50
+ for i in range(len(contours)):
51
+ area = cv2.contourArea(contours[i])
52
+ img_area = img.shape[0] * img.shape[1]
53
+
54
+ # if the contour is very small or very big, it's likely wrongly detected
55
+ if area < (self.min_panel * img_area) or area > (self.max_panel * img_area):
56
+ continue
57
+
58
+ x, y, w, h = cv2.boundingRect(contours[i])
59
+ masks.append(cv2.boundingRect(contours[i]))
60
+ # create panel mask
61
+ panel_mask = np.ones_like(block_mask, "int32")
62
+ cv2.fillPoly(panel_mask, [contours[i].astype("int32")], color=(0, 0, 0))
63
+ # Image.fromarray(panel_mask).show()
64
+ panel_mask = panel_mask[y:y + h, x:x + w].copy()
65
+ # Image.fromarray(panel_mask).show()
66
+
67
+ # apply panel mask
68
+ panel = img[y:y + h, x:x + w].copy()
69
+ # Image.fromarray(panel).show()
70
+ panel[panel_mask == 1] = 255
71
+ # Image.fromarray(panel).show()
72
+
73
+ panels.append(panel)
74
+ panel_masks.append(panel_mask)
75
+
76
+ return panels, masks, panel_masks
77
+
78
+ def extract(self, folder):
79
+ print("Loading images ... ", end="")
80
+ # image_list, _, _ = get_files(folder)
81
+ image_list = []
82
+ image_list.append(folder)
83
+ imgs = [load_image(x) for x in image_list]
84
+ print("Done!")
85
+
86
+ folder = os.path.dirname(folder)
87
+ # create panels dir
88
+ if not exists(join(folder, "panels")):
89
+ makedirs(join(folder, "panels"))
90
+ folder = join(folder, "panels")
91
+
92
+ # remove images with paper texture, not well segmented
93
+ paperless_imgs = []
94
+ for img in tqdm(imgs, desc="Removing images with paper texture"):
95
+ hist, bins = np.histogram(img.copy().ravel(), 256, [0, 256])
96
+ if np.sum(hist[50:200]) / np.sum(hist) < self.paper_th:
97
+ paperless_imgs.append(img)
98
+
99
+ if not paperless_imgs:
100
+ return imgs, [], []
101
+ for i, img in tqdm(enumerate(paperless_imgs), desc="extracting panels"):
102
+ panels, masks, panel_masks = self.generate_panels(img)
103
+ name, ext = splitext(basename(image_list[i]))
104
+ for j, panel in enumerate(panels):
105
+ cv2.imwrite(join(folder, f'{name}_{j}.{ext}'), panel)
106
+
107
+ # show the order of colorized panels
108
+ img = Image.fromarray(img)
109
+ draw = ImageDraw.Draw(img)
110
+ font = ImageFont.truetype('extractor/Open-Sans-Bold.ttf', 160)
111
+
112
+ def flatten(l):
113
+ for el in l:
114
+ if isinstance(el, list):
115
+ yield from flatten(el)
116
+ else:
117
+ yield el
118
+
119
+ for i, bbox in enumerate(flatten(masks), start=1):
120
+ w, h = draw.textsize(str(i), font=font)
121
+ y = (bbox[1] + bbox[3] / 2 - h / 2)
122
+ x = (bbox[0] + bbox[2] / 2 - w / 2)
123
+ draw.text((x, y), str(i), (255, 215, 0), font=font)
124
+ img.show()
125
+ return panels, masks, panel_masks
126
+
127
+ def concatPanels(self, img_file, fake_imgs, masks, panel_masks):
128
+ img = io.imread(img_file)
129
+ # out_imgs.append(f"D:\MyProject\Python\DL_learning\Manga-Panel-Extractor-master\out\in0_ref0.png")
130
+ # out_imgs.append(f"D:\MyProject\Python\DL_learning\Manga-Panel-Extractor-master\out\in1_ref1.png")
131
+ # out_imgs.append(f"D:\MyProject\Python\DL_learning\Manga-Panel-Extractor-master\out\in2_ref2.png")
132
+ for i in range(len(fake_imgs)):
133
+ x, y, w, h = masks[i]
134
+ # fake_img = io.imread(fake_imgs[i])
135
+ # fake_img = np.array(fake_img)
136
+ fake_img = fake_imgs[i]
137
+ panel_mask = panel_masks[i]
138
+ img[y:y + h, x:x + w][panel_mask == 0] = fake_img[panel_mask == 0]
139
+ # Image.fromarray(img).show()
140
+ out_folder = os.path.dirname(img_file)
141
+ out_name = os.path.basename(img_file)
142
+ out_name = os.path.splitext(out_name)[0]
143
+ out_img_path = os.path.join(out_folder,'color',f'{out_name}_color.png')
144
+
145
+ # show image
146
+ Image.fromarray(img).show()
147
+ # save image
148
+ folder_path = os.path.join(out_folder, 'color')
149
+ if not os.path.exists(folder_path):
150
+ os.mkdir(folder_path)
151
+ io.imsave(out_img_path, img)
152
+
153
+
154
+ def main(args):
155
+ panel_extractor = PanelExtractor(min_pct_panel=args.min_panel, max_pct_panel=args.max_panel)
156
+ panels, masks, panel_masks = panel_extractor.extract(args.folder)
157
+ panel_extractor.concatPanels(args.folder, [], masks, panel_masks)
158
+
159
+
160
+ if __name__ == "__main__":
161
+ parser = argparse.ArgumentParser(
162
+ description="Implementation of a Manga Panel Extractor and dialogue bubble text eraser.",
163
+ formatter_class=RawTextHelpFormatter
164
+ )
165
+ parser.add_argument("-minp", "--min_panel", type=int, choices=range(1, 99), default=5, metavar="[1-99]",
166
+ help="Percentage of minimum panel area in relation to total page area.")
167
+ parser.add_argument("-maxp", "--max_panel", type=int, choices=range(1, 99), default=90, metavar="[1-99]",
168
+ help="Percentage of minimum panel area in relation to total page area.")
169
+ parser.add_argument("-f", '--folder', default='./images/002.png', type=str,
170
+ help="""folder path to input manga pages.
171
+ Panels will be saved to a directory named `panels` in this folder.""")
172
+
173
+ args = parser.parse_args()
174
+ main(args)
inference.py ADDED
@@ -0,0 +1,229 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ import numpy as np
3
+ from skimage import color, io
4
+
5
+ import torch
6
+ import torch.nn.functional as F
7
+
8
+ from PIL import Image
9
+ from models import ColorEncoder, ColorUNet
10
+ from extractor.manga_panel_extractor import PanelExtractor
11
+ import argparse
12
+
13
+ os.environ["CUDA_VISIBLE_DEVICES"] = '0'
14
+
15
+ def mkdirs(path):
16
+ if not os.path.exists(path):
17
+ os.makedirs(path)
18
+
19
+ def Lab2RGB_out(img_lab):
20
+ img_lab = img_lab.detach().cpu()
21
+ img_l = img_lab[:,:1,:,:]
22
+ img_ab = img_lab[:,1:,:,:]
23
+ # print(torch.max(img_l), torch.min(img_l))
24
+ # print(torch.max(img_ab), torch.min(img_ab))
25
+ img_l = img_l + 50
26
+ pred_lab = torch.cat((img_l, img_ab), 1)[0,...].numpy()
27
+ # grid_lab = utils.make_grid(pred_lab, nrow=1).numpy().astype("float64")
28
+ # print(grid_lab.shape)
29
+ out = (np.clip(color.lab2rgb(pred_lab.transpose(1, 2, 0)), 0, 1)* 255).astype("uint8")
30
+ return out
31
+
32
+ def RGB2Lab(inputs):
33
+ return color.rgb2lab(inputs)
34
+
35
+ def Normalize(inputs):
36
+ l = inputs[:, :, 0:1]
37
+ ab = inputs[:, :, 1:3]
38
+ l = l - 50
39
+ lab = np.concatenate((l, ab), 2)
40
+
41
+ return lab.astype('float32')
42
+
43
+ def numpy2tensor(inputs):
44
+ out = torch.from_numpy(inputs.transpose(2,0,1))
45
+ return out
46
+
47
+ def tensor2numpy(inputs):
48
+ out = inputs[0,...].detach().cpu().numpy().transpose(1,2,0)
49
+ return out
50
+
51
+ def preprocessing(inputs):
52
+ # input: rgb, [0, 255], uint8
53
+ img_lab = Normalize(RGB2Lab(inputs))
54
+ img = np.array(inputs, 'float32') # [0, 255]
55
+ img = numpy2tensor(img)
56
+ img_lab = numpy2tensor(img_lab)
57
+ return img.unsqueeze(0), img_lab.unsqueeze(0)
58
+
59
+ if __name__ == "__main__":
60
+ device = "cuda"
61
+
62
+ # model_name = 'Color2Manga_sketch'
63
+ ckpt_path = 'experiments/Color2Manga_gray/074000_gray.pt'
64
+ test_dir_path = 'test_datasets/gray_test'
65
+ no_extractor = False
66
+ # imgs_num = len(os.listdir(test_dir_path)) // 2
67
+ imgsize = 256
68
+
69
+ parser = argparse.ArgumentParser()
70
+
71
+ parser.add_argument("--path", type=str, default=None, help="path of input image")
72
+ parser.add_argument("--size", type=int, default=None)
73
+ parser.add_argument("--ckpt", type=str, default=None, help="path of model weight")
74
+ parser.add_argument("-ne", "--no_extractor", action='store_true',
75
+ help="Do not segment the manga panels.")
76
+
77
+ args = parser.parse_args()
78
+
79
+ if args.path:
80
+ ckpt_path = args.path
81
+ if args.size:
82
+ imgsize = args.size
83
+ if args.ckpt:
84
+ test_dir_path = args.ckpt
85
+ if args.no_extractor:
86
+ no_extractor = args.no_extractor
87
+
88
+
89
+ ckpt = torch.load(ckpt_path, map_location=lambda storage, loc: storage)
90
+
91
+ colorEncoder = ColorEncoder().to(device)
92
+ colorEncoder.load_state_dict(ckpt["colorEncoder"])
93
+ colorEncoder.eval()
94
+
95
+ colorUNet = ColorUNet().to(device)
96
+ colorUNet.load_state_dict(ckpt["colorUNet"])
97
+ colorUNet.eval()
98
+
99
+ imgs = []
100
+ imgs_lab = []
101
+
102
+ # for i in range(imgs_num):
103
+ # idx = i
104
+ # print('Image', idx, 'Input Image', 'in%d.JPEG'%idx, 'Ref Image', 'ref%d.JPEG'%idx)
105
+
106
+ while 1:
107
+ print(f'make sure both manga image and reference images are under this path{test_dir_path}')
108
+ img_path = input("please input the name of image needed to be colorized(with file extension): ")
109
+ img_path = os.path.join(test_dir_path, img_path)
110
+ img_name = os.path.basename(img_path)
111
+ img_name = os.path.splitext(img_name)[0]
112
+
113
+ if no_extractor:
114
+ ref_img_path = os.path.join(test_dir_path, input(f"{1}/{1} reference image:"))
115
+
116
+ img1 = Image.open(img_path).convert("RGB")
117
+ width, height = img1.size
118
+ img2 = Image.open(ref_img_path).convert("RGB")
119
+
120
+ img1, img1_lab = preprocessing(img1)
121
+ img2, img2_lab = preprocessing(img2)
122
+
123
+ img1 = img1.to(device)
124
+ img1_lab = img1_lab.to(device)
125
+ img2 = img2.to(device)
126
+ img2_lab = img2_lab.to(device)
127
+
128
+ # print('-------',torch.max(img1_lab[:,:1,:,:]), torch.min(img1_lab[:,1:,:,:]))
129
+
130
+ with torch.no_grad():
131
+ img2_resize = F.interpolate(img2 / 255., size=(imgsize, imgsize), mode='bilinear',
132
+ recompute_scale_factor=False, align_corners=False)
133
+ img1_L_resize = F.interpolate(img1_lab[:, :1, :, :] / 50., size=(imgsize, imgsize), mode='bilinear',
134
+ recompute_scale_factor=False, align_corners=False)
135
+
136
+ color_vector = colorEncoder(img2_resize)
137
+
138
+ fake_ab = colorUNet((img1_L_resize, color_vector))
139
+ fake_ab = F.interpolate(fake_ab * 110, size=(height, width), mode='bilinear',
140
+ recompute_scale_factor=False, align_corners=False)
141
+
142
+ fake_img = torch.cat((img1_lab[:, :1, :, :], fake_ab), 1)
143
+ fake_img = Lab2RGB_out(fake_img)
144
+ # io.imsave(out_img_path, fake_img)
145
+
146
+ out_folder = os.path.dirname(img_path)
147
+ out_name = os.path.basename(img_path)
148
+ out_name = os.path.splitext(out_name)[0]
149
+ out_img_path = os.path.join(out_folder, 'color', f'{out_name}_color.png')
150
+
151
+ # show image
152
+ Image.fromarray(fake_img).show()
153
+ # save image
154
+ folder_path = os.path.join(out_folder, 'color')
155
+ if not os.path.exists(folder_path):
156
+ os.mkdir(folder_path)
157
+ io.imsave(out_img_path, fake_img)
158
+
159
+ continue
160
+
161
+
162
+
163
+ # extract panels from manga
164
+ panel_extractor = PanelExtractor(min_pct_panel=5, max_pct_panel=90)
165
+ panels, masks, panel_masks = panel_extractor.extract(img_path)
166
+ panel_num = len(panels)
167
+
168
+ ref_img_paths = []
169
+ # ref_img_path = os.path.join(test_dir_path, '%03d_ref.png' % idx)
170
+ print("Please enter the name of the reference image in order according to the number prompts on the picture")
171
+ for i in range(panel_num):
172
+ ref_img_path = os.path.join(test_dir_path, input(f"{i+1}/{panel_num} reference image:"))
173
+ ref_img_paths.append(ref_img_path)
174
+
175
+
176
+
177
+
178
+ fake_imgs = []
179
+ for i in range(panel_num):
180
+ img1 = Image.fromarray(panels[i]).convert("RGB")
181
+ width, height = img1.size
182
+ img2 = Image.open(ref_img_paths[i]).convert("RGB")
183
+
184
+ # img1 = Image.open(img_path).convert("RGB")
185
+ # width, height = img1.size
186
+ # img2 = Image.open(ref_img_path).convert("RGB")
187
+
188
+ img1, img1_lab = preprocessing(img1)
189
+ img2, img2_lab = preprocessing(img2)
190
+
191
+ img1 = img1.to(device)
192
+ img1_lab = img1_lab.to(device)
193
+ img2 = img2.to(device)
194
+ img2_lab = img2_lab.to(device)
195
+
196
+ # print('-------',torch.max(img1_lab[:,:1,:,:]), torch.min(img1_lab[:,1:,:,:]))
197
+
198
+ with torch.no_grad():
199
+ img2_resize = F.interpolate(img2 / 255., size=(imgsize, imgsize), mode='bilinear', recompute_scale_factor=False, align_corners=False)
200
+ img1_L_resize = F.interpolate(img1_lab[:,:1,:,:] / 50., size=(imgsize, imgsize), mode='bilinear', recompute_scale_factor=False, align_corners=False)
201
+
202
+ color_vector = colorEncoder(img2_resize)
203
+
204
+ fake_ab = colorUNet((img1_L_resize, color_vector))
205
+ fake_ab = F.interpolate(fake_ab*110, size=(height, width), mode='bilinear', recompute_scale_factor=False, align_corners=False)
206
+
207
+ fake_img = torch.cat((img1_lab[:,:1,:,:], fake_ab), 1)
208
+ fake_img = Lab2RGB_out(fake_img)
209
+ # io.imsave(f'test_datasets/gray_test/panels/{i}.png', fake_img)
210
+ fake_imgs.append(fake_img)
211
+
212
+ if panel_num == 1:
213
+ out_folder = os.path.dirname(img_path)
214
+ out_name = os.path.basename(img_path)
215
+ out_name = os.path.splitext(out_name)[0]
216
+ out_img_path = os.path.join(out_folder,'color',f'{out_name}_color.png')
217
+
218
+ # show image
219
+ Image.fromarray(fake_imgs[0]).show()
220
+ # save image
221
+ folder_path = os.path.join(out_folder, 'color')
222
+ if not os.path.exists(folder_path):
223
+ os.mkdir(folder_path)
224
+ io.imsave(out_img_path, fake_imgs[0])
225
+ else:
226
+ panel_extractor.concatPanels(img_path, fake_imgs, masks, panel_masks)
227
+
228
+ print(f'colored image has been put to: {test_dir_path}color')
229
+
models.py ADDED
@@ -0,0 +1,223 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import math
2
+
3
+ import torch
4
+ import torch.nn as nn
5
+ import torch.nn.functional as F
6
+
7
+ from vgg_model import vgg19
8
+
9
+ class DoubleConv(nn.Module):
10
+ """(convolution => [BN] => ReLU) * 2"""
11
+
12
+ def __init__(self, in_channels, out_channels, mid_channels=None):
13
+ super().__init__()
14
+ if not mid_channels:
15
+ mid_channels = out_channels
16
+ self.double_conv = nn.Sequential(
17
+ nn.Conv2d(in_channels, mid_channels, kernel_size=3, padding=1),
18
+ nn.BatchNorm2d(mid_channels),
19
+ nn.LeakyReLU(0.1, True),
20
+ nn.Conv2d(mid_channels, out_channels, kernel_size=3, padding=1),
21
+ nn.BatchNorm2d(out_channels),
22
+ nn.LeakyReLU(0.1, True)
23
+ )
24
+
25
+ def forward(self, x):
26
+ x = self.double_conv(x)
27
+ return x
28
+
29
+ class ResBlock(nn.Module):
30
+ """(convolution => [BN] => ReLU) * 2"""
31
+
32
+ def __init__(self, in_channels, out_channels):
33
+ super().__init__()
34
+ self.bottle_conv = nn.Conv2d(in_channels, out_channels, 1, 1, 0)
35
+ self.double_conv = nn.Sequential(
36
+ nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1),
37
+ nn.BatchNorm2d(out_channels),
38
+ nn.LeakyReLU(0.2, True),
39
+ nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1)
40
+ )
41
+
42
+ def forward(self, x):
43
+ x = self.bottle_conv(x)
44
+ x = self.double_conv(x) + x
45
+ return x / math.sqrt(2)
46
+
47
+
48
+ class Down(nn.Module):
49
+ """Downscaling with stride conv then double conv"""
50
+
51
+ def __init__(self, in_channels, out_channels):
52
+ super().__init__()
53
+ self.main = nn.Sequential(
54
+ nn.Conv2d(in_channels, in_channels, 4, 2, 1),
55
+ nn.LeakyReLU(0.1, True),
56
+ # DoubleConv(in_channels, out_channels)
57
+ ResBlock(in_channels, out_channels)
58
+ )
59
+
60
+
61
+ def forward(self, x):
62
+
63
+ x = self.main(x)
64
+
65
+ return x
66
+
67
+ class SDFT(nn.Module):
68
+
69
+ def __init__(self, color_dim, channels, kernel_size = 3):
70
+ super().__init__()
71
+
72
+ # generate global conv weights
73
+ fan_in = channels * kernel_size ** 2
74
+ self.kernel_size = kernel_size
75
+ self.padding = kernel_size // 2
76
+
77
+ self.scale = 1 / math.sqrt(fan_in)
78
+ self.modulation = nn.Conv2d(color_dim, channels, 1)
79
+ self.weight = nn.Parameter(
80
+ torch.randn(1, channels, channels, kernel_size, kernel_size)
81
+ )
82
+
83
+ def forward(self, fea, color_style):
84
+ # for global adjustation
85
+ B, C, H, W = fea.size()
86
+ # print(fea.shape, color_style.shape)
87
+ style = self.modulation(color_style).view(B, 1, C, 1, 1)
88
+ weight = self.scale * self.weight * style
89
+ # demodulation
90
+ demod = torch.rsqrt(weight.pow(2).sum([2, 3, 4]) + 1e-8)
91
+ weight = weight * demod.view(B, C, 1, 1, 1)
92
+
93
+ weight = weight.view(
94
+ B * C, C, self.kernel_size, self.kernel_size
95
+ )
96
+
97
+ fea = fea.view(1, B * C, H, W)
98
+ fea = F.conv2d(fea, weight, padding=self.padding, groups=B)
99
+ fea = fea.view(B, C, H, W)
100
+
101
+ return fea
102
+
103
+
104
+ class UpBlock(nn.Module):
105
+
106
+
107
+ def __init__(self, color_dim, in_channels, out_channels, kernel_size = 3, bilinear=True):
108
+ super().__init__()
109
+
110
+ # if bilinear, use the normal convolutions to reduce the number of channels
111
+ if bilinear:
112
+ self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False)
113
+
114
+ else:
115
+ self.up = nn.ConvTranspose2d(in_channels , in_channels // 2, kernel_size=2, stride=2)
116
+
117
+ self.conv_cat = nn.Sequential(
118
+ nn.Conv2d(in_channels // 2 + in_channels // 8, out_channels, 1, 1, 0),
119
+ nn.LeakyReLU(0.2, True),
120
+ nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1),
121
+ nn.LeakyReLU(0.2, True)
122
+ )
123
+
124
+ self.conv_s = nn.Conv2d(in_channels//2, out_channels, 1, 1, 0)
125
+
126
+ # generate global conv weights
127
+ self.SDFT = SDFT(color_dim, out_channels, kernel_size)
128
+
129
+
130
+ def forward(self, x1, x2, color_style):
131
+ # print(x1.shape, x2.shape, color_style.shape)
132
+ x1 = self.up(x1)
133
+ x1_s = self.conv_s(x1)
134
+
135
+ x = torch.cat([x1, x2[:, ::4, :, :]], dim=1)
136
+ x = self.conv_cat(x)
137
+ x = self.SDFT(x, color_style)
138
+
139
+ x = x + x1_s #ResBlock
140
+
141
+ return x
142
+
143
+
144
+ class ColorEncoder(nn.Module):
145
+ def __init__(self, color_dim=512):
146
+ super(ColorEncoder, self).__init__()
147
+
148
+ # self.vgg = vgg19(pretrained_path=None)
149
+ self.vgg = vgg19()
150
+
151
+ self.feature2vector = nn.Sequential(
152
+ nn.Conv2d(color_dim, color_dim, 4, 2, 2), # 8x8
153
+ nn.LeakyReLU(0.2, True),
154
+ nn.Conv2d(color_dim, color_dim, 3, 1, 1),
155
+ nn.LeakyReLU(0.2, True),
156
+ nn.Conv2d(color_dim, color_dim, 4, 2, 2), # 4x4
157
+ nn.LeakyReLU(0.2, True),
158
+ nn.Conv2d(color_dim, color_dim, 3, 1, 1),
159
+ nn.LeakyReLU(0.2, True),
160
+ nn.AdaptiveAvgPool2d((1, 1)), # 1x1
161
+ nn.Conv2d(color_dim, color_dim//2, 1), # linear-1
162
+ nn.LeakyReLU(0.2, True),
163
+ nn.Conv2d(color_dim//2, color_dim//2, 1), # linear-2
164
+ nn.LeakyReLU(0.2, True),
165
+ nn.Conv2d(color_dim//2, color_dim, 1), # linear-3
166
+ )
167
+
168
+ self.color_dim = color_dim
169
+
170
+ def forward(self, x):
171
+ # x #[0, 1] RGB
172
+ vgg_fea = self.vgg(x, layer_name='relu5_2') # [B, 512, 16, 16]
173
+
174
+ x_color = self.feature2vector(vgg_fea[-1]) # [B, 512, 1, 1]
175
+
176
+ return x_color
177
+
178
+
179
+ class ColorUNet(nn.Module):
180
+ ### this model output is ab
181
+ def __init__(self, n_channels=1, n_classes=3, bilinear=True):
182
+ super(ColorUNet, self).__init__()
183
+ self.n_channels = n_channels
184
+ self.n_classes = n_classes
185
+ self.bilinear = bilinear
186
+
187
+ self.inc = DoubleConv(n_channels, 64)
188
+ self.down1 = Down(64, 128)
189
+ self.down2 = Down(128, 256)
190
+ self.down3 = Down(256, 512)
191
+ factor = 2 if bilinear else 1
192
+ self.down4 = Down(512, 1024 // factor)
193
+
194
+ self.up1 = UpBlock(512, 1024, 512 // factor, 3, bilinear)
195
+ self.up2 = UpBlock(512, 512, 256 // factor, 3, bilinear)
196
+ self.up3 = UpBlock(512, 256, 128 // factor, 5, bilinear)
197
+ self.up4 = UpBlock(512, 128, 64, 5, bilinear)
198
+ self.outc = nn.Sequential(
199
+ nn.Conv2d(64, 64, 3, 1, 1),
200
+ nn.LeakyReLU(0.2, True),
201
+ nn.Conv2d(64, 2, 3, 1, 1),
202
+ nn.Tanh() # [-1,1]
203
+ )
204
+
205
+ def forward(self, x):
206
+ # print(torch.max(x[0]), torch.min(x[0])) #[-1, 1] gray image L
207
+ # print(torch.max(x[1]), torch.min(x[1])) # color vector
208
+
209
+ x_color = x[1] # [B, 512, 1, 1]
210
+
211
+ x1 = self.inc(x[0]) # [B, 64, 256, 256]
212
+ x2 = self.down1(x1) # [B, 128, 128, 128]
213
+ x3 = self.down2(x2) # [B, 256, 64, 64]
214
+ x4 = self.down3(x3) # [B, 512, 32, 32]
215
+ x5 = self.down4(x4) # [B, 512, 16, 16]
216
+
217
+ x6 = self.up1(x5, x4, x_color) # [B, 256, 32, 32]
218
+ x7 = self.up2(x6, x3, x_color) # [B, 128, 64, 64]
219
+ x8 = self.up3(x7, x2, x_color) # [B, 64, 128, 128]
220
+ x9 = self.up4(x8, x1, x_color) # [B, 64, 256, 256]
221
+ x_ab = self.outc(x9)
222
+
223
+ return x_ab
real_manga/class1/Color 1659315.jpg ADDED

Git LFS Details

  • SHA256: e36e5429d356c0b7ef109f0aacf584332314a619baba0bf9996fc9e7e338b35f
  • Pointer size: 133 Bytes
  • Size of remote file: 10.2 MB
real_manga/class1/Color 3223141571376159.jpg ADDED

Git LFS Details

  • SHA256: f487b137176cfa45dfceacfd4135fbe96b24e9b67d47b7fd826ae55148dabfbe
  • Pointer size: 132 Bytes
  • Size of remote file: 9.48 MB
real_manga/class1/Color 3486521.jpg ADDED

Git LFS Details

  • SHA256: ba0a92505ee915f6a615fa714f624afc609e8416133a65505d728658b75b0634
  • Pointer size: 133 Bytes
  • Size of remote file: 10.2 MB
real_manga/class1/Color 5102676.jpg ADDED

Git LFS Details

  • SHA256: bbdab465b40ab86b3e0e68d597e336456a3e9921c658e428ebfe543ddc4426ae
  • Pointer size: 132 Bytes
  • Size of remote file: 9.69 MB
real_manga/class1/Color 5570824.jpg ADDED

Git LFS Details

  • SHA256: f56c6d28bd482e9eb58c3f069d99c17f8e8b60c0669de8070d638f5c921547d6
  • Pointer size: 133 Bytes
  • Size of remote file: 10.7 MB
real_manga/class1/Color 5674950.jpg ADDED

Git LFS Details

  • SHA256: 7d75ce71d21323da983c95a24409bf23485484d52d293874eb45ab045376baaa
  • Pointer size: 133 Bytes
  • Size of remote file: 10.3 MB
real_manga/class1/Color 5828407151952509.jpg ADDED

Git LFS Details

  • SHA256: bd598bec0b867950c47415da3d8bbe7c24e0c58bba2df8389e82405032986016
  • Pointer size: 133 Bytes
  • Size of remote file: 15.5 MB