Spaces:
Runtime error
Runtime error
Zhouyan248
commited on
Commit
•
26555ee
1
Parent(s):
10a869a
Upload 86 files
Browse filesThis view is limited to 50 files because it contains too many changes.
See raw diff
- LICENSE +201 -0
- README.md +86 -12
- base/__pycache__/download.cpython-311.pyc +0 -0
- base/app.py +175 -0
- base/app.sh +1 -0
- base/configs/sample.yaml +28 -0
- base/download.py +18 -0
- base/gradio_cached_examples/14/Video result/35727d2ebeb816c94d68/laviea_confused_grizzly_bear_in_cal-1000-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/Video result/42b8a418d77480fcc8fc/laviea_panda_taking_a_selfie_2k_h-400-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/Video result/4b54d32b7e8f2a3cd333/lavieA_steam_train_moving_on_a_moun-230-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/Video result/625fb799abdbcb60fe2f/laviea_corgi_walking_in_the_park_at-400-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/Video result/75b535bd2f78c28d2789/laviea_teddy_bear_walking_on_the_st-100-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/Video result/767a0718deb1983b3d43/laviean_epic_tornado_attacking_abov-230-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/Video result/a15ccccc7c42e18cd062/laviea_jar_filled_with_fire_4K_vid-400-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/Video result/beedae14fa3a8f24e4ec/laviea_shark_swimming_in_clear_Carr-400-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/Video result/c2e7acb8ce5cb0a52899/laviea_teddy_bear_walking_in_the_pa-400-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/Video result/cecd7ff29690b876a418/laviejungle_river_at_sunset_ultra_-400-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/Video result/e67b3c12db1c38afd2c4/laviea_polar_bear_playing_drum_kit_-400-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/Video result/feeee8981f36b962bfe6/laviea_cut_teddy_bear_reading_a_boo-700-50-7-.mp4 +0 -0
- base/gradio_cached_examples/14/log.csv +13 -0
- base/huggingface-t2v/.DS_Store +0 -0
- base/huggingface-t2v/__init__.py +0 -0
- base/huggingface-t2v/requirements.txt +0 -0
- base/models/__init__.py +33 -0
- base/models/__pycache__/__init__.cpython-311.pyc +0 -0
- base/models/__pycache__/attention.cpython-311.pyc +0 -0
- base/models/__pycache__/resnet.cpython-311.pyc +0 -0
- base/models/__pycache__/unet.cpython-311.pyc +0 -0
- base/models/__pycache__/unet_blocks.cpython-311.pyc +0 -0
- base/models/attention.py +707 -0
- base/models/clip.py +120 -0
- base/models/resnet.py +212 -0
- base/models/temporal_attention.py +388 -0
- base/models/transformer_3d.py +367 -0
- base/models/unet.py +617 -0
- base/models/unet_blocks.py +648 -0
- base/models/utils.py +215 -0
- base/pipelines/__pycache__/pipeline_videogen.cpython-311.pyc +0 -0
- base/pipelines/pipeline_videogen.py +677 -0
- base/pipelines/sample.py +88 -0
- base/pipelines/sample.sh +2 -0
- base/text_to_video/__init__.py +44 -0
- base/text_to_video/__pycache__/__init__.cpython-311.pyc +0 -0
- base/try.py +5 -0
- environment.yml +27 -0
- interpolation/configs/sample.yaml +36 -0
- interpolation/datasets/__init__.py +1 -0
- interpolation/datasets/video_transforms.py +109 -0
- interpolation/diffusion/__init__.py +47 -0
- interpolation/diffusion/diffusion_utils.py +88 -0
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
CHANGED
@@ -1,12 +1,86 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# LaVie: High-Quality Video Generation with Cascaded Latent Diffusion Models
|
2 |
+
|
3 |
+
This repository is the official PyTorch implementation of [LaVie](https://arxiv.org/abs/2309.15103).
|
4 |
+
|
5 |
+
**LaVie** is a Text-to-Video (T2V) generation framework, and main part of video generation system [Vchitect](http://vchitect.intern-ai.org.cn/).
|
6 |
+
|
7 |
+
[![arXiv](https://img.shields.io/badge/arXiv-2307.04725-b31b1b.svg)](https://arxiv.org/abs/2309.15103)
|
8 |
+
[![Project Page](https://img.shields.io/badge/Project-Website-green)](https://vchitect.github.io/LaVie-project/)
|
9 |
+
<!--
|
10 |
+
[![Open in OpenXLab](https://cdn-static.openxlab.org.cn/app-center/openxlab_app.svg)]()
|
11 |
+
[![Hugging Face Spaces](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-yellow)]()
|
12 |
+
-->
|
13 |
+
|
14 |
+
<img src="lavie.gif" width="800">
|
15 |
+
|
16 |
+
## Installation
|
17 |
+
```
|
18 |
+
conda env create -f environment.yml
|
19 |
+
conda activate lavie
|
20 |
+
```
|
21 |
+
|
22 |
+
## Download Pre-Trained models
|
23 |
+
Download [pre-trained models](https://huggingface.co/YaohuiW/LaVie/tree/main), [stable diffusion 1.4](https://huggingface.co/CompVis/stable-diffusion-v1-4/tree/main), [stable-diffusion-x4-upscaler](https://huggingface.co/stabilityai/stable-diffusion-x4-upscaler/tree/main) to `./pretrained_models`. You should be able to see the following:
|
24 |
+
```
|
25 |
+
├── pretrained_models
|
26 |
+
│ ├── lavie_base.pt
|
27 |
+
│ ├── lavie_interpolation.pt
|
28 |
+
│ ├── lavie_vsr.pt
|
29 |
+
│ ├── stable-diffusion-v1-4
|
30 |
+
│ │ ├── ...
|
31 |
+
└── └── stable-diffusion-x4-upscaler
|
32 |
+
├── ...
|
33 |
+
```
|
34 |
+
|
35 |
+
## Inference
|
36 |
+
The inference contains **Base T2V**, **Video Interpolation** and **Video Super-Resolution** three steps. We provide several options to generate videos:
|
37 |
+
* **Step1**: 320 x 512 resolution, 16 frames
|
38 |
+
* **Step1+Step2**: 320 x 512 resolution, 61 frames
|
39 |
+
* **Step1+Step3**: 1280 x 2048 resolution, 16 frames
|
40 |
+
* **Step1+Step2+Step3**: 1280 x 2048 resolution, 61 frames
|
41 |
+
|
42 |
+
Feel free to try different options:)
|
43 |
+
|
44 |
+
|
45 |
+
### Step1. Base T2V
|
46 |
+
Run following command to generate videos from base T2V model.
|
47 |
+
```
|
48 |
+
cd base
|
49 |
+
python pipelines/sample.py --config configs/sample.yaml
|
50 |
+
```
|
51 |
+
Edit `text_prompt` in `configs/sample.yaml` to change prompt, results will be saved under `./res/base`.
|
52 |
+
|
53 |
+
### Step2 (optional). Video Interpolation
|
54 |
+
Run following command to conduct video interpolation.
|
55 |
+
```
|
56 |
+
cd interpolation
|
57 |
+
python sample.py --config configs/sample.yaml
|
58 |
+
```
|
59 |
+
The default input video path is `./res/base`, results will be saved under `./res/interpolation`. In `configs/sample.yaml`, you could modify default `input_folder` with `YOUR_INPUT_FOLDER` in `configs/sample.yaml`. Input videos should be named as `prompt1.mp4`, `prompt2.mp4`, ... and put under `YOUR_INPUT_FOLDER`. Launching the code will process all the input videos in `input_folder`.
|
60 |
+
|
61 |
+
|
62 |
+
### Step3 (optional). Video Super-Resolution
|
63 |
+
Run following command to conduct video super-resolution.
|
64 |
+
```
|
65 |
+
cd vsr
|
66 |
+
python sample.py --config configs/sample.yaml
|
67 |
+
```
|
68 |
+
The default input video path is `./res/base` and results will be saved under `./res/vsr`. You could modify default `input_path` with `YOUR_INPUT_FOLDER` in `configs/sample.yaml`. Smiliar to Step2, input videos should be named as `prompt1.mp4`, `prompt2.mp4`, ... and put under `YOUR_INPUT_FOLDER`. Launching the code will process all the input videos in `input_folder`.
|
69 |
+
|
70 |
+
|
71 |
+
## BibTex
|
72 |
+
```bibtex
|
73 |
+
@article{wang2023lavie,
|
74 |
+
title={LAVIE: High-Quality Video Generation with Cascaded Latent Diffusion Models},
|
75 |
+
author={Wang, Yaohui and Chen, Xinyuan and Ma, Xin and Zhou, Shangchen and Huang, Ziqi and Wang, Yi and Yang, Ceyuan and He, Yinan and Yu, Jiashuo and Yang, Peiqing and others},
|
76 |
+
journal={arXiv preprint arXiv:2309.15103},
|
77 |
+
year={2023}
|
78 |
+
}
|
79 |
+
```
|
80 |
+
|
81 |
+
## Acknowledgements
|
82 |
+
The code is buit upon [diffusers](https://github.com/huggingface/diffusers) and [Stable Diffusion](https://github.com/CompVis/stable-diffusion), we thank all the contributors for open-sourcing.
|
83 |
+
|
84 |
+
|
85 |
+
## License
|
86 |
+
The code is licensed under Apache-2.0, model weights are fully open for academic research and also allow **free** commercial usage. To apply for a commercial license, please fill in the [application form]().
|
base/__pycache__/download.cpython-311.pyc
ADDED
Binary file (815 Bytes). View file
|
|
base/app.py
ADDED
@@ -0,0 +1,175 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from text_to_video import model_t2v_fun,setup_seed
|
3 |
+
from omegaconf import OmegaConf
|
4 |
+
import torch
|
5 |
+
import imageio
|
6 |
+
import os
|
7 |
+
import cv2
|
8 |
+
import pandas as pd
|
9 |
+
import torchvision
|
10 |
+
import random
|
11 |
+
config_path = "/mnt/petrelfs/zhouyan/project/lavie-release/base/configs/sample.yaml"
|
12 |
+
args = OmegaConf.load("/mnt/petrelfs/zhouyan/project/lavie-release/base/configs/sample.yaml")
|
13 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
14 |
+
# ------- get model ---------------
|
15 |
+
model_t2V = model_t2v_fun(args)
|
16 |
+
model_t2V.to(device)
|
17 |
+
if device == "cuda":
|
18 |
+
model_t2V.enable_xformers_memory_efficient_attention()
|
19 |
+
|
20 |
+
# model_t2V.enable_xformers_memory_efficient_attention()
|
21 |
+
css = """
|
22 |
+
h1 {
|
23 |
+
text-align: center;
|
24 |
+
}
|
25 |
+
#component-0 {
|
26 |
+
max-width: 730px;
|
27 |
+
margin: auto;
|
28 |
+
}
|
29 |
+
"""
|
30 |
+
|
31 |
+
def infer(prompt, seed_inp, ddim_steps,cfg):
|
32 |
+
if seed_inp!=-1:
|
33 |
+
setup_seed(seed_inp)
|
34 |
+
else:
|
35 |
+
seed_inp = random.choice(range(10000000))
|
36 |
+
setup_seed(seed_inp)
|
37 |
+
videos = model_t2V(prompt, video_length=16, height = 320, width= 512, num_inference_steps=ddim_steps, guidance_scale=cfg).video
|
38 |
+
print(videos[0].shape)
|
39 |
+
if not os.path.exists(args.output_folder):
|
40 |
+
os.mkdir(args.output_folder)
|
41 |
+
torchvision.io.write_video(args.output_folder + prompt[0:30].replace(' ', '_') + '-'+str(seed_inp)+'-'+str(ddim_steps)+'-'+str(cfg)+ '-.mp4', videos[0], fps=8)
|
42 |
+
# imageio.mimwrite(args.output_folder + prompt.replace(' ', '_') + '.mp4', videos[0], fps=8)
|
43 |
+
# video = cv2.VideoCapture(args.output_folder + prompt.replace(' ', '_') + '.mp4')
|
44 |
+
# video = imageio.get_reader(args.output_folder + prompt.replace(' ', '_') + '.mp4', 'ffmpeg')
|
45 |
+
|
46 |
+
|
47 |
+
# video = model_t2V(prompt, seed_inp, ddim_steps)
|
48 |
+
|
49 |
+
return args.output_folder + prompt[0:30].replace(' ', '_') + '-'+str(seed_inp)+'-'+str(ddim_steps)+'-'+str(cfg)+ '-.mp4'
|
50 |
+
|
51 |
+
print(1)
|
52 |
+
|
53 |
+
# def clean():
|
54 |
+
# return gr.Image.update(value=None, visible=False), gr.Video.update(value=None)
|
55 |
+
def clean():
|
56 |
+
return gr.Video.update(value=None)
|
57 |
+
|
58 |
+
title = """
|
59 |
+
<div style="text-align: center; max-width: 700px; margin: 0 auto;">
|
60 |
+
<div
|
61 |
+
style="
|
62 |
+
display: inline-flex;
|
63 |
+
align-items: center;
|
64 |
+
gap: 0.8rem;
|
65 |
+
font-size: 1.75rem;
|
66 |
+
"
|
67 |
+
>
|
68 |
+
<h1 style="font-weight: 900; margin-bottom: 7px; margin-top: 5px;">
|
69 |
+
Intern·Vchitect (Text-to-Video)
|
70 |
+
</h1>
|
71 |
+
</div>
|
72 |
+
<p style="margin-bottom: 10px; font-size: 94%">
|
73 |
+
Apply Intern·Vchitect to generate a video
|
74 |
+
</p>
|
75 |
+
</div>
|
76 |
+
"""
|
77 |
+
|
78 |
+
# print(1)
|
79 |
+
with gr.Blocks(css='style.css') as demo:
|
80 |
+
gr.Markdown("<font color=red size=10><center>LaVie: Text-to-Video generation</center></font>")
|
81 |
+
with gr.Column():
|
82 |
+
with gr.Row(elem_id="col-container"):
|
83 |
+
# inputs = [prompt, seed_inp, ddim_steps]
|
84 |
+
# outputs = [video_out]
|
85 |
+
with gr.Column():
|
86 |
+
|
87 |
+
prompt = gr.Textbox(value="a teddy bear walking on the street", label="Prompt", placeholder="enter prompt", show_label=True, elem_id="prompt-in", min_width=200, lines=2)
|
88 |
+
|
89 |
+
ddim_steps = gr.Slider(label='Steps', minimum=50, maximum=300, value=50, step=1)
|
90 |
+
seed_inp = gr.Slider(value=-1,label="seed (for random generation, use -1)",show_label=True,minimum=-1,maximum=2147483647)
|
91 |
+
cfg = gr.Number(label="guidance_scale",value=7)
|
92 |
+
# seed_inp = gr.Slider(label="Seed", minimum=0, maximum=2147483647, step=1, value=400, elem_id="seed-in")
|
93 |
+
|
94 |
+
# with gr.Row():
|
95 |
+
# # control_task = gr.Dropdown(label="Task", choices=["Text-2-video", "Image-2-video"], value="Text-2-video", multiselect=False, elem_id="controltask-in")
|
96 |
+
# ddim_steps = gr.Slider(label='Steps', minimum=50, maximum=300, value=250, step=1)
|
97 |
+
# seed_inp = gr.Slider(label="Seed", minimum=0, maximum=2147483647, step=1, value=123456, elem_id="seed-in")
|
98 |
+
|
99 |
+
# ddim_steps = gr.Slider(label='Steps', minimum=50, maximum=300, value=250, step=1)
|
100 |
+
# ex = gr.Examples(
|
101 |
+
# examples = [['a corgi walking in the park at sunrise, oil painting style',400,50,7],
|
102 |
+
# ['a cut teddy bear reading a book in the park, oil painting style, high quality',700,50,7],
|
103 |
+
# ['an epic tornado attacking above a glowing city at night, the tornado is made of smoke, highly detailed',230,50,7],
|
104 |
+
# ['a jar filled with fire, 4K video, 3D rendered, well-rendered',400,50,7],
|
105 |
+
# ['a teddy bear walking in the park, oil painting style, high quality',400,50,7],
|
106 |
+
# ['a teddy bear walking on the street, 2k, high quality',100,50,7],
|
107 |
+
# ['a panda taking a selfie, 2k, high quality',400,50,7],
|
108 |
+
# ['a polar bear playing drum kit in NYC Times Square, 4k, high resolution',400,50,7],
|
109 |
+
# ['jungle river at sunset, ultra quality',400,50,7],
|
110 |
+
# ['a shark swimming in clear Carribean ocean, 2k, high quality',400,50,7],
|
111 |
+
# ['A steam train moving on a mountainside by Vincent van Gogh',230,50,7],
|
112 |
+
# ['a confused grizzly bear in calculus class',1000,50,7]],
|
113 |
+
# fn = infer,
|
114 |
+
# inputs=[prompt, seed_inp, ddim_steps,cfg],
|
115 |
+
# # outputs=[video_out],
|
116 |
+
# cache_examples=False,
|
117 |
+
# examples_per_page = 6
|
118 |
+
# )
|
119 |
+
# ex.dataset.headers = [""]
|
120 |
+
|
121 |
+
with gr.Column():
|
122 |
+
submit_btn = gr.Button("Generate video")
|
123 |
+
clean_btn = gr.Button("Clean video")
|
124 |
+
# submit_btn = gr.Button("Generate video", size='sm')
|
125 |
+
# video_out = gr.Video(label="Video result", elem_id="video-output", height=320, width=512)
|
126 |
+
video_out = gr.Video(label="Video result", elem_id="video-output")
|
127 |
+
# with gr.Row():
|
128 |
+
# video_out = gr.Video(label="Video result", elem_id="video-output", height=320, width=512)
|
129 |
+
# submit_btn = gr.Button("Generate video", size='sm')
|
130 |
+
|
131 |
+
|
132 |
+
# video_out = gr.Video(label="Video result", elem_id="video-output", height=320, width=512)
|
133 |
+
inputs = [prompt, seed_inp, ddim_steps,cfg]
|
134 |
+
outputs = [video_out]
|
135 |
+
# gr.Examples(
|
136 |
+
# value = [['An astronaut riding a horse',123,50],
|
137 |
+
# ['a panda eating bamboo on a rock',123,50],
|
138 |
+
# ['Spiderman is surfing',123,50]],
|
139 |
+
# label = "example of sampling",
|
140 |
+
# show_label = True,
|
141 |
+
# headers = ['prompt','seed','steps'],
|
142 |
+
# datatype = ['str','number','number'],
|
143 |
+
# row_count=4,
|
144 |
+
# col_count=(3,"fixed")
|
145 |
+
# )
|
146 |
+
ex = gr.Examples(
|
147 |
+
examples = [['a corgi walking in the park at sunrise, oil painting style',400,50,7],
|
148 |
+
['a cut teddy bear reading a book in the park, oil painting style, high quality',700,50,7],
|
149 |
+
['an epic tornado attacking above a glowing city at night, the tornado is made of smoke, highly detailed',230,50,7],
|
150 |
+
['a jar filled with fire, 4K video, 3D rendered, well-rendered',400,50,7],
|
151 |
+
['a teddy bear walking in the park, oil painting style, high quality',400,50,7],
|
152 |
+
['a teddy bear walking on the street, 2k, high quality',100,50,7],
|
153 |
+
['a panda taking a selfie, 2k, high quality',400,50,7],
|
154 |
+
['a polar bear playing drum kit in NYC Times Square, 4k, high resolution',400,50,7],
|
155 |
+
['jungle river at sunset, ultra quality',400,50,7],
|
156 |
+
['a shark swimming in clear Carribean ocean, 2k, high quality',400,50,7],
|
157 |
+
['A steam train moving on a mountainside by Vincent van Gogh',230,50,7],
|
158 |
+
['a confused grizzly bear in calculus class',1000,50,7]],
|
159 |
+
fn = infer,
|
160 |
+
inputs=[prompt, seed_inp, ddim_steps,cfg],
|
161 |
+
outputs=[video_out],
|
162 |
+
cache_examples=True,
|
163 |
+
)
|
164 |
+
ex.dataset.headers = [""]
|
165 |
+
|
166 |
+
# control_task.change(change_task_options, inputs=[control_task], outputs=[canny_opt, hough_opt, normal_opt], queue=False)
|
167 |
+
# submit_btn.click(clean, inputs=[], outputs=[video_out], queue=False)
|
168 |
+
clean_btn.click(clean, inputs=[], outputs=[video_out], queue=False)
|
169 |
+
submit_btn.click(infer, inputs, outputs)
|
170 |
+
# share_button.click(None, [], [], _js=share_js)
|
171 |
+
|
172 |
+
print(2)
|
173 |
+
demo.queue(max_size=12).launch(server_name="0.0.0.0", server_port=7860)
|
174 |
+
|
175 |
+
|
base/app.sh
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
srun -p aigc-video --gres=gpu:1 -n1 -N1 python app.py
|
base/configs/sample.yaml
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# path:
|
2 |
+
output_folder: "/mnt/petrelfs/share_data/zhouyan/gradio/lavie"
|
3 |
+
pretrained_path: "/mnt/petrelfs/zhouyan/models"
|
4 |
+
|
5 |
+
# model config:
|
6 |
+
model: UNet
|
7 |
+
video_length: 16
|
8 |
+
image_size: [320, 512]
|
9 |
+
|
10 |
+
# beta schedule
|
11 |
+
beta_start: 0.0001
|
12 |
+
beta_end: 0.02
|
13 |
+
beta_schedule: "linear"
|
14 |
+
|
15 |
+
# model speedup
|
16 |
+
use_compile: False
|
17 |
+
use_fp16: True
|
18 |
+
|
19 |
+
# sample config:
|
20 |
+
seed: 3
|
21 |
+
run_time: 0
|
22 |
+
guidance_scale: 7.0
|
23 |
+
sample_method: 'ddpm'
|
24 |
+
num_sampling_steps: 250
|
25 |
+
text_prompt: [
|
26 |
+
'a teddy bear walking on the street, high quality, 2k',
|
27 |
+
|
28 |
+
]
|
base/download.py
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# All rights reserved.
|
2 |
+
|
3 |
+
# This source code is licensed under the license found in the
|
4 |
+
# LICENSE file in the root directory of this source tree.
|
5 |
+
|
6 |
+
import torch
|
7 |
+
import os
|
8 |
+
|
9 |
+
|
10 |
+
def find_model(model_name):
|
11 |
+
"""
|
12 |
+
Finds a pre-trained model, downloading it if necessary. Alternatively, loads a model from a local path.
|
13 |
+
"""
|
14 |
+
checkpoint = torch.load(model_name, map_location=lambda storage, loc: storage)
|
15 |
+
if "ema" in checkpoint: # supports checkpoints from train.py
|
16 |
+
print('Ema existing!')
|
17 |
+
checkpoint = checkpoint["ema"]
|
18 |
+
return checkpoint
|
base/gradio_cached_examples/14/Video result/35727d2ebeb816c94d68/laviea_confused_grizzly_bear_in_cal-1000-50-7-.mp4
ADDED
Binary file (332 kB). View file
|
|
base/gradio_cached_examples/14/Video result/42b8a418d77480fcc8fc/laviea_panda_taking_a_selfie_2k_h-400-50-7-.mp4
ADDED
Binary file (288 kB). View file
|
|
base/gradio_cached_examples/14/Video result/4b54d32b7e8f2a3cd333/lavieA_steam_train_moving_on_a_moun-230-50-7-.mp4
ADDED
Binary file (316 kB). View file
|
|
base/gradio_cached_examples/14/Video result/625fb799abdbcb60fe2f/laviea_corgi_walking_in_the_park_at-400-50-7-.mp4
ADDED
Binary file (258 kB). View file
|
|
base/gradio_cached_examples/14/Video result/75b535bd2f78c28d2789/laviea_teddy_bear_walking_on_the_st-100-50-7-.mp4
ADDED
Binary file (352 kB). View file
|
|
base/gradio_cached_examples/14/Video result/767a0718deb1983b3d43/laviean_epic_tornado_attacking_abov-230-50-7-.mp4
ADDED
Binary file (247 kB). View file
|
|
base/gradio_cached_examples/14/Video result/a15ccccc7c42e18cd062/laviea_jar_filled_with_fire_4K_vid-400-50-7-.mp4
ADDED
Binary file (267 kB). View file
|
|
base/gradio_cached_examples/14/Video result/beedae14fa3a8f24e4ec/laviea_shark_swimming_in_clear_Carr-400-50-7-.mp4
ADDED
Binary file (307 kB). View file
|
|
base/gradio_cached_examples/14/Video result/c2e7acb8ce5cb0a52899/laviea_teddy_bear_walking_in_the_pa-400-50-7-.mp4
ADDED
Binary file (338 kB). View file
|
|
base/gradio_cached_examples/14/Video result/cecd7ff29690b876a418/laviejungle_river_at_sunset_ultra_-400-50-7-.mp4
ADDED
Binary file (261 kB). View file
|
|
base/gradio_cached_examples/14/Video result/e67b3c12db1c38afd2c4/laviea_polar_bear_playing_drum_kit_-400-50-7-.mp4
ADDED
Binary file (372 kB). View file
|
|
base/gradio_cached_examples/14/Video result/feeee8981f36b962bfe6/laviea_cut_teddy_bear_reading_a_boo-700-50-7-.mp4
ADDED
Binary file (381 kB). View file
|
|
base/gradio_cached_examples/14/log.csv
ADDED
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
Video result,flag,username,timestamp
|
2 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/625fb799abdbcb60fe2f/laviea_corgi_walking_in_the_park_at-400-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""laviea_corgi_walking_in_the_park_at-400-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:01:13.139609
|
3 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/feeee8981f36b962bfe6/laviea_cut_teddy_bear_reading_a_boo-700-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""laviea_cut_teddy_bear_reading_a_boo-700-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:01:23.543257
|
4 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/767a0718deb1983b3d43/laviean_epic_tornado_attacking_abov-230-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""laviean_epic_tornado_attacking_abov-230-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:01:33.942899
|
5 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/a15ccccc7c42e18cd062/laviea_jar_filled_with_fire_4K_vid-400-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""laviea_jar_filled_with_fire,_4K_vid-400-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:01:44.348969
|
6 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/c2e7acb8ce5cb0a52899/laviea_teddy_bear_walking_in_the_pa-400-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""laviea_teddy_bear_walking_in_the_pa-400-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:01:54.765554
|
7 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/75b535bd2f78c28d2789/laviea_teddy_bear_walking_on_the_st-100-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""laviea_teddy_bear_walking_on_the_st-100-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:02:05.255612
|
8 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/42b8a418d77480fcc8fc/laviea_panda_taking_a_selfie_2k_h-400-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""laviea_panda_taking_a_selfie,_2k,_h-400-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:02:15.694357
|
9 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/e67b3c12db1c38afd2c4/laviea_polar_bear_playing_drum_kit_-400-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""laviea_polar_bear_playing_drum_kit_-400-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:02:26.121546
|
10 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/cecd7ff29690b876a418/laviejungle_river_at_sunset_ultra_-400-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""laviejungle_river_at_sunset,_ultra_-400-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:02:36.540682
|
11 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/beedae14fa3a8f24e4ec/laviea_shark_swimming_in_clear_Carr-400-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""laviea_shark_swimming_in_clear_Carr-400-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:02:46.992686
|
12 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/4b54d32b7e8f2a3cd333/lavieA_steam_train_moving_on_a_moun-230-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""lavieA_steam_train_moving_on_a_moun-230-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:02:57.458758
|
13 |
+
"{""video"":{""path"":""gradio_cached_examples/14/Video result/35727d2ebeb816c94d68/laviea_confused_grizzly_bear_in_cal-1000-50-7-.mp4"",""url"":null,""size"":null,""orig_name"":""laviea_confused_grizzly_bear_in_cal-1000-50-7-.mp4"",""mime_type"":null},""subtitles"":null}",,,2023-11-27 14:03:07.878403
|
base/huggingface-t2v/.DS_Store
ADDED
Binary file (6.15 kB). View file
|
|
base/huggingface-t2v/__init__.py
ADDED
File without changes
|
base/huggingface-t2v/requirements.txt
ADDED
File without changes
|
base/models/__init__.py
ADDED
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import sys
|
3 |
+
sys.path.append(os.path.split(sys.path[0])[0])
|
4 |
+
|
5 |
+
from .unet import UNet3DConditionModel
|
6 |
+
from torch.optim.lr_scheduler import LambdaLR
|
7 |
+
|
8 |
+
def customized_lr_scheduler(optimizer, warmup_steps=5000): # 5000 from u-vit
|
9 |
+
from torch.optim.lr_scheduler import LambdaLR
|
10 |
+
def fn(step):
|
11 |
+
if warmup_steps > 0:
|
12 |
+
return min(step / warmup_steps, 1)
|
13 |
+
else:
|
14 |
+
return 1
|
15 |
+
return LambdaLR(optimizer, fn)
|
16 |
+
|
17 |
+
|
18 |
+
def get_lr_scheduler(optimizer, name, **kwargs):
|
19 |
+
if name == 'warmup':
|
20 |
+
return customized_lr_scheduler(optimizer, **kwargs)
|
21 |
+
elif name == 'cosine':
|
22 |
+
from torch.optim.lr_scheduler import CosineAnnealingLR
|
23 |
+
return CosineAnnealingLR(optimizer, **kwargs)
|
24 |
+
else:
|
25 |
+
raise NotImplementedError(name)
|
26 |
+
|
27 |
+
def get_models(args, sd_path):
|
28 |
+
|
29 |
+
if 'UNet' in args.model:
|
30 |
+
return UNet3DConditionModel.from_pretrained_2d(sd_path, subfolder="unet")
|
31 |
+
else:
|
32 |
+
raise '{} Model Not Supported!'.format(args.model)
|
33 |
+
|
base/models/__pycache__/__init__.cpython-311.pyc
ADDED
Binary file (1.9 kB). View file
|
|
base/models/__pycache__/attention.cpython-311.pyc
ADDED
Binary file (33.7 kB). View file
|
|
base/models/__pycache__/resnet.cpython-311.pyc
ADDED
Binary file (9.76 kB). View file
|
|
base/models/__pycache__/unet.cpython-311.pyc
ADDED
Binary file (27.3 kB). View file
|
|
base/models/__pycache__/unet_blocks.cpython-311.pyc
ADDED
Binary file (20.3 kB). View file
|
|
base/models/attention.py
ADDED
@@ -0,0 +1,707 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention.py
|
2 |
+
import os
|
3 |
+
import sys
|
4 |
+
sys.path.append(os.path.split(sys.path[0])[0])
|
5 |
+
|
6 |
+
from dataclasses import dataclass
|
7 |
+
from typing import Optional
|
8 |
+
|
9 |
+
import math
|
10 |
+
import torch
|
11 |
+
import torch.nn.functional as F
|
12 |
+
from torch import nn
|
13 |
+
|
14 |
+
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
15 |
+
from diffusers.utils import BaseOutput
|
16 |
+
from diffusers.utils.import_utils import is_xformers_available
|
17 |
+
from diffusers.models.attention import FeedForward, AdaLayerNorm
|
18 |
+
from rotary_embedding_torch import RotaryEmbedding
|
19 |
+
from typing import Callable, Optional
|
20 |
+
from einops import rearrange, repeat
|
21 |
+
|
22 |
+
try:
|
23 |
+
from diffusers.models.modeling_utils import ModelMixin
|
24 |
+
except:
|
25 |
+
from diffusers.modeling_utils import ModelMixin # 0.11.1
|
26 |
+
|
27 |
+
|
28 |
+
@dataclass
|
29 |
+
class Transformer3DModelOutput(BaseOutput):
|
30 |
+
sample: torch.FloatTensor
|
31 |
+
|
32 |
+
|
33 |
+
if is_xformers_available():
|
34 |
+
import xformers
|
35 |
+
import xformers.ops
|
36 |
+
else:
|
37 |
+
xformers = None
|
38 |
+
|
39 |
+
def exists(x):
|
40 |
+
return x is not None
|
41 |
+
|
42 |
+
|
43 |
+
class CrossAttention(nn.Module):
|
44 |
+
r"""
|
45 |
+
copy from diffuser 0.11.1
|
46 |
+
A cross attention layer.
|
47 |
+
Parameters:
|
48 |
+
query_dim (`int`): The number of channels in the query.
|
49 |
+
cross_attention_dim (`int`, *optional*):
|
50 |
+
The number of channels in the encoder_hidden_states. If not given, defaults to `query_dim`.
|
51 |
+
heads (`int`, *optional*, defaults to 8): The number of heads to use for multi-head attention.
|
52 |
+
dim_head (`int`, *optional*, defaults to 64): The number of channels in each head.
|
53 |
+
dropout (`float`, *optional*, defaults to 0.0): The dropout probability to use.
|
54 |
+
bias (`bool`, *optional*, defaults to False):
|
55 |
+
Set to `True` for the query, key, and value linear layers to contain a bias parameter.
|
56 |
+
"""
|
57 |
+
|
58 |
+
def __init__(
|
59 |
+
self,
|
60 |
+
query_dim: int,
|
61 |
+
cross_attention_dim: Optional[int] = None,
|
62 |
+
heads: int = 8,
|
63 |
+
dim_head: int = 64,
|
64 |
+
dropout: float = 0.0,
|
65 |
+
bias=False,
|
66 |
+
upcast_attention: bool = False,
|
67 |
+
upcast_softmax: bool = False,
|
68 |
+
added_kv_proj_dim: Optional[int] = None,
|
69 |
+
norm_num_groups: Optional[int] = None,
|
70 |
+
use_relative_position: bool = False,
|
71 |
+
):
|
72 |
+
super().__init__()
|
73 |
+
inner_dim = dim_head * heads
|
74 |
+
cross_attention_dim = cross_attention_dim if cross_attention_dim is not None else query_dim
|
75 |
+
self.upcast_attention = upcast_attention
|
76 |
+
self.upcast_softmax = upcast_softmax
|
77 |
+
|
78 |
+
self.scale = dim_head**-0.5
|
79 |
+
|
80 |
+
self.heads = heads
|
81 |
+
self.dim_head = dim_head
|
82 |
+
# for slice_size > 0 the attention score computation
|
83 |
+
# is split across the batch axis to save memory
|
84 |
+
# You can set slice_size with `set_attention_slice`
|
85 |
+
self.sliceable_head_dim = heads
|
86 |
+
self._slice_size = None
|
87 |
+
self._use_memory_efficient_attention_xformers = False
|
88 |
+
self.added_kv_proj_dim = added_kv_proj_dim
|
89 |
+
|
90 |
+
if norm_num_groups is not None:
|
91 |
+
self.group_norm = nn.GroupNorm(num_channels=inner_dim, num_groups=norm_num_groups, eps=1e-5, affine=True)
|
92 |
+
else:
|
93 |
+
self.group_norm = None
|
94 |
+
|
95 |
+
self.to_q = nn.Linear(query_dim, inner_dim, bias=bias)
|
96 |
+
self.to_k = nn.Linear(cross_attention_dim, inner_dim, bias=bias)
|
97 |
+
self.to_v = nn.Linear(cross_attention_dim, inner_dim, bias=bias)
|
98 |
+
|
99 |
+
if self.added_kv_proj_dim is not None:
|
100 |
+
self.add_k_proj = nn.Linear(added_kv_proj_dim, cross_attention_dim)
|
101 |
+
self.add_v_proj = nn.Linear(added_kv_proj_dim, cross_attention_dim)
|
102 |
+
|
103 |
+
self.to_out = nn.ModuleList([])
|
104 |
+
self.to_out.append(nn.Linear(inner_dim, query_dim))
|
105 |
+
self.to_out.append(nn.Dropout(dropout))
|
106 |
+
|
107 |
+
self.use_relative_position = use_relative_position
|
108 |
+
if self.use_relative_position:
|
109 |
+
self.rotary_emb = RotaryEmbedding(min(32, dim_head))
|
110 |
+
|
111 |
+
|
112 |
+
def reshape_heads_to_batch_dim(self, tensor):
|
113 |
+
batch_size, seq_len, dim = tensor.shape
|
114 |
+
head_size = self.heads
|
115 |
+
tensor = tensor.reshape(batch_size, seq_len, head_size, dim // head_size)
|
116 |
+
tensor = tensor.permute(0, 2, 1, 3).reshape(batch_size * head_size, seq_len, dim // head_size)
|
117 |
+
return tensor
|
118 |
+
|
119 |
+
def reshape_batch_dim_to_heads(self, tensor):
|
120 |
+
batch_size, seq_len, dim = tensor.shape
|
121 |
+
head_size = self.heads
|
122 |
+
tensor = tensor.reshape(batch_size // head_size, head_size, seq_len, dim)
|
123 |
+
tensor = tensor.permute(0, 2, 1, 3).reshape(batch_size // head_size, seq_len, dim * head_size)
|
124 |
+
return tensor
|
125 |
+
|
126 |
+
def reshape_for_scores(self, tensor):
|
127 |
+
# split heads and dims
|
128 |
+
# tensor should be [b (h w)] f (d nd)
|
129 |
+
batch_size, seq_len, dim = tensor.shape
|
130 |
+
head_size = self.heads
|
131 |
+
tensor = tensor.reshape(batch_size, seq_len, head_size, dim // head_size)
|
132 |
+
tensor = tensor.permute(0, 2, 1, 3).contiguous()
|
133 |
+
return tensor
|
134 |
+
|
135 |
+
def same_batch_dim_to_heads(self, tensor):
|
136 |
+
batch_size, head_size, seq_len, dim = tensor.shape # [b (h w)] nd f d
|
137 |
+
tensor = tensor.reshape(batch_size, seq_len, dim * head_size)
|
138 |
+
return tensor
|
139 |
+
|
140 |
+
def set_attention_slice(self, slice_size):
|
141 |
+
if slice_size is not None and slice_size > self.sliceable_head_dim:
|
142 |
+
raise ValueError(f"slice_size {slice_size} has to be smaller or equal to {self.sliceable_head_dim}.")
|
143 |
+
|
144 |
+
self._slice_size = slice_size
|
145 |
+
|
146 |
+
def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None, use_image_num=None):
|
147 |
+
batch_size, sequence_length, _ = hidden_states.shape
|
148 |
+
|
149 |
+
encoder_hidden_states = encoder_hidden_states
|
150 |
+
|
151 |
+
if self.group_norm is not None:
|
152 |
+
hidden_states = self.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
|
153 |
+
|
154 |
+
query = self.to_q(hidden_states) # [b (h w)] f (nd * d)
|
155 |
+
|
156 |
+
# print('before reshpape query shape', query.shape)
|
157 |
+
dim = query.shape[-1]
|
158 |
+
if not self.use_relative_position:
|
159 |
+
query = self.reshape_heads_to_batch_dim(query) # [b (h w) nd] f d
|
160 |
+
# print('after reshape query shape', query.shape)
|
161 |
+
|
162 |
+
if self.added_kv_proj_dim is not None:
|
163 |
+
key = self.to_k(hidden_states)
|
164 |
+
value = self.to_v(hidden_states)
|
165 |
+
encoder_hidden_states_key_proj = self.add_k_proj(encoder_hidden_states)
|
166 |
+
encoder_hidden_states_value_proj = self.add_v_proj(encoder_hidden_states)
|
167 |
+
|
168 |
+
key = self.reshape_heads_to_batch_dim(key)
|
169 |
+
value = self.reshape_heads_to_batch_dim(value)
|
170 |
+
encoder_hidden_states_key_proj = self.reshape_heads_to_batch_dim(encoder_hidden_states_key_proj)
|
171 |
+
encoder_hidden_states_value_proj = self.reshape_heads_to_batch_dim(encoder_hidden_states_value_proj)
|
172 |
+
|
173 |
+
key = torch.concat([encoder_hidden_states_key_proj, key], dim=1)
|
174 |
+
value = torch.concat([encoder_hidden_states_value_proj, value], dim=1)
|
175 |
+
else:
|
176 |
+
encoder_hidden_states = encoder_hidden_states if encoder_hidden_states is not None else hidden_states
|
177 |
+
key = self.to_k(encoder_hidden_states)
|
178 |
+
value = self.to_v(encoder_hidden_states)
|
179 |
+
|
180 |
+
if not self.use_relative_position:
|
181 |
+
key = self.reshape_heads_to_batch_dim(key)
|
182 |
+
value = self.reshape_heads_to_batch_dim(value)
|
183 |
+
|
184 |
+
if attention_mask is not None:
|
185 |
+
if attention_mask.shape[-1] != query.shape[1]:
|
186 |
+
target_length = query.shape[1]
|
187 |
+
attention_mask = F.pad(attention_mask, (0, target_length), value=0.0)
|
188 |
+
attention_mask = attention_mask.repeat_interleave(self.heads, dim=0)
|
189 |
+
|
190 |
+
# attention, what we cannot get enough of
|
191 |
+
if self._use_memory_efficient_attention_xformers:
|
192 |
+
hidden_states = self._memory_efficient_attention_xformers(query, key, value, attention_mask)
|
193 |
+
# Some versions of xformers return output in fp32, cast it back to the dtype of the input
|
194 |
+
hidden_states = hidden_states.to(query.dtype)
|
195 |
+
else:
|
196 |
+
if self._slice_size is None or query.shape[0] // self._slice_size == 1:
|
197 |
+
hidden_states = self._attention(query, key, value, attention_mask)
|
198 |
+
else:
|
199 |
+
hidden_states = self._sliced_attention(query, key, value, sequence_length, dim, attention_mask)
|
200 |
+
|
201 |
+
# linear proj
|
202 |
+
hidden_states = self.to_out[0](hidden_states)
|
203 |
+
|
204 |
+
# dropout
|
205 |
+
hidden_states = self.to_out[1](hidden_states)
|
206 |
+
return hidden_states
|
207 |
+
|
208 |
+
|
209 |
+
def _attention(self, query, key, value, attention_mask=None):
|
210 |
+
if self.upcast_attention:
|
211 |
+
query = query.float()
|
212 |
+
key = key.float()
|
213 |
+
|
214 |
+
attention_scores = torch.baddbmm(
|
215 |
+
torch.empty(query.shape[0], query.shape[1], key.shape[1], dtype=query.dtype, device=query.device),
|
216 |
+
query,
|
217 |
+
key.transpose(-1, -2),
|
218 |
+
beta=0,
|
219 |
+
alpha=self.scale,
|
220 |
+
)
|
221 |
+
|
222 |
+
if attention_mask is not None:
|
223 |
+
attention_scores = attention_scores + attention_mask
|
224 |
+
|
225 |
+
if self.upcast_softmax:
|
226 |
+
attention_scores = attention_scores.float()
|
227 |
+
|
228 |
+
attention_probs = attention_scores.softmax(dim=-1)
|
229 |
+
|
230 |
+
# cast back to the original dtype
|
231 |
+
attention_probs = attention_probs.to(value.dtype)
|
232 |
+
|
233 |
+
# compute attention output
|
234 |
+
hidden_states = torch.bmm(attention_probs, value)
|
235 |
+
|
236 |
+
# reshape hidden_states
|
237 |
+
hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
|
238 |
+
|
239 |
+
return hidden_states
|
240 |
+
|
241 |
+
def _sliced_attention(self, query, key, value, sequence_length, dim, attention_mask):
|
242 |
+
batch_size_attention = query.shape[0]
|
243 |
+
hidden_states = torch.zeros(
|
244 |
+
(batch_size_attention, sequence_length, dim // self.heads), device=query.device, dtype=query.dtype
|
245 |
+
)
|
246 |
+
slice_size = self._slice_size if self._slice_size is not None else hidden_states.shape[0]
|
247 |
+
for i in range(hidden_states.shape[0] // slice_size):
|
248 |
+
start_idx = i * slice_size
|
249 |
+
end_idx = (i + 1) * slice_size
|
250 |
+
|
251 |
+
query_slice = query[start_idx:end_idx]
|
252 |
+
key_slice = key[start_idx:end_idx]
|
253 |
+
|
254 |
+
if self.upcast_attention:
|
255 |
+
query_slice = query_slice.float()
|
256 |
+
key_slice = key_slice.float()
|
257 |
+
|
258 |
+
attn_slice = torch.baddbmm(
|
259 |
+
torch.empty(slice_size, query.shape[1], key.shape[1], dtype=query_slice.dtype, device=query.device),
|
260 |
+
query_slice,
|
261 |
+
key_slice.transpose(-1, -2),
|
262 |
+
beta=0,
|
263 |
+
alpha=self.scale,
|
264 |
+
)
|
265 |
+
|
266 |
+
if attention_mask is not None:
|
267 |
+
attn_slice = attn_slice + attention_mask[start_idx:end_idx]
|
268 |
+
|
269 |
+
if self.upcast_softmax:
|
270 |
+
attn_slice = attn_slice.float()
|
271 |
+
|
272 |
+
attn_slice = attn_slice.softmax(dim=-1)
|
273 |
+
|
274 |
+
# cast back to the original dtype
|
275 |
+
attn_slice = attn_slice.to(value.dtype)
|
276 |
+
attn_slice = torch.bmm(attn_slice, value[start_idx:end_idx])
|
277 |
+
|
278 |
+
hidden_states[start_idx:end_idx] = attn_slice
|
279 |
+
|
280 |
+
# reshape hidden_states
|
281 |
+
hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
|
282 |
+
return hidden_states
|
283 |
+
|
284 |
+
def _memory_efficient_attention_xformers(self, query, key, value, attention_mask):
|
285 |
+
# TODO attention_mask
|
286 |
+
query = query.contiguous()
|
287 |
+
key = key.contiguous()
|
288 |
+
value = value.contiguous()
|
289 |
+
hidden_states = xformers.ops.memory_efficient_attention(query, key, value, attn_bias=attention_mask)
|
290 |
+
hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
|
291 |
+
return hidden_states
|
292 |
+
|
293 |
+
|
294 |
+
class Transformer3DModel(ModelMixin, ConfigMixin):
|
295 |
+
@register_to_config
|
296 |
+
def __init__(
|
297 |
+
self,
|
298 |
+
num_attention_heads: int = 16,
|
299 |
+
attention_head_dim: int = 88,
|
300 |
+
in_channels: Optional[int] = None,
|
301 |
+
num_layers: int = 1,
|
302 |
+
dropout: float = 0.0,
|
303 |
+
norm_num_groups: int = 32,
|
304 |
+
cross_attention_dim: Optional[int] = None,
|
305 |
+
attention_bias: bool = False,
|
306 |
+
activation_fn: str = "geglu",
|
307 |
+
num_embeds_ada_norm: Optional[int] = None,
|
308 |
+
use_linear_projection: bool = False,
|
309 |
+
only_cross_attention: bool = False,
|
310 |
+
upcast_attention: bool = False,
|
311 |
+
use_first_frame: bool = False,
|
312 |
+
use_relative_position: bool = False,
|
313 |
+
rotary_emb: bool = None,
|
314 |
+
):
|
315 |
+
super().__init__()
|
316 |
+
self.use_linear_projection = use_linear_projection
|
317 |
+
self.num_attention_heads = num_attention_heads
|
318 |
+
self.attention_head_dim = attention_head_dim
|
319 |
+
inner_dim = num_attention_heads * attention_head_dim
|
320 |
+
|
321 |
+
# Define input layers
|
322 |
+
self.in_channels = in_channels
|
323 |
+
|
324 |
+
self.norm = torch.nn.GroupNorm(num_groups=norm_num_groups, num_channels=in_channels, eps=1e-6, affine=True)
|
325 |
+
if use_linear_projection:
|
326 |
+
self.proj_in = nn.Linear(in_channels, inner_dim)
|
327 |
+
else:
|
328 |
+
self.proj_in = nn.Conv2d(in_channels, inner_dim, kernel_size=1, stride=1, padding=0)
|
329 |
+
|
330 |
+
# Define transformers blocks
|
331 |
+
self.transformer_blocks = nn.ModuleList(
|
332 |
+
[
|
333 |
+
BasicTransformerBlock(
|
334 |
+
inner_dim,
|
335 |
+
num_attention_heads,
|
336 |
+
attention_head_dim,
|
337 |
+
dropout=dropout,
|
338 |
+
cross_attention_dim=cross_attention_dim,
|
339 |
+
activation_fn=activation_fn,
|
340 |
+
num_embeds_ada_norm=num_embeds_ada_norm,
|
341 |
+
attention_bias=attention_bias,
|
342 |
+
only_cross_attention=only_cross_attention,
|
343 |
+
upcast_attention=upcast_attention,
|
344 |
+
use_first_frame=use_first_frame,
|
345 |
+
use_relative_position=use_relative_position,
|
346 |
+
rotary_emb=rotary_emb,
|
347 |
+
)
|
348 |
+
for d in range(num_layers)
|
349 |
+
]
|
350 |
+
)
|
351 |
+
|
352 |
+
# 4. Define output layers
|
353 |
+
if use_linear_projection:
|
354 |
+
self.proj_out = nn.Linear(in_channels, inner_dim)
|
355 |
+
else:
|
356 |
+
self.proj_out = nn.Conv2d(inner_dim, in_channels, kernel_size=1, stride=1, padding=0)
|
357 |
+
|
358 |
+
def forward(self, hidden_states, encoder_hidden_states=None, timestep=None, use_image_num=None, return_dict: bool = True):
|
359 |
+
# Input
|
360 |
+
assert hidden_states.dim() == 5, f"Expected hidden_states to have ndim=5, but got ndim={hidden_states.dim()}."
|
361 |
+
|
362 |
+
video_length = hidden_states.shape[2]
|
363 |
+
hidden_states = rearrange(hidden_states, "b c f h w -> (b f) c h w").contiguous()
|
364 |
+
encoder_hidden_states = repeat(encoder_hidden_states, 'b n c -> (b f) n c', f=video_length).contiguous()
|
365 |
+
|
366 |
+
batch, channel, height, weight = hidden_states.shape
|
367 |
+
residual = hidden_states
|
368 |
+
|
369 |
+
hidden_states = self.norm(hidden_states)
|
370 |
+
if not self.use_linear_projection:
|
371 |
+
hidden_states = self.proj_in(hidden_states)
|
372 |
+
inner_dim = hidden_states.shape[1]
|
373 |
+
hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(batch, height * weight, inner_dim)
|
374 |
+
else:
|
375 |
+
inner_dim = hidden_states.shape[1]
|
376 |
+
hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(batch, height * weight, inner_dim)
|
377 |
+
hidden_states = self.proj_in(hidden_states)
|
378 |
+
|
379 |
+
# Blocks
|
380 |
+
for block in self.transformer_blocks:
|
381 |
+
hidden_states = block(
|
382 |
+
hidden_states,
|
383 |
+
encoder_hidden_states=encoder_hidden_states,
|
384 |
+
timestep=timestep,
|
385 |
+
video_length=video_length,
|
386 |
+
use_image_num=use_image_num,
|
387 |
+
)
|
388 |
+
|
389 |
+
# Output
|
390 |
+
if not self.use_linear_projection:
|
391 |
+
hidden_states = (
|
392 |
+
hidden_states.reshape(batch, height, weight, inner_dim).permute(0, 3, 1, 2).contiguous()
|
393 |
+
)
|
394 |
+
hidden_states = self.proj_out(hidden_states)
|
395 |
+
else:
|
396 |
+
hidden_states = self.proj_out(hidden_states)
|
397 |
+
hidden_states = (
|
398 |
+
hidden_states.reshape(batch, height, weight, inner_dim).permute(0, 3, 1, 2).contiguous()
|
399 |
+
)
|
400 |
+
|
401 |
+
output = hidden_states + residual
|
402 |
+
|
403 |
+
output = rearrange(output, "(b f) c h w -> b c f h w", f=video_length + use_image_num).contiguous()
|
404 |
+
if not return_dict:
|
405 |
+
return (output,)
|
406 |
+
|
407 |
+
return Transformer3DModelOutput(sample=output)
|
408 |
+
|
409 |
+
|
410 |
+
class BasicTransformerBlock(nn.Module):
|
411 |
+
def __init__(
|
412 |
+
self,
|
413 |
+
dim: int,
|
414 |
+
num_attention_heads: int,
|
415 |
+
attention_head_dim: int,
|
416 |
+
dropout=0.0,
|
417 |
+
cross_attention_dim: Optional[int] = None,
|
418 |
+
activation_fn: str = "geglu",
|
419 |
+
num_embeds_ada_norm: Optional[int] = None,
|
420 |
+
attention_bias: bool = False,
|
421 |
+
only_cross_attention: bool = False,
|
422 |
+
upcast_attention: bool = False,
|
423 |
+
use_first_frame: bool = False,
|
424 |
+
use_relative_position: bool = False,
|
425 |
+
rotary_emb: bool = False,
|
426 |
+
):
|
427 |
+
super().__init__()
|
428 |
+
self.only_cross_attention = only_cross_attention
|
429 |
+
self.use_ada_layer_norm = num_embeds_ada_norm is not None
|
430 |
+
self.use_first_frame = use_first_frame
|
431 |
+
|
432 |
+
# Spatial-Attn
|
433 |
+
self.attn1 = CrossAttention(
|
434 |
+
query_dim=dim,
|
435 |
+
heads=num_attention_heads,
|
436 |
+
dim_head=attention_head_dim,
|
437 |
+
dropout=dropout,
|
438 |
+
bias=attention_bias,
|
439 |
+
cross_attention_dim=None,
|
440 |
+
upcast_attention=upcast_attention,
|
441 |
+
)
|
442 |
+
self.norm1 = AdaLayerNorm(dim, num_embeds_ada_norm) if self.use_ada_layer_norm else nn.LayerNorm(dim)
|
443 |
+
|
444 |
+
# Text Cross-Attn
|
445 |
+
if cross_attention_dim is not None:
|
446 |
+
self.attn2 = CrossAttention(
|
447 |
+
query_dim=dim,
|
448 |
+
cross_attention_dim=cross_attention_dim,
|
449 |
+
heads=num_attention_heads,
|
450 |
+
dim_head=attention_head_dim,
|
451 |
+
dropout=dropout,
|
452 |
+
bias=attention_bias,
|
453 |
+
upcast_attention=upcast_attention,
|
454 |
+
)
|
455 |
+
else:
|
456 |
+
self.attn2 = None
|
457 |
+
|
458 |
+
if cross_attention_dim is not None:
|
459 |
+
self.norm2 = AdaLayerNorm(dim, num_embeds_ada_norm) if self.use_ada_layer_norm else nn.LayerNorm(dim)
|
460 |
+
else:
|
461 |
+
self.norm2 = None
|
462 |
+
|
463 |
+
# Temp
|
464 |
+
self.attn_temp = TemporalAttention(
|
465 |
+
query_dim=dim,
|
466 |
+
heads=num_attention_heads,
|
467 |
+
dim_head=attention_head_dim,
|
468 |
+
dropout=dropout,
|
469 |
+
bias=attention_bias,
|
470 |
+
cross_attention_dim=None,
|
471 |
+
upcast_attention=upcast_attention,
|
472 |
+
rotary_emb=rotary_emb,
|
473 |
+
)
|
474 |
+
self.norm_temp = AdaLayerNorm(dim, num_embeds_ada_norm) if self.use_ada_layer_norm else nn.LayerNorm(dim)
|
475 |
+
nn.init.zeros_(self.attn_temp.to_out[0].weight.data)
|
476 |
+
|
477 |
+
|
478 |
+
# Feed-forward
|
479 |
+
self.ff = FeedForward(dim, dropout=dropout, activation_fn=activation_fn)
|
480 |
+
self.norm3 = nn.LayerNorm(dim)
|
481 |
+
|
482 |
+
def set_use_memory_efficient_attention_xformers(self, use_memory_efficient_attention_xformers: bool, op=None):
|
483 |
+
|
484 |
+
if not is_xformers_available():
|
485 |
+
print("Here is how to install it")
|
486 |
+
raise ModuleNotFoundError(
|
487 |
+
"Refer to https://github.com/facebookresearch/xformers for more information on how to install"
|
488 |
+
" xformers",
|
489 |
+
name="xformers",
|
490 |
+
)
|
491 |
+
elif not torch.cuda.is_available():
|
492 |
+
raise ValueError(
|
493 |
+
"torch.cuda.is_available() should be True but is False. xformers' memory efficient attention is only"
|
494 |
+
" available for GPU "
|
495 |
+
)
|
496 |
+
else:
|
497 |
+
try:
|
498 |
+
# Make sure we can run the memory efficient attention
|
499 |
+
_ = xformers.ops.memory_efficient_attention(
|
500 |
+
torch.randn((1, 2, 40), device="cuda"),
|
501 |
+
torch.randn((1, 2, 40), device="cuda"),
|
502 |
+
torch.randn((1, 2, 40), device="cuda"),
|
503 |
+
)
|
504 |
+
except Exception as e:
|
505 |
+
raise e
|
506 |
+
self.attn1._use_memory_efficient_attention_xformers = use_memory_efficient_attention_xformers
|
507 |
+
if self.attn2 is not None:
|
508 |
+
self.attn2._use_memory_efficient_attention_xformers = use_memory_efficient_attention_xformers
|
509 |
+
# self.attn_temp._use_memory_efficient_attention_xformers = use_memory_efficient_attention_xformers
|
510 |
+
|
511 |
+
def forward(self, hidden_states, encoder_hidden_states=None, timestep=None, attention_mask=None, video_length=None, use_image_num=None):
|
512 |
+
# SparseCausal-Attention
|
513 |
+
norm_hidden_states = (
|
514 |
+
self.norm1(hidden_states, timestep) if self.use_ada_layer_norm else self.norm1(hidden_states)
|
515 |
+
)
|
516 |
+
|
517 |
+
if self.only_cross_attention:
|
518 |
+
hidden_states = (
|
519 |
+
self.attn1(norm_hidden_states, encoder_hidden_states, attention_mask=attention_mask) + hidden_states
|
520 |
+
)
|
521 |
+
else:
|
522 |
+
hidden_states = self.attn1(norm_hidden_states, attention_mask=attention_mask, use_image_num=use_image_num) + hidden_states
|
523 |
+
|
524 |
+
if self.attn2 is not None:
|
525 |
+
# Cross-Attention
|
526 |
+
norm_hidden_states = (
|
527 |
+
self.norm2(hidden_states, timestep) if self.use_ada_layer_norm else self.norm2(hidden_states)
|
528 |
+
)
|
529 |
+
hidden_states = (
|
530 |
+
self.attn2(
|
531 |
+
norm_hidden_states, encoder_hidden_states=encoder_hidden_states, attention_mask=attention_mask
|
532 |
+
)
|
533 |
+
+ hidden_states
|
534 |
+
)
|
535 |
+
|
536 |
+
# Temporal Attention
|
537 |
+
if self.training:
|
538 |
+
d = hidden_states.shape[1]
|
539 |
+
hidden_states = rearrange(hidden_states, "(b f) d c -> (b d) f c", f=video_length + use_image_num).contiguous()
|
540 |
+
hidden_states_video = hidden_states[:, :video_length, :]
|
541 |
+
hidden_states_image = hidden_states[:, video_length:, :]
|
542 |
+
norm_hidden_states_video = (
|
543 |
+
self.norm_temp(hidden_states_video, timestep) if self.use_ada_layer_norm else self.norm_temp(hidden_states_video)
|
544 |
+
)
|
545 |
+
hidden_states_video = self.attn_temp(norm_hidden_states_video) + hidden_states_video
|
546 |
+
hidden_states = torch.cat([hidden_states_video, hidden_states_image], dim=1)
|
547 |
+
hidden_states = rearrange(hidden_states, "(b d) f c -> (b f) d c", d=d).contiguous()
|
548 |
+
else:
|
549 |
+
d = hidden_states.shape[1]
|
550 |
+
hidden_states = rearrange(hidden_states, "(b f) d c -> (b d) f c", f=video_length + use_image_num).contiguous()
|
551 |
+
norm_hidden_states = (
|
552 |
+
self.norm_temp(hidden_states, timestep) if self.use_ada_layer_norm else self.norm_temp(hidden_states)
|
553 |
+
)
|
554 |
+
hidden_states = self.attn_temp(norm_hidden_states) + hidden_states
|
555 |
+
hidden_states = rearrange(hidden_states, "(b d) f c -> (b f) d c", d=d).contiguous()
|
556 |
+
|
557 |
+
# Feed-forward
|
558 |
+
hidden_states = self.ff(self.norm3(hidden_states)) + hidden_states
|
559 |
+
|
560 |
+
return hidden_states
|
561 |
+
|
562 |
+
class TemporalAttention(CrossAttention):
|
563 |
+
def __init__(self,
|
564 |
+
query_dim: int,
|
565 |
+
cross_attention_dim: Optional[int] = None,
|
566 |
+
heads: int = 8,
|
567 |
+
dim_head: int = 64,
|
568 |
+
dropout: float = 0.0,
|
569 |
+
bias=False,
|
570 |
+
upcast_attention: bool = False,
|
571 |
+
upcast_softmax: bool = False,
|
572 |
+
added_kv_proj_dim: Optional[int] = None,
|
573 |
+
norm_num_groups: Optional[int] = None,
|
574 |
+
rotary_emb=None):
|
575 |
+
super().__init__(query_dim, cross_attention_dim, heads, dim_head, dropout, bias, upcast_attention, upcast_softmax, added_kv_proj_dim, norm_num_groups)
|
576 |
+
# relative time positional embeddings
|
577 |
+
self.time_rel_pos_bias = RelativePositionBias(heads=heads, max_distance=32) # realistically will not be able to generate that many frames of video... yet
|
578 |
+
self.rotary_emb = rotary_emb
|
579 |
+
|
580 |
+
def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None):
|
581 |
+
time_rel_pos_bias = self.time_rel_pos_bias(hidden_states.shape[1], device=hidden_states.device)
|
582 |
+
batch_size, sequence_length, _ = hidden_states.shape
|
583 |
+
|
584 |
+
encoder_hidden_states = encoder_hidden_states
|
585 |
+
|
586 |
+
if self.group_norm is not None:
|
587 |
+
hidden_states = self.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
|
588 |
+
|
589 |
+
query = self.to_q(hidden_states) # [b (h w)] f (nd * d)
|
590 |
+
dim = query.shape[-1]
|
591 |
+
|
592 |
+
if self.added_kv_proj_dim is not None:
|
593 |
+
key = self.to_k(hidden_states)
|
594 |
+
value = self.to_v(hidden_states)
|
595 |
+
encoder_hidden_states_key_proj = self.add_k_proj(encoder_hidden_states)
|
596 |
+
encoder_hidden_states_value_proj = self.add_v_proj(encoder_hidden_states)
|
597 |
+
|
598 |
+
key = self.reshape_heads_to_batch_dim(key)
|
599 |
+
value = self.reshape_heads_to_batch_dim(value)
|
600 |
+
encoder_hidden_states_key_proj = self.reshape_heads_to_batch_dim(encoder_hidden_states_key_proj)
|
601 |
+
encoder_hidden_states_value_proj = self.reshape_heads_to_batch_dim(encoder_hidden_states_value_proj)
|
602 |
+
|
603 |
+
key = torch.concat([encoder_hidden_states_key_proj, key], dim=1)
|
604 |
+
value = torch.concat([encoder_hidden_states_value_proj, value], dim=1)
|
605 |
+
else:
|
606 |
+
encoder_hidden_states = encoder_hidden_states if encoder_hidden_states is not None else hidden_states
|
607 |
+
key = self.to_k(encoder_hidden_states)
|
608 |
+
value = self.to_v(encoder_hidden_states)
|
609 |
+
|
610 |
+
if attention_mask is not None:
|
611 |
+
if attention_mask.shape[-1] != query.shape[1]:
|
612 |
+
target_length = query.shape[1]
|
613 |
+
attention_mask = F.pad(attention_mask, (0, target_length), value=0.0)
|
614 |
+
attention_mask = attention_mask.repeat_interleave(self.heads, dim=0)
|
615 |
+
|
616 |
+
# attention, what we cannot get enough of
|
617 |
+
if self._use_memory_efficient_attention_xformers:
|
618 |
+
hidden_states = self._memory_efficient_attention_xformers(query, key, value, attention_mask)
|
619 |
+
# Some versions of xformers return output in fp32, cast it back to the dtype of the input
|
620 |
+
hidden_states = hidden_states.to(query.dtype)
|
621 |
+
else:
|
622 |
+
if self._slice_size is None or query.shape[0] // self._slice_size == 1:
|
623 |
+
hidden_states = self._attention(query, key, value, attention_mask, time_rel_pos_bias)
|
624 |
+
else:
|
625 |
+
hidden_states = self._sliced_attention(query, key, value, sequence_length, dim, attention_mask)
|
626 |
+
|
627 |
+
# linear proj
|
628 |
+
hidden_states = self.to_out[0](hidden_states)
|
629 |
+
|
630 |
+
# dropout
|
631 |
+
hidden_states = self.to_out[1](hidden_states)
|
632 |
+
return hidden_states
|
633 |
+
|
634 |
+
def _attention(self, query, key, value, attention_mask=None, time_rel_pos_bias=None):
|
635 |
+
if self.upcast_attention:
|
636 |
+
query = query.float()
|
637 |
+
key = key.float()
|
638 |
+
|
639 |
+
# reshape for adding time positional bais
|
640 |
+
query = self.scale * rearrange(query, 'b f (h d) -> b h f d', h=self.heads) # d: dim_head; n: heads
|
641 |
+
key = rearrange(key, 'b f (h d) -> b h f d', h=self.heads) # d: dim_head; n: heads
|
642 |
+
value = rearrange(value, 'b f (h d) -> b h f d', h=self.heads) # d: dim_head; n: heads
|
643 |
+
|
644 |
+
if exists(self.rotary_emb):
|
645 |
+
query = self.rotary_emb.rotate_queries_or_keys(query)
|
646 |
+
key = self.rotary_emb.rotate_queries_or_keys(key)
|
647 |
+
|
648 |
+
attention_scores = torch.einsum('... h i d, ... h j d -> ... h i j', query, key)
|
649 |
+
|
650 |
+
attention_scores = attention_scores + time_rel_pos_bias
|
651 |
+
|
652 |
+
if attention_mask is not None:
|
653 |
+
# add attention mask
|
654 |
+
attention_scores = attention_scores + attention_mask
|
655 |
+
|
656 |
+
attention_scores = attention_scores - attention_scores.amax(dim = -1, keepdim = True).detach()
|
657 |
+
|
658 |
+
attention_probs = nn.functional.softmax(attention_scores, dim=-1)
|
659 |
+
# print(attention_probs[0][0])
|
660 |
+
|
661 |
+
# cast back to the original dtype
|
662 |
+
attention_probs = attention_probs.to(value.dtype)
|
663 |
+
|
664 |
+
# compute attention output
|
665 |
+
hidden_states = torch.einsum('... h i j, ... h j d -> ... h i d', attention_probs, value)
|
666 |
+
hidden_states = rearrange(hidden_states, 'b h f d -> b f (h d)')
|
667 |
+
return hidden_states
|
668 |
+
|
669 |
+
class RelativePositionBias(nn.Module):
|
670 |
+
def __init__(
|
671 |
+
self,
|
672 |
+
heads=8,
|
673 |
+
num_buckets=32,
|
674 |
+
max_distance=128,
|
675 |
+
):
|
676 |
+
super().__init__()
|
677 |
+
self.num_buckets = num_buckets
|
678 |
+
self.max_distance = max_distance
|
679 |
+
self.relative_attention_bias = nn.Embedding(num_buckets, heads)
|
680 |
+
|
681 |
+
@staticmethod
|
682 |
+
def _relative_position_bucket(relative_position, num_buckets=32, max_distance=128):
|
683 |
+
ret = 0
|
684 |
+
n = -relative_position
|
685 |
+
|
686 |
+
num_buckets //= 2
|
687 |
+
ret += (n < 0).long() * num_buckets
|
688 |
+
n = torch.abs(n)
|
689 |
+
|
690 |
+
max_exact = num_buckets // 2
|
691 |
+
is_small = n < max_exact
|
692 |
+
|
693 |
+
val_if_large = max_exact + (
|
694 |
+
torch.log(n.float() / max_exact) / math.log(max_distance / max_exact) * (num_buckets - max_exact)
|
695 |
+
).long()
|
696 |
+
val_if_large = torch.min(val_if_large, torch.full_like(val_if_large, num_buckets - 1))
|
697 |
+
|
698 |
+
ret += torch.where(is_small, n, val_if_large)
|
699 |
+
return ret
|
700 |
+
|
701 |
+
def forward(self, n, device):
|
702 |
+
q_pos = torch.arange(n, dtype = torch.long, device = device)
|
703 |
+
k_pos = torch.arange(n, dtype = torch.long, device = device)
|
704 |
+
rel_pos = rearrange(k_pos, 'j -> 1 j') - rearrange(q_pos, 'i -> i 1')
|
705 |
+
rp_bucket = self._relative_position_bucket(rel_pos, num_buckets = self.num_buckets, max_distance = self.max_distance)
|
706 |
+
values = self.relative_attention_bias(rp_bucket)
|
707 |
+
return rearrange(values, 'i j h -> h i j') # num_heads, num_frames, num_frames
|
base/models/clip.py
ADDED
@@ -0,0 +1,120 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy
|
2 |
+
import torch.nn as nn
|
3 |
+
from transformers import CLIPTokenizer, CLIPTextModel
|
4 |
+
|
5 |
+
import transformers
|
6 |
+
transformers.logging.set_verbosity_error()
|
7 |
+
|
8 |
+
"""
|
9 |
+
Will encounter following warning:
|
10 |
+
- This IS expected if you are initializing CLIPTextModel from the checkpoint of a model trained on another task
|
11 |
+
or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
|
12 |
+
- This IS NOT expected if you are initializing CLIPTextModel from the checkpoint of a model
|
13 |
+
that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
|
14 |
+
|
15 |
+
https://github.com/CompVis/stable-diffusion/issues/97
|
16 |
+
according to this issue, this warning is safe.
|
17 |
+
|
18 |
+
This is expected since the vision backbone of the CLIP model is not needed to run Stable Diffusion.
|
19 |
+
You can safely ignore the warning, it is not an error.
|
20 |
+
|
21 |
+
This clip usage is from U-ViT and same with Stable Diffusion.
|
22 |
+
"""
|
23 |
+
|
24 |
+
class AbstractEncoder(nn.Module):
|
25 |
+
def __init__(self):
|
26 |
+
super().__init__()
|
27 |
+
|
28 |
+
def encode(self, *args, **kwargs):
|
29 |
+
raise NotImplementedError
|
30 |
+
|
31 |
+
|
32 |
+
class FrozenCLIPEmbedder(AbstractEncoder):
|
33 |
+
"""Uses the CLIP transformer encoder for text (from Hugging Face)"""
|
34 |
+
# def __init__(self, version="openai/clip-vit-huge-patch14", device="cuda", max_length=77):
|
35 |
+
def __init__(self, path, device="cuda", max_length=77):
|
36 |
+
super().__init__()
|
37 |
+
self.tokenizer = CLIPTokenizer.from_pretrained(path, subfolder="tokenizer")
|
38 |
+
self.transformer = CLIPTextModel.from_pretrained(path, subfolder='text_encoder')
|
39 |
+
self.device = device
|
40 |
+
self.max_length = max_length
|
41 |
+
self.freeze()
|
42 |
+
|
43 |
+
def freeze(self):
|
44 |
+
self.transformer = self.transformer.eval()
|
45 |
+
for param in self.parameters():
|
46 |
+
param.requires_grad = False
|
47 |
+
|
48 |
+
def forward(self, text):
|
49 |
+
batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
|
50 |
+
return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
|
51 |
+
tokens = batch_encoding["input_ids"].to(self.device)
|
52 |
+
outputs = self.transformer(input_ids=tokens)
|
53 |
+
|
54 |
+
z = outputs.last_hidden_state
|
55 |
+
return z
|
56 |
+
|
57 |
+
def encode(self, text):
|
58 |
+
return self(text)
|
59 |
+
|
60 |
+
|
61 |
+
class TextEmbedder(nn.Module):
|
62 |
+
"""
|
63 |
+
Embeds text prompt into vector representations. Also handles text dropout for classifier-free guidance.
|
64 |
+
"""
|
65 |
+
def __init__(self, path, dropout_prob=0.1):
|
66 |
+
super().__init__()
|
67 |
+
self.text_encodder = FrozenCLIPEmbedder(path=path)
|
68 |
+
self.dropout_prob = dropout_prob
|
69 |
+
|
70 |
+
def token_drop(self, text_prompts, force_drop_ids=None):
|
71 |
+
"""
|
72 |
+
Drops text to enable classifier-free guidance.
|
73 |
+
"""
|
74 |
+
if force_drop_ids is None:
|
75 |
+
drop_ids = numpy.random.uniform(0, 1, len(text_prompts)) < self.dropout_prob
|
76 |
+
else:
|
77 |
+
# TODO
|
78 |
+
drop_ids = force_drop_ids == 1
|
79 |
+
labels = list(numpy.where(drop_ids, "", text_prompts))
|
80 |
+
# print(labels)
|
81 |
+
return labels
|
82 |
+
|
83 |
+
def forward(self, text_prompts, train, force_drop_ids=None):
|
84 |
+
use_dropout = self.dropout_prob > 0
|
85 |
+
if (train and use_dropout) or (force_drop_ids is not None):
|
86 |
+
text_prompts = self.token_drop(text_prompts, force_drop_ids)
|
87 |
+
embeddings = self.text_encodder(text_prompts)
|
88 |
+
return embeddings
|
89 |
+
|
90 |
+
|
91 |
+
if __name__ == '__main__':
|
92 |
+
|
93 |
+
r"""
|
94 |
+
Returns:
|
95 |
+
|
96 |
+
Examples from CLIPTextModel:
|
97 |
+
|
98 |
+
```python
|
99 |
+
>>> from transformers import AutoTokenizer, CLIPTextModel
|
100 |
+
|
101 |
+
>>> model = CLIPTextModel.from_pretrained("openai/clip-vit-base-patch32")
|
102 |
+
>>> tokenizer = AutoTokenizer.from_pretrained("openai/clip-vit-base-patch32")
|
103 |
+
|
104 |
+
>>> inputs = tokenizer(["a photo of a cat", "a photo of a dog"], padding=True, return_tensors="pt")
|
105 |
+
|
106 |
+
>>> outputs = model(**inputs)
|
107 |
+
>>> last_hidden_state = outputs.last_hidden_state
|
108 |
+
>>> pooled_output = outputs.pooler_output # pooled (EOS token) states
|
109 |
+
```"""
|
110 |
+
|
111 |
+
import torch
|
112 |
+
|
113 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
114 |
+
|
115 |
+
text_encoder = TextEmbedder(path='/mnt/petrelfs/maxin/work/pretrained/stable-diffusion-2-1-base',
|
116 |
+
dropout_prob=0.00001).to(device)
|
117 |
+
|
118 |
+
text_prompt = [["a photo of a cat", "a photo of a cat"], ["a photo of a dog", "a photo of a cat"], ['a photo of a dog human', "a photo of a cat"]]
|
119 |
+
output = text_encoder(text_prompts=text_prompt, train=False)
|
120 |
+
print(output.shape)
|
base/models/resnet.py
ADDED
@@ -0,0 +1,212 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/resnet.py
|
2 |
+
import os
|
3 |
+
import sys
|
4 |
+
sys.path.append(os.path.split(sys.path[0])[0])
|
5 |
+
|
6 |
+
import torch
|
7 |
+
import torch.nn as nn
|
8 |
+
import torch.nn.functional as F
|
9 |
+
|
10 |
+
from einops import rearrange
|
11 |
+
|
12 |
+
|
13 |
+
class InflatedConv3d(nn.Conv2d):
|
14 |
+
def forward(self, x):
|
15 |
+
video_length = x.shape[2]
|
16 |
+
|
17 |
+
x = rearrange(x, "b c f h w -> (b f) c h w")
|
18 |
+
x = super().forward(x)
|
19 |
+
x = rearrange(x, "(b f) c h w -> b c f h w", f=video_length)
|
20 |
+
|
21 |
+
return x
|
22 |
+
|
23 |
+
|
24 |
+
class Upsample3D(nn.Module):
|
25 |
+
def __init__(self, channels, use_conv=False, use_conv_transpose=False, out_channels=None, name="conv"):
|
26 |
+
super().__init__()
|
27 |
+
self.channels = channels
|
28 |
+
self.out_channels = out_channels or channels
|
29 |
+
self.use_conv = use_conv
|
30 |
+
self.use_conv_transpose = use_conv_transpose
|
31 |
+
self.name = name
|
32 |
+
|
33 |
+
conv = None
|
34 |
+
if use_conv_transpose:
|
35 |
+
raise NotImplementedError
|
36 |
+
elif use_conv:
|
37 |
+
conv = InflatedConv3d(self.channels, self.out_channels, 3, padding=1)
|
38 |
+
|
39 |
+
if name == "conv":
|
40 |
+
self.conv = conv
|
41 |
+
else:
|
42 |
+
self.Conv2d_0 = conv
|
43 |
+
|
44 |
+
def forward(self, hidden_states, output_size=None):
|
45 |
+
assert hidden_states.shape[1] == self.channels
|
46 |
+
|
47 |
+
if self.use_conv_transpose:
|
48 |
+
raise NotImplementedError
|
49 |
+
|
50 |
+
# Cast to float32 to as 'upsample_nearest2d_out_frame' op does not support bfloat16
|
51 |
+
dtype = hidden_states.dtype
|
52 |
+
if dtype == torch.bfloat16:
|
53 |
+
hidden_states = hidden_states.to(torch.float32)
|
54 |
+
|
55 |
+
# upsample_nearest_nhwc fails with large batch sizes. see https://github.com/huggingface/diffusers/issues/984
|
56 |
+
if hidden_states.shape[0] >= 64:
|
57 |
+
hidden_states = hidden_states.contiguous()
|
58 |
+
|
59 |
+
# if `output_size` is passed we force the interpolation output
|
60 |
+
# size and do not make use of `scale_factor=2`
|
61 |
+
if output_size is None:
|
62 |
+
hidden_states = F.interpolate(hidden_states, scale_factor=[1.0, 2.0, 2.0], mode="nearest")
|
63 |
+
else:
|
64 |
+
hidden_states = F.interpolate(hidden_states, size=output_size, mode="nearest")
|
65 |
+
|
66 |
+
# If the input is bfloat16, we cast back to bfloat16
|
67 |
+
if dtype == torch.bfloat16:
|
68 |
+
hidden_states = hidden_states.to(dtype)
|
69 |
+
|
70 |
+
if self.use_conv:
|
71 |
+
if self.name == "conv":
|
72 |
+
hidden_states = self.conv(hidden_states)
|
73 |
+
else:
|
74 |
+
hidden_states = self.Conv2d_0(hidden_states)
|
75 |
+
|
76 |
+
return hidden_states
|
77 |
+
|
78 |
+
|
79 |
+
class Downsample3D(nn.Module):
|
80 |
+
def __init__(self, channels, use_conv=False, out_channels=None, padding=1, name="conv"):
|
81 |
+
super().__init__()
|
82 |
+
self.channels = channels
|
83 |
+
self.out_channels = out_channels or channels
|
84 |
+
self.use_conv = use_conv
|
85 |
+
self.padding = padding
|
86 |
+
stride = 2
|
87 |
+
self.name = name
|
88 |
+
|
89 |
+
if use_conv:
|
90 |
+
conv = InflatedConv3d(self.channels, self.out_channels, 3, stride=stride, padding=padding)
|
91 |
+
else:
|
92 |
+
raise NotImplementedError
|
93 |
+
|
94 |
+
if name == "conv":
|
95 |
+
self.Conv2d_0 = conv
|
96 |
+
self.conv = conv
|
97 |
+
elif name == "Conv2d_0":
|
98 |
+
self.conv = conv
|
99 |
+
else:
|
100 |
+
self.conv = conv
|
101 |
+
|
102 |
+
def forward(self, hidden_states):
|
103 |
+
assert hidden_states.shape[1] == self.channels
|
104 |
+
if self.use_conv and self.padding == 0:
|
105 |
+
raise NotImplementedError
|
106 |
+
|
107 |
+
assert hidden_states.shape[1] == self.channels
|
108 |
+
hidden_states = self.conv(hidden_states)
|
109 |
+
|
110 |
+
return hidden_states
|
111 |
+
|
112 |
+
|
113 |
+
class ResnetBlock3D(nn.Module):
|
114 |
+
def __init__(
|
115 |
+
self,
|
116 |
+
*,
|
117 |
+
in_channels,
|
118 |
+
out_channels=None,
|
119 |
+
conv_shortcut=False,
|
120 |
+
dropout=0.0,
|
121 |
+
temb_channels=512,
|
122 |
+
groups=32,
|
123 |
+
groups_out=None,
|
124 |
+
pre_norm=True,
|
125 |
+
eps=1e-6,
|
126 |
+
non_linearity="swish",
|
127 |
+
time_embedding_norm="default",
|
128 |
+
output_scale_factor=1.0,
|
129 |
+
use_in_shortcut=None,
|
130 |
+
):
|
131 |
+
super().__init__()
|
132 |
+
self.pre_norm = pre_norm
|
133 |
+
self.pre_norm = True
|
134 |
+
self.in_channels = in_channels
|
135 |
+
out_channels = in_channels if out_channels is None else out_channels
|
136 |
+
self.out_channels = out_channels
|
137 |
+
self.use_conv_shortcut = conv_shortcut
|
138 |
+
self.time_embedding_norm = time_embedding_norm
|
139 |
+
self.output_scale_factor = output_scale_factor
|
140 |
+
|
141 |
+
if groups_out is None:
|
142 |
+
groups_out = groups
|
143 |
+
|
144 |
+
self.norm1 = torch.nn.GroupNorm(num_groups=groups, num_channels=in_channels, eps=eps, affine=True)
|
145 |
+
|
146 |
+
self.conv1 = InflatedConv3d(in_channels, out_channels, kernel_size=3, stride=1, padding=1)
|
147 |
+
|
148 |
+
if temb_channels is not None:
|
149 |
+
if self.time_embedding_norm == "default":
|
150 |
+
time_emb_proj_out_channels = out_channels
|
151 |
+
elif self.time_embedding_norm == "scale_shift":
|
152 |
+
time_emb_proj_out_channels = out_channels * 2
|
153 |
+
else:
|
154 |
+
raise ValueError(f"unknown time_embedding_norm : {self.time_embedding_norm} ")
|
155 |
+
|
156 |
+
self.time_emb_proj = torch.nn.Linear(temb_channels, time_emb_proj_out_channels)
|
157 |
+
else:
|
158 |
+
self.time_emb_proj = None
|
159 |
+
|
160 |
+
self.norm2 = torch.nn.GroupNorm(num_groups=groups_out, num_channels=out_channels, eps=eps, affine=True)
|
161 |
+
self.dropout = torch.nn.Dropout(dropout)
|
162 |
+
self.conv2 = InflatedConv3d(out_channels, out_channels, kernel_size=3, stride=1, padding=1)
|
163 |
+
|
164 |
+
if non_linearity == "swish":
|
165 |
+
self.nonlinearity = lambda x: F.silu(x)
|
166 |
+
elif non_linearity == "mish":
|
167 |
+
self.nonlinearity = Mish()
|
168 |
+
elif non_linearity == "silu":
|
169 |
+
self.nonlinearity = nn.SiLU()
|
170 |
+
|
171 |
+
self.use_in_shortcut = self.in_channels != self.out_channels if use_in_shortcut is None else use_in_shortcut
|
172 |
+
|
173 |
+
self.conv_shortcut = None
|
174 |
+
if self.use_in_shortcut:
|
175 |
+
self.conv_shortcut = InflatedConv3d(in_channels, out_channels, kernel_size=1, stride=1, padding=0)
|
176 |
+
|
177 |
+
def forward(self, input_tensor, temb):
|
178 |
+
hidden_states = input_tensor
|
179 |
+
|
180 |
+
hidden_states = self.norm1(hidden_states)
|
181 |
+
hidden_states = self.nonlinearity(hidden_states)
|
182 |
+
|
183 |
+
hidden_states = self.conv1(hidden_states)
|
184 |
+
|
185 |
+
if temb is not None:
|
186 |
+
temb = self.time_emb_proj(self.nonlinearity(temb))[:, :, None, None, None]
|
187 |
+
|
188 |
+
if temb is not None and self.time_embedding_norm == "default":
|
189 |
+
hidden_states = hidden_states + temb
|
190 |
+
|
191 |
+
hidden_states = self.norm2(hidden_states)
|
192 |
+
|
193 |
+
if temb is not None and self.time_embedding_norm == "scale_shift":
|
194 |
+
scale, shift = torch.chunk(temb, 2, dim=1)
|
195 |
+
hidden_states = hidden_states * (1 + scale) + shift
|
196 |
+
|
197 |
+
hidden_states = self.nonlinearity(hidden_states)
|
198 |
+
|
199 |
+
hidden_states = self.dropout(hidden_states)
|
200 |
+
hidden_states = self.conv2(hidden_states)
|
201 |
+
|
202 |
+
if self.conv_shortcut is not None:
|
203 |
+
input_tensor = self.conv_shortcut(input_tensor)
|
204 |
+
|
205 |
+
output_tensor = (input_tensor + hidden_states) / self.output_scale_factor
|
206 |
+
|
207 |
+
return output_tensor
|
208 |
+
|
209 |
+
|
210 |
+
class Mish(torch.nn.Module):
|
211 |
+
def forward(self, hidden_states):
|
212 |
+
return hidden_states * torch.tanh(torch.nn.functional.softplus(hidden_states))
|
base/models/temporal_attention.py
ADDED
@@ -0,0 +1,388 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from torch import nn
|
3 |
+
from typing import Optional
|
4 |
+
from rotary_embedding_torch import RotaryEmbedding
|
5 |
+
from dataclasses import dataclass
|
6 |
+
from diffusers.utils import BaseOutput
|
7 |
+
from diffusers.utils.import_utils import is_xformers_available
|
8 |
+
import torch.nn.functional as F
|
9 |
+
from einops import rearrange, repeat
|
10 |
+
import math
|
11 |
+
|
12 |
+
@dataclass
|
13 |
+
class Transformer3DModelOutput(BaseOutput):
|
14 |
+
sample: torch.FloatTensor
|
15 |
+
|
16 |
+
|
17 |
+
if is_xformers_available():
|
18 |
+
import xformers
|
19 |
+
import xformers.ops
|
20 |
+
else:
|
21 |
+
xformers = None
|
22 |
+
|
23 |
+
def exists(x):
|
24 |
+
return x is not None
|
25 |
+
|
26 |
+
class CrossAttention(nn.Module):
|
27 |
+
r"""
|
28 |
+
copy from diffuser 0.11.1
|
29 |
+
A cross attention layer.
|
30 |
+
Parameters:
|
31 |
+
query_dim (`int`): The number of channels in the query.
|
32 |
+
cross_attention_dim (`int`, *optional*):
|
33 |
+
The number of channels in the encoder_hidden_states. If not given, defaults to `query_dim`.
|
34 |
+
heads (`int`, *optional*, defaults to 8): The number of heads to use for multi-head attention.
|
35 |
+
dim_head (`int`, *optional*, defaults to 64): The number of channels in each head.
|
36 |
+
dropout (`float`, *optional*, defaults to 0.0): The dropout probability to use.
|
37 |
+
bias (`bool`, *optional*, defaults to False):
|
38 |
+
Set to `True` for the query, key, and value linear layers to contain a bias parameter.
|
39 |
+
"""
|
40 |
+
|
41 |
+
def __init__(
|
42 |
+
self,
|
43 |
+
query_dim: int,
|
44 |
+
cross_attention_dim: Optional[int] = None,
|
45 |
+
heads: int = 8,
|
46 |
+
dim_head: int = 64,
|
47 |
+
dropout: float = 0.0,
|
48 |
+
bias=False,
|
49 |
+
upcast_attention: bool = False,
|
50 |
+
upcast_softmax: bool = False,
|
51 |
+
added_kv_proj_dim: Optional[int] = None,
|
52 |
+
norm_num_groups: Optional[int] = None,
|
53 |
+
use_relative_position: bool = False,
|
54 |
+
):
|
55 |
+
super().__init__()
|
56 |
+
# print('num head', heads)
|
57 |
+
inner_dim = dim_head * heads
|
58 |
+
cross_attention_dim = cross_attention_dim if cross_attention_dim is not None else query_dim
|
59 |
+
self.upcast_attention = upcast_attention
|
60 |
+
self.upcast_softmax = upcast_softmax
|
61 |
+
|
62 |
+
self.scale = dim_head**-0.5
|
63 |
+
|
64 |
+
self.heads = heads
|
65 |
+
self.dim_head = dim_head
|
66 |
+
# for slice_size > 0 the attention score computation
|
67 |
+
# is split across the batch axis to save memory
|
68 |
+
# You can set slice_size with `set_attention_slice`
|
69 |
+
self.sliceable_head_dim = heads
|
70 |
+
self._slice_size = None
|
71 |
+
self._use_memory_efficient_attention_xformers = False # No use xformers for temporal attention
|
72 |
+
self.added_kv_proj_dim = added_kv_proj_dim
|
73 |
+
|
74 |
+
if norm_num_groups is not None:
|
75 |
+
self.group_norm = nn.GroupNorm(num_channels=inner_dim, num_groups=norm_num_groups, eps=1e-5, affine=True)
|
76 |
+
else:
|
77 |
+
self.group_norm = None
|
78 |
+
|
79 |
+
self.to_q = nn.Linear(query_dim, inner_dim, bias=bias)
|
80 |
+
self.to_k = nn.Linear(cross_attention_dim, inner_dim, bias=bias)
|
81 |
+
self.to_v = nn.Linear(cross_attention_dim, inner_dim, bias=bias)
|
82 |
+
|
83 |
+
if self.added_kv_proj_dim is not None:
|
84 |
+
self.add_k_proj = nn.Linear(added_kv_proj_dim, cross_attention_dim)
|
85 |
+
self.add_v_proj = nn.Linear(added_kv_proj_dim, cross_attention_dim)
|
86 |
+
|
87 |
+
self.to_out = nn.ModuleList([])
|
88 |
+
self.to_out.append(nn.Linear(inner_dim, query_dim))
|
89 |
+
self.to_out.append(nn.Dropout(dropout))
|
90 |
+
|
91 |
+
def reshape_heads_to_batch_dim(self, tensor):
|
92 |
+
batch_size, seq_len, dim = tensor.shape
|
93 |
+
head_size = self.heads
|
94 |
+
tensor = tensor.reshape(batch_size, seq_len, head_size, dim // head_size)
|
95 |
+
tensor = tensor.permute(0, 2, 1, 3).reshape(batch_size * head_size, seq_len, dim // head_size)
|
96 |
+
return tensor
|
97 |
+
|
98 |
+
def reshape_batch_dim_to_heads(self, tensor):
|
99 |
+
batch_size, seq_len, dim = tensor.shape
|
100 |
+
head_size = self.heads
|
101 |
+
tensor = tensor.reshape(batch_size // head_size, head_size, seq_len, dim)
|
102 |
+
tensor = tensor.permute(0, 2, 1, 3).reshape(batch_size // head_size, seq_len, dim * head_size)
|
103 |
+
return tensor
|
104 |
+
|
105 |
+
def reshape_for_scores(self, tensor):
|
106 |
+
# split heads and dims
|
107 |
+
# tensor should be [b (h w)] f (d nd)
|
108 |
+
batch_size, seq_len, dim = tensor.shape
|
109 |
+
head_size = self.heads
|
110 |
+
tensor = tensor.reshape(batch_size, seq_len, head_size, dim // head_size)
|
111 |
+
tensor = tensor.permute(0, 2, 1, 3).contiguous()
|
112 |
+
return tensor
|
113 |
+
|
114 |
+
def same_batch_dim_to_heads(self, tensor):
|
115 |
+
batch_size, head_size, seq_len, dim = tensor.shape # [b (h w)] nd f d
|
116 |
+
tensor = tensor.reshape(batch_size, seq_len, dim * head_size)
|
117 |
+
return tensor
|
118 |
+
|
119 |
+
def set_attention_slice(self, slice_size):
|
120 |
+
if slice_size is not None and slice_size > self.sliceable_head_dim:
|
121 |
+
raise ValueError(f"slice_size {slice_size} has to be smaller or equal to {self.sliceable_head_dim}.")
|
122 |
+
|
123 |
+
self._slice_size = slice_size
|
124 |
+
|
125 |
+
def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None, use_image_num=None):
|
126 |
+
batch_size, sequence_length, _ = hidden_states.shape
|
127 |
+
|
128 |
+
encoder_hidden_states = encoder_hidden_states
|
129 |
+
|
130 |
+
if self.group_norm is not None:
|
131 |
+
hidden_states = self.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
|
132 |
+
|
133 |
+
query = self.to_q(hidden_states) # [b (h w)] f (nd * d)
|
134 |
+
|
135 |
+
# print('before reshpape query shape', query.shape)
|
136 |
+
dim = query.shape[-1]
|
137 |
+
query = self.reshape_heads_to_batch_dim(query) # [b (h w) nd] f d
|
138 |
+
# print('after reshape query shape', query.shape)
|
139 |
+
|
140 |
+
if self.added_kv_proj_dim is not None:
|
141 |
+
key = self.to_k(hidden_states)
|
142 |
+
value = self.to_v(hidden_states)
|
143 |
+
encoder_hidden_states_key_proj = self.add_k_proj(encoder_hidden_states)
|
144 |
+
encoder_hidden_states_value_proj = self.add_v_proj(encoder_hidden_states)
|
145 |
+
|
146 |
+
key = self.reshape_heads_to_batch_dim(key)
|
147 |
+
value = self.reshape_heads_to_batch_dim(value)
|
148 |
+
encoder_hidden_states_key_proj = self.reshape_heads_to_batch_dim(encoder_hidden_states_key_proj)
|
149 |
+
encoder_hidden_states_value_proj = self.reshape_heads_to_batch_dim(encoder_hidden_states_value_proj)
|
150 |
+
|
151 |
+
key = torch.concat([encoder_hidden_states_key_proj, key], dim=1)
|
152 |
+
value = torch.concat([encoder_hidden_states_value_proj, value], dim=1)
|
153 |
+
else:
|
154 |
+
encoder_hidden_states = encoder_hidden_states if encoder_hidden_states is not None else hidden_states
|
155 |
+
key = self.to_k(encoder_hidden_states)
|
156 |
+
value = self.to_v(encoder_hidden_states)
|
157 |
+
|
158 |
+
key = self.reshape_heads_to_batch_dim(key)
|
159 |
+
value = self.reshape_heads_to_batch_dim(value)
|
160 |
+
|
161 |
+
if attention_mask is not None:
|
162 |
+
if attention_mask.shape[-1] != query.shape[1]:
|
163 |
+
target_length = query.shape[1]
|
164 |
+
attention_mask = F.pad(attention_mask, (0, target_length), value=0.0)
|
165 |
+
attention_mask = attention_mask.repeat_interleave(self.heads, dim=0)
|
166 |
+
|
167 |
+
hidden_states = self._attention(query, key, value, attention_mask)
|
168 |
+
|
169 |
+
# linear proj
|
170 |
+
hidden_states = self.to_out[0](hidden_states)
|
171 |
+
|
172 |
+
# dropout
|
173 |
+
hidden_states = self.to_out[1](hidden_states)
|
174 |
+
return hidden_states
|
175 |
+
|
176 |
+
|
177 |
+
def _attention(self, query, key, value, attention_mask=None):
|
178 |
+
if self.upcast_attention:
|
179 |
+
query = query.float()
|
180 |
+
key = key.float()
|
181 |
+
|
182 |
+
attention_scores = torch.baddbmm(
|
183 |
+
torch.empty(query.shape[0], query.shape[1], key.shape[1], dtype=query.dtype, device=query.device),
|
184 |
+
query,
|
185 |
+
key.transpose(-1, -2),
|
186 |
+
beta=0,
|
187 |
+
alpha=self.scale,
|
188 |
+
)
|
189 |
+
|
190 |
+
if attention_mask is not None:
|
191 |
+
attention_scores = attention_scores + attention_mask
|
192 |
+
|
193 |
+
if self.upcast_softmax:
|
194 |
+
attention_scores = attention_scores.float()
|
195 |
+
|
196 |
+
attention_probs = attention_scores.softmax(dim=-1)
|
197 |
+
attention_probs = attention_probs.to(value.dtype)
|
198 |
+
# compute attention output
|
199 |
+
hidden_states = torch.bmm(attention_probs, value)
|
200 |
+
# reshape hidden_states
|
201 |
+
hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
|
202 |
+
return hidden_states
|
203 |
+
|
204 |
+
def _sliced_attention(self, query, key, value, sequence_length, dim, attention_mask):
|
205 |
+
batch_size_attention = query.shape[0]
|
206 |
+
hidden_states = torch.zeros(
|
207 |
+
(batch_size_attention, sequence_length, dim // self.heads), device=query.device, dtype=query.dtype
|
208 |
+
)
|
209 |
+
slice_size = self._slice_size if self._slice_size is not None else hidden_states.shape[0]
|
210 |
+
for i in range(hidden_states.shape[0] // slice_size):
|
211 |
+
start_idx = i * slice_size
|
212 |
+
end_idx = (i + 1) * slice_size
|
213 |
+
|
214 |
+
query_slice = query[start_idx:end_idx]
|
215 |
+
key_slice = key[start_idx:end_idx]
|
216 |
+
|
217 |
+
if self.upcast_attention:
|
218 |
+
query_slice = query_slice.float()
|
219 |
+
key_slice = key_slice.float()
|
220 |
+
|
221 |
+
attn_slice = torch.baddbmm(
|
222 |
+
torch.empty(slice_size, query.shape[1], key.shape[1], dtype=query_slice.dtype, device=query.device),
|
223 |
+
query_slice,
|
224 |
+
key_slice.transpose(-1, -2),
|
225 |
+
beta=0,
|
226 |
+
alpha=self.scale,
|
227 |
+
)
|
228 |
+
|
229 |
+
if attention_mask is not None:
|
230 |
+
attn_slice = attn_slice + attention_mask[start_idx:end_idx]
|
231 |
+
|
232 |
+
if self.upcast_softmax:
|
233 |
+
attn_slice = attn_slice.float()
|
234 |
+
|
235 |
+
attn_slice = attn_slice.softmax(dim=-1)
|
236 |
+
|
237 |
+
# cast back to the original dtype
|
238 |
+
attn_slice = attn_slice.to(value.dtype)
|
239 |
+
attn_slice = torch.bmm(attn_slice, value[start_idx:end_idx])
|
240 |
+
|
241 |
+
hidden_states[start_idx:end_idx] = attn_slice
|
242 |
+
|
243 |
+
# reshape hidden_states
|
244 |
+
hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
|
245 |
+
return hidden_states
|
246 |
+
|
247 |
+
def _memory_efficient_attention_xformers(self, query, key, value, attention_mask):
|
248 |
+
# TODO attention_mask
|
249 |
+
query = query.contiguous()
|
250 |
+
key = key.contiguous()
|
251 |
+
value = value.contiguous()
|
252 |
+
hidden_states = xformers.ops.memory_efficient_attention(query, key, value, attn_bias=attention_mask)
|
253 |
+
hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
|
254 |
+
return hidden_states
|
255 |
+
|
256 |
+
class TemporalAttention(CrossAttention):
|
257 |
+
def __init__(self,
|
258 |
+
query_dim: int,
|
259 |
+
cross_attention_dim: Optional[int] = None,
|
260 |
+
heads: int = 8,
|
261 |
+
dim_head: int = 64,
|
262 |
+
dropout: float = 0.0,
|
263 |
+
bias=False,
|
264 |
+
upcast_attention: bool = False,
|
265 |
+
upcast_softmax: bool = False,
|
266 |
+
added_kv_proj_dim: Optional[int] = None,
|
267 |
+
norm_num_groups: Optional[int] = None,
|
268 |
+
rotary_emb=None):
|
269 |
+
super().__init__(query_dim, cross_attention_dim, heads, dim_head, dropout, bias, upcast_attention, upcast_softmax, added_kv_proj_dim, norm_num_groups)
|
270 |
+
# relative time positional embeddings
|
271 |
+
self.time_rel_pos_bias = RelativePositionBias(heads=heads, max_distance=32) # realistically will not be able to generate that many frames of video... yet
|
272 |
+
self.rotary_emb = rotary_emb
|
273 |
+
|
274 |
+
def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None):
|
275 |
+
time_rel_pos_bias = self.time_rel_pos_bias(hidden_states.shape[1], device=hidden_states.device)
|
276 |
+
batch_size, sequence_length, _ = hidden_states.shape
|
277 |
+
|
278 |
+
encoder_hidden_states = encoder_hidden_states
|
279 |
+
|
280 |
+
if self.group_norm is not None:
|
281 |
+
hidden_states = self.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
|
282 |
+
|
283 |
+
query = self.to_q(hidden_states) # [b (h w)] f (nd * d)
|
284 |
+
dim = query.shape[-1]
|
285 |
+
|
286 |
+
if self.added_kv_proj_dim is not None:
|
287 |
+
key = self.to_k(hidden_states)
|
288 |
+
value = self.to_v(hidden_states)
|
289 |
+
encoder_hidden_states_key_proj = self.add_k_proj(encoder_hidden_states)
|
290 |
+
encoder_hidden_states_value_proj = self.add_v_proj(encoder_hidden_states)
|
291 |
+
|
292 |
+
key = self.reshape_heads_to_batch_dim(key)
|
293 |
+
value = self.reshape_heads_to_batch_dim(value)
|
294 |
+
encoder_hidden_states_key_proj = self.reshape_heads_to_batch_dim(encoder_hidden_states_key_proj)
|
295 |
+
encoder_hidden_states_value_proj = self.reshape_heads_to_batch_dim(encoder_hidden_states_value_proj)
|
296 |
+
|
297 |
+
key = torch.concat([encoder_hidden_states_key_proj, key], dim=1)
|
298 |
+
value = torch.concat([encoder_hidden_states_value_proj, value], dim=1)
|
299 |
+
else:
|
300 |
+
encoder_hidden_states = encoder_hidden_states if encoder_hidden_states is not None else hidden_states
|
301 |
+
key = self.to_k(encoder_hidden_states)
|
302 |
+
value = self.to_v(encoder_hidden_states)
|
303 |
+
|
304 |
+
if attention_mask is not None:
|
305 |
+
if attention_mask.shape[-1] != query.shape[1]:
|
306 |
+
target_length = query.shape[1]
|
307 |
+
attention_mask = F.pad(attention_mask, (0, target_length), value=0.0)
|
308 |
+
attention_mask = attention_mask.repeat_interleave(self.heads, dim=0)
|
309 |
+
|
310 |
+
if self._slice_size is None or query.shape[0] // self._slice_size == 1:
|
311 |
+
hidden_states = self._attention(query, key, value, attention_mask, time_rel_pos_bias)
|
312 |
+
else:
|
313 |
+
hidden_states = self._sliced_attention(query, key, value, sequence_length, dim, attention_mask)
|
314 |
+
|
315 |
+
# linear proj
|
316 |
+
hidden_states = self.to_out[0](hidden_states)
|
317 |
+
|
318 |
+
# dropout
|
319 |
+
hidden_states = self.to_out[1](hidden_states)
|
320 |
+
return hidden_states
|
321 |
+
|
322 |
+
|
323 |
+
def _attention(self, query, key, value, attention_mask=None, time_rel_pos_bias=None):
|
324 |
+
if self.upcast_attention:
|
325 |
+
query = query.float()
|
326 |
+
key = key.float()
|
327 |
+
|
328 |
+
query = self.scale * rearrange(query, 'b f (h d) -> b h f d', h=self.heads) # d: dim_head; n: heads
|
329 |
+
key = rearrange(key, 'b f (h d) -> b h f d', h=self.heads) # d: dim_head; n: heads
|
330 |
+
value = rearrange(value, 'b f (h d) -> b h f d', h=self.heads) # d: dim_head; n: heads
|
331 |
+
if exists(self.rotary_emb):
|
332 |
+
query = self.rotary_emb.rotate_queries_or_keys(query)
|
333 |
+
key = self.rotary_emb.rotate_queries_or_keys(key)
|
334 |
+
|
335 |
+
attention_scores = torch.einsum('... h i d, ... h j d -> ... h i j', query, key)
|
336 |
+
attention_scores = attention_scores + time_rel_pos_bias
|
337 |
+
|
338 |
+
if attention_mask is not None:
|
339 |
+
# add attention mask
|
340 |
+
attention_scores = attention_scores + attention_mask
|
341 |
+
|
342 |
+
attention_scores = attention_scores - attention_scores.amax(dim = -1, keepdim = True).detach()
|
343 |
+
attention_probs = nn.functional.softmax(attention_scores, dim=-1)
|
344 |
+
|
345 |
+
attention_probs = attention_probs.to(value.dtype)
|
346 |
+
hidden_states = torch.einsum('... h i j, ... h j d -> ... h i d', attention_probs, value)
|
347 |
+
hidden_states = rearrange(hidden_states, 'b h f d -> b f (h d)')
|
348 |
+
return hidden_states
|
349 |
+
|
350 |
+
class RelativePositionBias(nn.Module):
|
351 |
+
def __init__(
|
352 |
+
self,
|
353 |
+
heads=8,
|
354 |
+
num_buckets=32,
|
355 |
+
max_distance=128,
|
356 |
+
):
|
357 |
+
super().__init__()
|
358 |
+
self.num_buckets = num_buckets
|
359 |
+
self.max_distance = max_distance
|
360 |
+
self.relative_attention_bias = nn.Embedding(num_buckets, heads)
|
361 |
+
|
362 |
+
@staticmethod
|
363 |
+
def _relative_position_bucket(relative_position, num_buckets=32, max_distance=128):
|
364 |
+
ret = 0
|
365 |
+
n = -relative_position
|
366 |
+
|
367 |
+
num_buckets //= 2
|
368 |
+
ret += (n < 0).long() * num_buckets
|
369 |
+
n = torch.abs(n)
|
370 |
+
|
371 |
+
max_exact = num_buckets // 2
|
372 |
+
is_small = n < max_exact
|
373 |
+
|
374 |
+
val_if_large = max_exact + (
|
375 |
+
torch.log(n.float() / max_exact) / math.log(max_distance / max_exact) * (num_buckets - max_exact)
|
376 |
+
).long()
|
377 |
+
val_if_large = torch.min(val_if_large, torch.full_like(val_if_large, num_buckets - 1))
|
378 |
+
|
379 |
+
ret += torch.where(is_small, n, val_if_large)
|
380 |
+
return ret
|
381 |
+
|
382 |
+
def forward(self, n, device):
|
383 |
+
q_pos = torch.arange(n, dtype = torch.long, device = device)
|
384 |
+
k_pos = torch.arange(n, dtype = torch.long, device = device)
|
385 |
+
rel_pos = rearrange(k_pos, 'j -> 1 j') - rearrange(q_pos, 'i -> i 1')
|
386 |
+
rp_bucket = self._relative_position_bucket(rel_pos, num_buckets = self.num_buckets, max_distance = self.max_distance)
|
387 |
+
values = self.relative_attention_bias(rp_bucket)
|
388 |
+
return rearrange(values, 'i j h -> h i j') # num_heads, num_frames, num_frames
|
base/models/transformer_3d.py
ADDED
@@ -0,0 +1,367 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
3 |
+
# you may not use this file except in compliance with the License.
|
4 |
+
# You may obtain a copy of the License at
|
5 |
+
#
|
6 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
7 |
+
#
|
8 |
+
# Unless required by applicable law or agreed to in writing, software
|
9 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
10 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
11 |
+
# See the License for the specific language governing permissions and
|
12 |
+
# limitations under the License.
|
13 |
+
from dataclasses import dataclass
|
14 |
+
from typing import Any, Dict, Optional
|
15 |
+
|
16 |
+
import torch
|
17 |
+
import torch.nn.functional as F
|
18 |
+
from torch import nn
|
19 |
+
|
20 |
+
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
21 |
+
from diffusers.models.embeddings import ImagePositionalEmbeddings
|
22 |
+
from diffusers.utils import BaseOutput, deprecate
|
23 |
+
from diffusers.models.embeddings import PatchEmbed
|
24 |
+
from diffusers.models.lora import LoRACompatibleConv, LoRACompatibleLinear
|
25 |
+
from diffusers.models.modeling_utils import ModelMixin
|
26 |
+
from einops import rearrange, repeat
|
27 |
+
|
28 |
+
try:
|
29 |
+
from attention import BasicTransformerBlock
|
30 |
+
except:
|
31 |
+
from .attention import BasicTransformerBlock
|
32 |
+
|
33 |
+
@dataclass
|
34 |
+
class Transformer3DModelOutput(BaseOutput):
|
35 |
+
"""
|
36 |
+
The output of [`Transformer2DModel`].
|
37 |
+
|
38 |
+
Args:
|
39 |
+
sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` or `(batch size, num_vector_embeds - 1, num_latent_pixels)` if [`Transformer2DModel`] is discrete):
|
40 |
+
The hidden states output conditioned on the `encoder_hidden_states` input. If discrete, returns probability
|
41 |
+
distributions for the unnoised latent pixels.
|
42 |
+
"""
|
43 |
+
|
44 |
+
sample: torch.FloatTensor
|
45 |
+
|
46 |
+
|
47 |
+
class Transformer3DModel(ModelMixin, ConfigMixin):
|
48 |
+
"""
|
49 |
+
A 2D Transformer model for image-like data.
|
50 |
+
|
51 |
+
Parameters:
|
52 |
+
num_attention_heads (`int`, *optional*, defaults to 16): The number of heads to use for multi-head attention.
|
53 |
+
attention_head_dim (`int`, *optional*, defaults to 88): The number of channels in each head.
|
54 |
+
in_channels (`int`, *optional*):
|
55 |
+
The number of channels in the input and output (specify if the input is **continuous**).
|
56 |
+
num_layers (`int`, *optional*, defaults to 1): The number of layers of Transformer blocks to use.
|
57 |
+
dropout (`float`, *optional*, defaults to 0.0): The dropout probability to use.
|
58 |
+
cross_attention_dim (`int`, *optional*): The number of `encoder_hidden_states` dimensions to use.
|
59 |
+
sample_size (`int`, *optional*): The width of the latent images (specify if the input is **discrete**).
|
60 |
+
This is fixed during training since it is used to learn a number of position embeddings.
|
61 |
+
num_vector_embeds (`int`, *optional*):
|
62 |
+
The number of classes of the vector embeddings of the latent pixels (specify if the input is **discrete**).
|
63 |
+
Includes the class for the masked latent pixel.
|
64 |
+
activation_fn (`str`, *optional*, defaults to `"geglu"`): Activation function to use in feed-forward.
|
65 |
+
num_embeds_ada_norm ( `int`, *optional*):
|
66 |
+
The number of diffusion steps used during training. Pass if at least one of the norm_layers is
|
67 |
+
`AdaLayerNorm`. This is fixed during training since it is used to learn a number of embeddings that are
|
68 |
+
added to the hidden states.
|
69 |
+
|
70 |
+
During inference, you can denoise for up to but not more steps than `num_embeds_ada_norm`.
|
71 |
+
attention_bias (`bool`, *optional*):
|
72 |
+
Configure if the `TransformerBlocks` attention should contain a bias parameter.
|
73 |
+
"""
|
74 |
+
|
75 |
+
@register_to_config
|
76 |
+
def __init__(
|
77 |
+
self,
|
78 |
+
num_attention_heads: int = 16,
|
79 |
+
attention_head_dim: int = 88,
|
80 |
+
in_channels: Optional[int] = None,
|
81 |
+
out_channels: Optional[int] = None,
|
82 |
+
num_layers: int = 1,
|
83 |
+
dropout: float = 0.0,
|
84 |
+
norm_num_groups: int = 32,
|
85 |
+
cross_attention_dim: Optional[int] = None,
|
86 |
+
attention_bias: bool = False,
|
87 |
+
sample_size: Optional[int] = None,
|
88 |
+
num_vector_embeds: Optional[int] = None,
|
89 |
+
patch_size: Optional[int] = None,
|
90 |
+
activation_fn: str = "geglu",
|
91 |
+
num_embeds_ada_norm: Optional[int] = None,
|
92 |
+
use_linear_projection: bool = False,
|
93 |
+
only_cross_attention: bool = False,
|
94 |
+
upcast_attention: bool = False,
|
95 |
+
norm_type: str = "layer_norm",
|
96 |
+
norm_elementwise_affine: bool = True,
|
97 |
+
rotary_emb=None,
|
98 |
+
):
|
99 |
+
super().__init__()
|
100 |
+
self.use_linear_projection = use_linear_projection
|
101 |
+
self.num_attention_heads = num_attention_heads
|
102 |
+
self.attention_head_dim = attention_head_dim
|
103 |
+
inner_dim = num_attention_heads * attention_head_dim
|
104 |
+
|
105 |
+
# 1. Transformer2DModel can process both standard continuous images of shape `(batch_size, num_channels, width, height)` as well as quantized image embeddings of shape `(batch_size, num_image_vectors)`
|
106 |
+
# Define whether input is continuous or discrete depending on configuration
|
107 |
+
self.is_input_continuous = (in_channels is not None) and (patch_size is None)
|
108 |
+
self.is_input_vectorized = num_vector_embeds is not None
|
109 |
+
self.is_input_patches = in_channels is not None and patch_size is not None
|
110 |
+
|
111 |
+
if norm_type == "layer_norm" and num_embeds_ada_norm is not None:
|
112 |
+
deprecation_message = (
|
113 |
+
f"The configuration file of this model: {self.__class__} is outdated. `norm_type` is either not set or"
|
114 |
+
" incorrectly set to `'layer_norm'`.Make sure to set `norm_type` to `'ada_norm'` in the config."
|
115 |
+
" Please make sure to update the config accordingly as leaving `norm_type` might led to incorrect"
|
116 |
+
" results in future versions. If you have downloaded this checkpoint from the Hugging Face Hub, it"
|
117 |
+
" would be very nice if you could open a Pull request for the `transformer/config.json` file"
|
118 |
+
)
|
119 |
+
deprecate("norm_type!=num_embeds_ada_norm", "1.0.0", deprecation_message, standard_warn=False)
|
120 |
+
norm_type = "ada_norm"
|
121 |
+
|
122 |
+
if self.is_input_continuous and self.is_input_vectorized:
|
123 |
+
raise ValueError(
|
124 |
+
f"Cannot define both `in_channels`: {in_channels} and `num_vector_embeds`: {num_vector_embeds}. Make"
|
125 |
+
" sure that either `in_channels` or `num_vector_embeds` is None."
|
126 |
+
)
|
127 |
+
elif self.is_input_vectorized and self.is_input_patches:
|
128 |
+
raise ValueError(
|
129 |
+
f"Cannot define both `num_vector_embeds`: {num_vector_embeds} and `patch_size`: {patch_size}. Make"
|
130 |
+
" sure that either `num_vector_embeds` or `num_patches` is None."
|
131 |
+
)
|
132 |
+
elif not self.is_input_continuous and not self.is_input_vectorized and not self.is_input_patches:
|
133 |
+
raise ValueError(
|
134 |
+
f"Has to define `in_channels`: {in_channels}, `num_vector_embeds`: {num_vector_embeds}, or patch_size:"
|
135 |
+
f" {patch_size}. Make sure that `in_channels`, `num_vector_embeds` or `num_patches` is not None."
|
136 |
+
)
|
137 |
+
|
138 |
+
# 2. Define input layers
|
139 |
+
if self.is_input_continuous:
|
140 |
+
self.in_channels = in_channels
|
141 |
+
|
142 |
+
self.norm = torch.nn.GroupNorm(num_groups=norm_num_groups, num_channels=in_channels, eps=1e-6, affine=True)
|
143 |
+
if use_linear_projection:
|
144 |
+
self.proj_in = LoRACompatibleLinear(in_channels, inner_dim)
|
145 |
+
else:
|
146 |
+
self.proj_in = LoRACompatibleConv(in_channels, inner_dim, kernel_size=1, stride=1, padding=0)
|
147 |
+
elif self.is_input_vectorized:
|
148 |
+
assert sample_size is not None, "Transformer2DModel over discrete input must provide sample_size"
|
149 |
+
assert num_vector_embeds is not None, "Transformer2DModel over discrete input must provide num_embed"
|
150 |
+
|
151 |
+
self.height = sample_size
|
152 |
+
self.width = sample_size
|
153 |
+
self.num_vector_embeds = num_vector_embeds
|
154 |
+
self.num_latent_pixels = self.height * self.width
|
155 |
+
|
156 |
+
self.latent_image_embedding = ImagePositionalEmbeddings(
|
157 |
+
num_embed=num_vector_embeds, embed_dim=inner_dim, height=self.height, width=self.width
|
158 |
+
)
|
159 |
+
elif self.is_input_patches:
|
160 |
+
assert sample_size is not None, "Transformer2DModel over patched input must provide sample_size"
|
161 |
+
|
162 |
+
self.height = sample_size
|
163 |
+
self.width = sample_size
|
164 |
+
|
165 |
+
self.patch_size = patch_size
|
166 |
+
self.pos_embed = PatchEmbed(
|
167 |
+
height=sample_size,
|
168 |
+
width=sample_size,
|
169 |
+
patch_size=patch_size,
|
170 |
+
in_channels=in_channels,
|
171 |
+
embed_dim=inner_dim,
|
172 |
+
)
|
173 |
+
|
174 |
+
# 3. Define transformers blocks
|
175 |
+
self.transformer_blocks = nn.ModuleList(
|
176 |
+
[
|
177 |
+
BasicTransformerBlock(
|
178 |
+
inner_dim,
|
179 |
+
num_attention_heads,
|
180 |
+
attention_head_dim,
|
181 |
+
dropout=dropout,
|
182 |
+
cross_attention_dim=cross_attention_dim,
|
183 |
+
activation_fn=activation_fn,
|
184 |
+
num_embeds_ada_norm=num_embeds_ada_norm,
|
185 |
+
attention_bias=attention_bias,
|
186 |
+
only_cross_attention=only_cross_attention,
|
187 |
+
upcast_attention=upcast_attention,
|
188 |
+
norm_type=norm_type,
|
189 |
+
norm_elementwise_affine=norm_elementwise_affine,
|
190 |
+
rotary_emb=rotary_emb,
|
191 |
+
)
|
192 |
+
for d in range(num_layers)
|
193 |
+
]
|
194 |
+
)
|
195 |
+
|
196 |
+
# 4. Define output layers
|
197 |
+
self.out_channels = in_channels if out_channels is None else out_channels
|
198 |
+
if self.is_input_continuous:
|
199 |
+
# TODO: should use out_channels for continuous projections
|
200 |
+
if use_linear_projection:
|
201 |
+
self.proj_out = LoRACompatibleLinear(inner_dim, in_channels)
|
202 |
+
else:
|
203 |
+
self.proj_out = LoRACompatibleConv(inner_dim, in_channels, kernel_size=1, stride=1, padding=0)
|
204 |
+
elif self.is_input_vectorized:
|
205 |
+
self.norm_out = nn.LayerNorm(inner_dim)
|
206 |
+
self.out = nn.Linear(inner_dim, self.num_vector_embeds - 1)
|
207 |
+
elif self.is_input_patches:
|
208 |
+
self.norm_out = nn.LayerNorm(inner_dim, elementwise_affine=False, eps=1e-6)
|
209 |
+
self.proj_out_1 = nn.Linear(inner_dim, 2 * inner_dim)
|
210 |
+
self.proj_out_2 = nn.Linear(inner_dim, patch_size * patch_size * self.out_channels)
|
211 |
+
|
212 |
+
def forward(
|
213 |
+
self,
|
214 |
+
hidden_states: torch.Tensor,
|
215 |
+
encoder_hidden_states: Optional[torch.Tensor] = None,
|
216 |
+
timestep: Optional[torch.LongTensor] = None,
|
217 |
+
class_labels: Optional[torch.LongTensor] = None,
|
218 |
+
cross_attention_kwargs: Dict[str, Any] = None,
|
219 |
+
attention_mask: Optional[torch.Tensor] = None,
|
220 |
+
encoder_attention_mask: Optional[torch.Tensor] = None,
|
221 |
+
return_dict: bool = True,
|
222 |
+
use_image_num=None,
|
223 |
+
):
|
224 |
+
"""
|
225 |
+
The [`Transformer2DModel`] forward method.
|
226 |
+
|
227 |
+
Args:
|
228 |
+
hidden_states (`torch.LongTensor` of shape `(batch size, num latent pixels)` if discrete, `torch.FloatTensor` of shape `(batch size, channel, height, width)` if continuous):
|
229 |
+
Input `hidden_states`.
|
230 |
+
encoder_hidden_states ( `torch.FloatTensor` of shape `(batch size, sequence len, embed dims)`, *optional*):
|
231 |
+
Conditional embeddings for cross attention layer. If not given, cross-attention defaults to
|
232 |
+
self-attention.
|
233 |
+
timestep ( `torch.LongTensor`, *optional*):
|
234 |
+
Used to indicate denoising step. Optional timestep to be applied as an embedding in `AdaLayerNorm`.
|
235 |
+
class_labels ( `torch.LongTensor` of shape `(batch size, num classes)`, *optional*):
|
236 |
+
Used to indicate class labels conditioning. Optional class labels to be applied as an embedding in
|
237 |
+
`AdaLayerZeroNorm`.
|
238 |
+
encoder_attention_mask ( `torch.Tensor`, *optional*):
|
239 |
+
Cross-attention mask applied to `encoder_hidden_states`. Two formats supported:
|
240 |
+
|
241 |
+
* Mask `(batch, sequence_length)` True = keep, False = discard.
|
242 |
+
* Bias `(batch, 1, sequence_length)` 0 = keep, -10000 = discard.
|
243 |
+
|
244 |
+
If `ndim == 2`: will be interpreted as a mask, then converted into a bias consistent with the format
|
245 |
+
above. This bias will be added to the cross-attention scores.
|
246 |
+
return_dict (`bool`, *optional*, defaults to `True`):
|
247 |
+
Whether or not to return a [`~models.unet_2d_condition.UNet2DConditionOutput`] instead of a plain
|
248 |
+
tuple.
|
249 |
+
|
250 |
+
Returns:
|
251 |
+
If `return_dict` is True, an [`~models.transformer_2d.Transformer2DModelOutput`] is returned, otherwise a
|
252 |
+
`tuple` where the first element is the sample tensor.
|
253 |
+
"""
|
254 |
+
# ensure attention_mask is a bias, and give it a singleton query_tokens dimension.
|
255 |
+
# we may have done this conversion already, e.g. if we came here via UNet2DConditionModel#forward.
|
256 |
+
# we can tell by counting dims; if ndim == 2: it's a mask rather than a bias.
|
257 |
+
# expects mask of shape:
|
258 |
+
# [batch, key_tokens]
|
259 |
+
# adds singleton query_tokens dimension:
|
260 |
+
# [batch, 1, key_tokens]
|
261 |
+
# this helps to broadcast it as a bias over attention scores, which will be in one of the following shapes:
|
262 |
+
# [batch, heads, query_tokens, key_tokens] (e.g. torch sdp attn)
|
263 |
+
# [batch * heads, query_tokens, key_tokens] (e.g. xformers or classic attn)
|
264 |
+
if attention_mask is not None and attention_mask.ndim == 2:
|
265 |
+
# assume that mask is expressed as:
|
266 |
+
# (1 = keep, 0 = discard)
|
267 |
+
# convert mask into a bias that can be added to attention scores:
|
268 |
+
# (keep = +0, discard = -10000.0)
|
269 |
+
attention_mask = (1 - attention_mask.to(hidden_states.dtype)) * -10000.0
|
270 |
+
attention_mask = attention_mask.unsqueeze(1)
|
271 |
+
|
272 |
+
# convert encoder_attention_mask to a bias the same way we do for attention_mask
|
273 |
+
if encoder_attention_mask is not None and encoder_attention_mask.ndim == 2:
|
274 |
+
encoder_attention_mask = (1 - encoder_attention_mask.to(hidden_states.dtype)) * -10000.0
|
275 |
+
encoder_attention_mask = encoder_attention_mask.unsqueeze(1)
|
276 |
+
|
277 |
+
# 1. Input
|
278 |
+
if self.is_input_continuous: # True
|
279 |
+
|
280 |
+
assert hidden_states.dim() == 5, f"Expected hidden_states to have ndim=5, but got ndim={hidden_states.dim()}."
|
281 |
+
if self.training:
|
282 |
+
video_length = hidden_states.shape[2] - use_image_num
|
283 |
+
hidden_states = rearrange(hidden_states, "b c f h w -> (b f) c h w").contiguous()
|
284 |
+
encoder_hidden_states_length = encoder_hidden_states.shape[1]
|
285 |
+
encoder_hidden_states_video = encoder_hidden_states[:, :encoder_hidden_states_length - use_image_num, ...]
|
286 |
+
encoder_hidden_states_video = repeat(encoder_hidden_states_video, 'b m n c -> b (m f) n c', f=video_length).contiguous()
|
287 |
+
encoder_hidden_states_image = encoder_hidden_states[:, encoder_hidden_states_length - use_image_num:, ...]
|
288 |
+
encoder_hidden_states = torch.cat([encoder_hidden_states_video, encoder_hidden_states_image], dim=1)
|
289 |
+
encoder_hidden_states = rearrange(encoder_hidden_states, 'b m n c -> (b m) n c').contiguous()
|
290 |
+
else:
|
291 |
+
video_length = hidden_states.shape[2]
|
292 |
+
hidden_states = rearrange(hidden_states, "b c f h w -> (b f) c h w").contiguous()
|
293 |
+
encoder_hidden_states = repeat(encoder_hidden_states, 'b n c -> (b f) n c', f=video_length).contiguous()
|
294 |
+
|
295 |
+
batch, _, height, width = hidden_states.shape
|
296 |
+
residual = hidden_states
|
297 |
+
|
298 |
+
hidden_states = self.norm(hidden_states)
|
299 |
+
if not self.use_linear_projection:
|
300 |
+
hidden_states = self.proj_in(hidden_states)
|
301 |
+
inner_dim = hidden_states.shape[1]
|
302 |
+
hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(batch, height * width, inner_dim)
|
303 |
+
else:
|
304 |
+
inner_dim = hidden_states.shape[1]
|
305 |
+
hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(batch, height * width, inner_dim)
|
306 |
+
hidden_states = self.proj_in(hidden_states)
|
307 |
+
elif self.is_input_vectorized:
|
308 |
+
hidden_states = self.latent_image_embedding(hidden_states)
|
309 |
+
elif self.is_input_patches:
|
310 |
+
hidden_states = self.pos_embed(hidden_states)
|
311 |
+
|
312 |
+
# 2. Blocks
|
313 |
+
for block in self.transformer_blocks:
|
314 |
+
hidden_states = block(
|
315 |
+
hidden_states,
|
316 |
+
attention_mask=attention_mask,
|
317 |
+
encoder_hidden_states=encoder_hidden_states,
|
318 |
+
encoder_attention_mask=encoder_attention_mask,
|
319 |
+
timestep=timestep,
|
320 |
+
cross_attention_kwargs=cross_attention_kwargs,
|
321 |
+
class_labels=class_labels,
|
322 |
+
video_length=video_length,
|
323 |
+
use_image_num=use_image_num,
|
324 |
+
)
|
325 |
+
|
326 |
+
# 3. Output
|
327 |
+
if self.is_input_continuous:
|
328 |
+
if not self.use_linear_projection:
|
329 |
+
hidden_states = hidden_states.reshape(batch, height, width, inner_dim).permute(0, 3, 1, 2).contiguous()
|
330 |
+
hidden_states = self.proj_out(hidden_states)
|
331 |
+
else:
|
332 |
+
hidden_states = self.proj_out(hidden_states)
|
333 |
+
hidden_states = hidden_states.reshape(batch, height, width, inner_dim).permute(0, 3, 1, 2).contiguous()
|
334 |
+
|
335 |
+
output = hidden_states + residual
|
336 |
+
elif self.is_input_vectorized:
|
337 |
+
hidden_states = self.norm_out(hidden_states)
|
338 |
+
logits = self.out(hidden_states)
|
339 |
+
# (batch, self.num_vector_embeds - 1, self.num_latent_pixels)
|
340 |
+
logits = logits.permute(0, 2, 1)
|
341 |
+
|
342 |
+
# log(p(x_0))
|
343 |
+
output = F.log_softmax(logits.double(), dim=1).float()
|
344 |
+
elif self.is_input_patches:
|
345 |
+
# TODO: cleanup!
|
346 |
+
conditioning = self.transformer_blocks[0].norm1.emb(
|
347 |
+
timestep, class_labels, hidden_dtype=hidden_states.dtype
|
348 |
+
)
|
349 |
+
shift, scale = self.proj_out_1(F.silu(conditioning)).chunk(2, dim=1)
|
350 |
+
hidden_states = self.norm_out(hidden_states) * (1 + scale[:, None]) + shift[:, None]
|
351 |
+
hidden_states = self.proj_out_2(hidden_states)
|
352 |
+
|
353 |
+
# unpatchify
|
354 |
+
height = width = int(hidden_states.shape[1] ** 0.5)
|
355 |
+
hidden_states = hidden_states.reshape(
|
356 |
+
shape=(-1, height, width, self.patch_size, self.patch_size, self.out_channels)
|
357 |
+
)
|
358 |
+
hidden_states = torch.einsum("nhwpqc->nchpwq", hidden_states)
|
359 |
+
output = hidden_states.reshape(
|
360 |
+
shape=(-1, self.out_channels, height * self.patch_size, width * self.patch_size)
|
361 |
+
)
|
362 |
+
|
363 |
+
output = rearrange(output, "(b f) c h w -> b c f h w", f=video_length + use_image_num).contiguous()
|
364 |
+
if not return_dict:
|
365 |
+
return (output,)
|
366 |
+
|
367 |
+
return Transformer3DModelOutput(sample=output)
|
base/models/unet.py
ADDED
@@ -0,0 +1,617 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unet_2d_condition.py
|
2 |
+
|
3 |
+
from dataclasses import dataclass
|
4 |
+
from typing import List, Optional, Tuple, Union
|
5 |
+
|
6 |
+
import os
|
7 |
+
import sys
|
8 |
+
sys.path.append(os.path.split(sys.path[0])[0])
|
9 |
+
|
10 |
+
import math
|
11 |
+
import json
|
12 |
+
import torch
|
13 |
+
import einops
|
14 |
+
import torch.nn as nn
|
15 |
+
import torch.utils.checkpoint
|
16 |
+
|
17 |
+
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
18 |
+
from diffusers.utils import BaseOutput, logging
|
19 |
+
from diffusers.models.embeddings import TimestepEmbedding, Timesteps
|
20 |
+
|
21 |
+
try:
|
22 |
+
from diffusers.models.modeling_utils import ModelMixin
|
23 |
+
except:
|
24 |
+
from diffusers.modeling_utils import ModelMixin # 0.11.1
|
25 |
+
|
26 |
+
try:
|
27 |
+
from .unet_blocks import (
|
28 |
+
CrossAttnDownBlock3D,
|
29 |
+
CrossAttnUpBlock3D,
|
30 |
+
DownBlock3D,
|
31 |
+
UNetMidBlock3DCrossAttn,
|
32 |
+
UpBlock3D,
|
33 |
+
get_down_block,
|
34 |
+
get_up_block,
|
35 |
+
)
|
36 |
+
from .resnet import InflatedConv3d
|
37 |
+
except:
|
38 |
+
from unet_blocks import (
|
39 |
+
CrossAttnDownBlock3D,
|
40 |
+
CrossAttnUpBlock3D,
|
41 |
+
DownBlock3D,
|
42 |
+
UNetMidBlock3DCrossAttn,
|
43 |
+
UpBlock3D,
|
44 |
+
get_down_block,
|
45 |
+
get_up_block,
|
46 |
+
)
|
47 |
+
from resnet import InflatedConv3d
|
48 |
+
|
49 |
+
from rotary_embedding_torch import RotaryEmbedding
|
50 |
+
|
51 |
+
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
|
52 |
+
|
53 |
+
class RelativePositionBias(nn.Module):
|
54 |
+
def __init__(
|
55 |
+
self,
|
56 |
+
heads=8,
|
57 |
+
num_buckets=32,
|
58 |
+
max_distance=128,
|
59 |
+
):
|
60 |
+
super().__init__()
|
61 |
+
self.num_buckets = num_buckets
|
62 |
+
self.max_distance = max_distance
|
63 |
+
self.relative_attention_bias = nn.Embedding(num_buckets, heads)
|
64 |
+
|
65 |
+
@staticmethod
|
66 |
+
def _relative_position_bucket(relative_position, num_buckets=32, max_distance=128):
|
67 |
+
ret = 0
|
68 |
+
n = -relative_position
|
69 |
+
|
70 |
+
num_buckets //= 2
|
71 |
+
ret += (n < 0).long() * num_buckets
|
72 |
+
n = torch.abs(n)
|
73 |
+
|
74 |
+
max_exact = num_buckets // 2
|
75 |
+
is_small = n < max_exact
|
76 |
+
|
77 |
+
val_if_large = max_exact + (
|
78 |
+
torch.log(n.float() / max_exact) / math.log(max_distance / max_exact) * (num_buckets - max_exact)
|
79 |
+
).long()
|
80 |
+
val_if_large = torch.min(val_if_large, torch.full_like(val_if_large, num_buckets - 1))
|
81 |
+
|
82 |
+
ret += torch.where(is_small, n, val_if_large)
|
83 |
+
return ret
|
84 |
+
|
85 |
+
def forward(self, n, device):
|
86 |
+
q_pos = torch.arange(n, dtype = torch.long, device = device)
|
87 |
+
k_pos = torch.arange(n, dtype = torch.long, device = device)
|
88 |
+
rel_pos = einops.rearrange(k_pos, 'j -> 1 j') - einops.rearrange(q_pos, 'i -> i 1')
|
89 |
+
rp_bucket = self._relative_position_bucket(rel_pos, num_buckets = self.num_buckets, max_distance = self.max_distance)
|
90 |
+
values = self.relative_attention_bias(rp_bucket)
|
91 |
+
return einops.rearrange(values, 'i j h -> h i j') # num_heads, num_frames, num_frames
|
92 |
+
|
93 |
+
@dataclass
|
94 |
+
class UNet3DConditionOutput(BaseOutput):
|
95 |
+
sample: torch.FloatTensor
|
96 |
+
|
97 |
+
|
98 |
+
class UNet3DConditionModel(ModelMixin, ConfigMixin):
|
99 |
+
_supports_gradient_checkpointing = True
|
100 |
+
|
101 |
+
@register_to_config
|
102 |
+
def __init__(
|
103 |
+
self,
|
104 |
+
sample_size: Optional[int] = None, # 64
|
105 |
+
in_channels: int = 4,
|
106 |
+
out_channels: int = 4,
|
107 |
+
center_input_sample: bool = False,
|
108 |
+
flip_sin_to_cos: bool = True,
|
109 |
+
freq_shift: int = 0,
|
110 |
+
down_block_types: Tuple[str] = (
|
111 |
+
"CrossAttnDownBlock3D",
|
112 |
+
"CrossAttnDownBlock3D",
|
113 |
+
"CrossAttnDownBlock3D",
|
114 |
+
"DownBlock3D",
|
115 |
+
),
|
116 |
+
mid_block_type: str = "UNetMidBlock3DCrossAttn",
|
117 |
+
up_block_types: Tuple[str] = (
|
118 |
+
"UpBlock3D",
|
119 |
+
"CrossAttnUpBlock3D",
|
120 |
+
"CrossAttnUpBlock3D",
|
121 |
+
"CrossAttnUpBlock3D"
|
122 |
+
),
|
123 |
+
only_cross_attention: Union[bool, Tuple[bool]] = False,
|
124 |
+
block_out_channels: Tuple[int] = (320, 640, 1280, 1280),
|
125 |
+
layers_per_block: int = 2,
|
126 |
+
downsample_padding: int = 1,
|
127 |
+
mid_block_scale_factor: float = 1,
|
128 |
+
act_fn: str = "silu",
|
129 |
+
norm_num_groups: int = 32,
|
130 |
+
norm_eps: float = 1e-5,
|
131 |
+
cross_attention_dim: int = 1280,
|
132 |
+
attention_head_dim: Union[int, Tuple[int]] = 8,
|
133 |
+
dual_cross_attention: bool = False,
|
134 |
+
use_linear_projection: bool = False,
|
135 |
+
class_embed_type: Optional[str] = None,
|
136 |
+
num_class_embeds: Optional[int] = None,
|
137 |
+
upcast_attention: bool = False,
|
138 |
+
resnet_time_scale_shift: str = "default",
|
139 |
+
use_first_frame: bool = False,
|
140 |
+
use_relative_position: bool = False,
|
141 |
+
):
|
142 |
+
super().__init__()
|
143 |
+
|
144 |
+
# print(use_first_frame)
|
145 |
+
|
146 |
+
self.sample_size = sample_size
|
147 |
+
time_embed_dim = block_out_channels[0] * 4
|
148 |
+
|
149 |
+
# input
|
150 |
+
self.conv_in = InflatedConv3d(in_channels, block_out_channels[0], kernel_size=3, padding=(1, 1))
|
151 |
+
|
152 |
+
# time
|
153 |
+
self.time_proj = Timesteps(block_out_channels[0], flip_sin_to_cos, freq_shift)
|
154 |
+
timestep_input_dim = block_out_channels[0]
|
155 |
+
|
156 |
+
self.time_embedding = TimestepEmbedding(timestep_input_dim, time_embed_dim)
|
157 |
+
|
158 |
+
# class embedding
|
159 |
+
if class_embed_type is None and num_class_embeds is not None:
|
160 |
+
self.class_embedding = nn.Embedding(num_class_embeds, time_embed_dim)
|
161 |
+
elif class_embed_type == "timestep":
|
162 |
+
self.class_embedding = TimestepEmbedding(timestep_input_dim, time_embed_dim)
|
163 |
+
elif class_embed_type == "identity":
|
164 |
+
self.class_embedding = nn.Identity(time_embed_dim, time_embed_dim)
|
165 |
+
else:
|
166 |
+
self.class_embedding = None
|
167 |
+
|
168 |
+
self.down_blocks = nn.ModuleList([])
|
169 |
+
self.mid_block = None
|
170 |
+
self.up_blocks = nn.ModuleList([])
|
171 |
+
|
172 |
+
# print(only_cross_attention)
|
173 |
+
# print(type(only_cross_attention))
|
174 |
+
# exit()
|
175 |
+
if isinstance(only_cross_attention, bool):
|
176 |
+
only_cross_attention = [only_cross_attention] * len(down_block_types)
|
177 |
+
# print(only_cross_attention)
|
178 |
+
# exit()
|
179 |
+
|
180 |
+
if isinstance(attention_head_dim, int):
|
181 |
+
attention_head_dim = (attention_head_dim,) * len(down_block_types)
|
182 |
+
# print(attention_head_dim)
|
183 |
+
# exit()
|
184 |
+
|
185 |
+
rotary_emb = RotaryEmbedding(32)
|
186 |
+
|
187 |
+
# down
|
188 |
+
output_channel = block_out_channels[0]
|
189 |
+
for i, down_block_type in enumerate(down_block_types):
|
190 |
+
input_channel = output_channel
|
191 |
+
output_channel = block_out_channels[i]
|
192 |
+
is_final_block = i == len(block_out_channels) - 1
|
193 |
+
|
194 |
+
down_block = get_down_block(
|
195 |
+
down_block_type,
|
196 |
+
num_layers=layers_per_block,
|
197 |
+
in_channels=input_channel,
|
198 |
+
out_channels=output_channel,
|
199 |
+
temb_channels=time_embed_dim,
|
200 |
+
add_downsample=not is_final_block,
|
201 |
+
resnet_eps=norm_eps,
|
202 |
+
resnet_act_fn=act_fn,
|
203 |
+
resnet_groups=norm_num_groups,
|
204 |
+
cross_attention_dim=cross_attention_dim,
|
205 |
+
attn_num_head_channels=attention_head_dim[i],
|
206 |
+
downsample_padding=downsample_padding,
|
207 |
+
dual_cross_attention=dual_cross_attention,
|
208 |
+
use_linear_projection=use_linear_projection,
|
209 |
+
only_cross_attention=only_cross_attention[i],
|
210 |
+
upcast_attention=upcast_attention,
|
211 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
212 |
+
use_first_frame=use_first_frame,
|
213 |
+
use_relative_position=use_relative_position,
|
214 |
+
rotary_emb=rotary_emb,
|
215 |
+
)
|
216 |
+
self.down_blocks.append(down_block)
|
217 |
+
|
218 |
+
# mid
|
219 |
+
if mid_block_type == "UNetMidBlock3DCrossAttn":
|
220 |
+
self.mid_block = UNetMidBlock3DCrossAttn(
|
221 |
+
in_channels=block_out_channels[-1],
|
222 |
+
temb_channels=time_embed_dim,
|
223 |
+
resnet_eps=norm_eps,
|
224 |
+
resnet_act_fn=act_fn,
|
225 |
+
output_scale_factor=mid_block_scale_factor,
|
226 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
227 |
+
cross_attention_dim=cross_attention_dim,
|
228 |
+
attn_num_head_channels=attention_head_dim[-1],
|
229 |
+
resnet_groups=norm_num_groups,
|
230 |
+
dual_cross_attention=dual_cross_attention,
|
231 |
+
use_linear_projection=use_linear_projection,
|
232 |
+
upcast_attention=upcast_attention,
|
233 |
+
use_first_frame=use_first_frame,
|
234 |
+
use_relative_position=use_relative_position,
|
235 |
+
rotary_emb=rotary_emb,
|
236 |
+
)
|
237 |
+
else:
|
238 |
+
raise ValueError(f"unknown mid_block_type : {mid_block_type}")
|
239 |
+
|
240 |
+
# count how many layers upsample the videos
|
241 |
+
self.num_upsamplers = 0
|
242 |
+
|
243 |
+
# up
|
244 |
+
reversed_block_out_channels = list(reversed(block_out_channels))
|
245 |
+
reversed_attention_head_dim = list(reversed(attention_head_dim))
|
246 |
+
only_cross_attention = list(reversed(only_cross_attention))
|
247 |
+
output_channel = reversed_block_out_channels[0]
|
248 |
+
for i, up_block_type in enumerate(up_block_types):
|
249 |
+
is_final_block = i == len(block_out_channels) - 1
|
250 |
+
|
251 |
+
prev_output_channel = output_channel
|
252 |
+
output_channel = reversed_block_out_channels[i]
|
253 |
+
input_channel = reversed_block_out_channels[min(i + 1, len(block_out_channels) - 1)]
|
254 |
+
|
255 |
+
# add upsample block for all BUT final layer
|
256 |
+
if not is_final_block:
|
257 |
+
add_upsample = True
|
258 |
+
self.num_upsamplers += 1
|
259 |
+
else:
|
260 |
+
add_upsample = False
|
261 |
+
|
262 |
+
up_block = get_up_block(
|
263 |
+
up_block_type,
|
264 |
+
num_layers=layers_per_block + 1,
|
265 |
+
in_channels=input_channel,
|
266 |
+
out_channels=output_channel,
|
267 |
+
prev_output_channel=prev_output_channel,
|
268 |
+
temb_channels=time_embed_dim,
|
269 |
+
add_upsample=add_upsample,
|
270 |
+
resnet_eps=norm_eps,
|
271 |
+
resnet_act_fn=act_fn,
|
272 |
+
resnet_groups=norm_num_groups,
|
273 |
+
cross_attention_dim=cross_attention_dim,
|
274 |
+
attn_num_head_channels=reversed_attention_head_dim[i],
|
275 |
+
dual_cross_attention=dual_cross_attention,
|
276 |
+
use_linear_projection=use_linear_projection,
|
277 |
+
only_cross_attention=only_cross_attention[i],
|
278 |
+
upcast_attention=upcast_attention,
|
279 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
280 |
+
use_first_frame=use_first_frame,
|
281 |
+
use_relative_position=use_relative_position,
|
282 |
+
rotary_emb=rotary_emb,
|
283 |
+
)
|
284 |
+
self.up_blocks.append(up_block)
|
285 |
+
prev_output_channel = output_channel
|
286 |
+
|
287 |
+
# out
|
288 |
+
self.conv_norm_out = nn.GroupNorm(num_channels=block_out_channels[0], num_groups=norm_num_groups, eps=norm_eps)
|
289 |
+
self.conv_act = nn.SiLU()
|
290 |
+
self.conv_out = InflatedConv3d(block_out_channels[0], out_channels, kernel_size=3, padding=1)
|
291 |
+
|
292 |
+
# relative time positional embeddings
|
293 |
+
self.use_relative_position = use_relative_position
|
294 |
+
if self.use_relative_position:
|
295 |
+
self.time_rel_pos_bias = RelativePositionBias(heads=8, max_distance=32) # realistically will not be able to generate that many frames of video... yet
|
296 |
+
|
297 |
+
def set_attention_slice(self, slice_size):
|
298 |
+
r"""
|
299 |
+
Enable sliced attention computation.
|
300 |
+
|
301 |
+
When this option is enabled, the attention module will split the input tensor in slices, to compute attention
|
302 |
+
in several steps. This is useful to save some memory in exchange for a small speed decrease.
|
303 |
+
|
304 |
+
Args:
|
305 |
+
slice_size (`str` or `int` or `list(int)`, *optional*, defaults to `"auto"`):
|
306 |
+
When `"auto"`, halves the input to the attention heads, so attention will be computed in two steps. If
|
307 |
+
`"max"`, maxium amount of memory will be saved by running only one slice at a time. If a number is
|
308 |
+
provided, uses as many slices as `attention_head_dim // slice_size`. In this case, `attention_head_dim`
|
309 |
+
must be a multiple of `slice_size`.
|
310 |
+
"""
|
311 |
+
sliceable_head_dims = []
|
312 |
+
|
313 |
+
def fn_recursive_retrieve_slicable_dims(module: torch.nn.Module):
|
314 |
+
if hasattr(module, "set_attention_slice"):
|
315 |
+
sliceable_head_dims.append(module.sliceable_head_dim)
|
316 |
+
|
317 |
+
for child in module.children():
|
318 |
+
fn_recursive_retrieve_slicable_dims(child)
|
319 |
+
|
320 |
+
# retrieve number of attention layers
|
321 |
+
for module in self.children():
|
322 |
+
fn_recursive_retrieve_slicable_dims(module)
|
323 |
+
|
324 |
+
num_slicable_layers = len(sliceable_head_dims)
|
325 |
+
|
326 |
+
if slice_size == "auto":
|
327 |
+
# half the attention head size is usually a good trade-off between
|
328 |
+
# speed and memory
|
329 |
+
slice_size = [dim // 2 for dim in sliceable_head_dims]
|
330 |
+
elif slice_size == "max":
|
331 |
+
# make smallest slice possible
|
332 |
+
slice_size = num_slicable_layers * [1]
|
333 |
+
|
334 |
+
slice_size = num_slicable_layers * [slice_size] if not isinstance(slice_size, list) else slice_size
|
335 |
+
|
336 |
+
if len(slice_size) != len(sliceable_head_dims):
|
337 |
+
raise ValueError(
|
338 |
+
f"You have provided {len(slice_size)}, but {self.config} has {len(sliceable_head_dims)} different"
|
339 |
+
f" attention layers. Make sure to match `len(slice_size)` to be {len(sliceable_head_dims)}."
|
340 |
+
)
|
341 |
+
|
342 |
+
for i in range(len(slice_size)):
|
343 |
+
size = slice_size[i]
|
344 |
+
dim = sliceable_head_dims[i]
|
345 |
+
if size is not None and size > dim:
|
346 |
+
raise ValueError(f"size {size} has to be smaller or equal to {dim}.")
|
347 |
+
|
348 |
+
# Recursively walk through all the children.
|
349 |
+
# Any children which exposes the set_attention_slice method
|
350 |
+
# gets the message
|
351 |
+
def fn_recursive_set_attention_slice(module: torch.nn.Module, slice_size: List[int]):
|
352 |
+
if hasattr(module, "set_attention_slice"):
|
353 |
+
module.set_attention_slice(slice_size.pop())
|
354 |
+
|
355 |
+
for child in module.children():
|
356 |
+
fn_recursive_set_attention_slice(child, slice_size)
|
357 |
+
|
358 |
+
reversed_slice_size = list(reversed(slice_size))
|
359 |
+
for module in self.children():
|
360 |
+
fn_recursive_set_attention_slice(module, reversed_slice_size)
|
361 |
+
|
362 |
+
def _set_gradient_checkpointing(self, module, value=False):
|
363 |
+
if isinstance(module, (CrossAttnDownBlock3D, DownBlock3D, CrossAttnUpBlock3D, UpBlock3D)):
|
364 |
+
module.gradient_checkpointing = value
|
365 |
+
|
366 |
+
def forward(
|
367 |
+
self,
|
368 |
+
sample: torch.FloatTensor,
|
369 |
+
timestep: Union[torch.Tensor, float, int],
|
370 |
+
encoder_hidden_states: torch.Tensor = None,
|
371 |
+
class_labels: Optional[torch.Tensor] = None,
|
372 |
+
attention_mask: Optional[torch.Tensor] = None,
|
373 |
+
use_image_num: int = 0,
|
374 |
+
return_dict: bool = True,
|
375 |
+
) -> Union[UNet3DConditionOutput, Tuple]:
|
376 |
+
r"""
|
377 |
+
Args:
|
378 |
+
sample (`torch.FloatTensor`): (batch, channel, height, width) noisy inputs tensor
|
379 |
+
timestep (`torch.FloatTensor` or `float` or `int`): (batch) timesteps
|
380 |
+
encoder_hidden_states (`torch.FloatTensor`): (batch, sequence_length, feature_dim) encoder hidden states
|
381 |
+
return_dict (`bool`, *optional*, defaults to `True`):
|
382 |
+
Whether or not to return a [`models.unet_2d_condition.UNet2DConditionOutput`] instead of a plain tuple.
|
383 |
+
|
384 |
+
Returns:
|
385 |
+
[`~models.unet_2d_condition.UNet2DConditionOutput`] or `tuple`:
|
386 |
+
[`~models.unet_2d_condition.UNet2DConditionOutput`] if `return_dict` is True, otherwise a `tuple`. When
|
387 |
+
returning a tuple, the first element is the sample tensor.
|
388 |
+
"""
|
389 |
+
# By default samples have to be AT least a multiple of the overall upsampling factor.
|
390 |
+
# The overall upsampling factor is equal to 2 ** (# num of upsampling layears).
|
391 |
+
# However, the upsampling interpolation output size can be forced to fit any upsampling size
|
392 |
+
# on the fly if necessary.
|
393 |
+
default_overall_up_factor = 2**self.num_upsamplers
|
394 |
+
|
395 |
+
# upsample size should be forwarded when sample is not a multiple of `default_overall_up_factor`
|
396 |
+
forward_upsample_size = False
|
397 |
+
upsample_size = None
|
398 |
+
|
399 |
+
if any(s % default_overall_up_factor != 0 for s in sample.shape[-2:]):
|
400 |
+
logger.info("Forward upsample size to force interpolation output size.")
|
401 |
+
forward_upsample_size = True
|
402 |
+
|
403 |
+
# prepare attention_mask
|
404 |
+
if attention_mask is not None:
|
405 |
+
attention_mask = (1 - attention_mask.to(sample.dtype)) * -10000.0
|
406 |
+
attention_mask = attention_mask.unsqueeze(1)
|
407 |
+
|
408 |
+
# center input if necessary
|
409 |
+
if self.config.center_input_sample:
|
410 |
+
sample = 2 * sample - 1.0
|
411 |
+
|
412 |
+
# time
|
413 |
+
timesteps = timestep
|
414 |
+
if not torch.is_tensor(timesteps):
|
415 |
+
# This would be a good case for the `match` statement (Python 3.10+)
|
416 |
+
is_mps = sample.device.type == "mps"
|
417 |
+
if isinstance(timestep, float):
|
418 |
+
dtype = torch.float32 if is_mps else torch.float64
|
419 |
+
else:
|
420 |
+
dtype = torch.int32 if is_mps else torch.int64
|
421 |
+
timesteps = torch.tensor([timesteps], dtype=dtype, device=sample.device)
|
422 |
+
elif len(timesteps.shape) == 0:
|
423 |
+
timesteps = timesteps[None].to(sample.device)
|
424 |
+
|
425 |
+
# broadcast to batch dimension in a way that's compatible with ONNX/Core ML
|
426 |
+
timesteps = timesteps.expand(sample.shape[0])
|
427 |
+
|
428 |
+
t_emb = self.time_proj(timesteps)
|
429 |
+
|
430 |
+
# timesteps does not contain any weights and will always return f32 tensors
|
431 |
+
# but time_embedding might actually be running in fp16. so we need to cast here.
|
432 |
+
# there might be better ways to encapsulate this.
|
433 |
+
t_emb = t_emb.to(dtype=self.dtype)
|
434 |
+
emb = self.time_embedding(t_emb)
|
435 |
+
|
436 |
+
if self.class_embedding is not None:
|
437 |
+
if class_labels is None:
|
438 |
+
raise ValueError("class_labels should be provided when num_class_embeds > 0")
|
439 |
+
|
440 |
+
if self.config.class_embed_type == "timestep":
|
441 |
+
class_labels = self.time_proj(class_labels)
|
442 |
+
|
443 |
+
class_emb = self.class_embedding(class_labels).to(dtype=self.dtype)
|
444 |
+
# print(emb.shape) # torch.Size([3, 1280])
|
445 |
+
# print(class_emb.shape) # torch.Size([3, 1280])
|
446 |
+
emb = emb + class_emb
|
447 |
+
|
448 |
+
if self.use_relative_position:
|
449 |
+
frame_rel_pos_bias = self.time_rel_pos_bias(sample.shape[2], device=sample.device)
|
450 |
+
else:
|
451 |
+
frame_rel_pos_bias = None
|
452 |
+
|
453 |
+
# pre-process
|
454 |
+
sample = self.conv_in(sample)
|
455 |
+
|
456 |
+
# down
|
457 |
+
down_block_res_samples = (sample,)
|
458 |
+
for downsample_block in self.down_blocks:
|
459 |
+
if hasattr(downsample_block, "has_cross_attention") and downsample_block.has_cross_attention:
|
460 |
+
sample, res_samples = downsample_block(
|
461 |
+
hidden_states=sample,
|
462 |
+
temb=emb,
|
463 |
+
encoder_hidden_states=encoder_hidden_states,
|
464 |
+
attention_mask=attention_mask,
|
465 |
+
use_image_num=use_image_num,
|
466 |
+
)
|
467 |
+
else:
|
468 |
+
sample, res_samples = downsample_block(hidden_states=sample, temb=emb)
|
469 |
+
|
470 |
+
down_block_res_samples += res_samples
|
471 |
+
|
472 |
+
# mid
|
473 |
+
sample = self.mid_block(
|
474 |
+
sample, emb, encoder_hidden_states=encoder_hidden_states, attention_mask=attention_mask, use_image_num=use_image_num,
|
475 |
+
)
|
476 |
+
|
477 |
+
# up
|
478 |
+
for i, upsample_block in enumerate(self.up_blocks):
|
479 |
+
is_final_block = i == len(self.up_blocks) - 1
|
480 |
+
|
481 |
+
res_samples = down_block_res_samples[-len(upsample_block.resnets) :]
|
482 |
+
down_block_res_samples = down_block_res_samples[: -len(upsample_block.resnets)]
|
483 |
+
|
484 |
+
# if we have not reached the final block and need to forward the
|
485 |
+
# upsample size, we do it here
|
486 |
+
if not is_final_block and forward_upsample_size:
|
487 |
+
upsample_size = down_block_res_samples[-1].shape[2:]
|
488 |
+
|
489 |
+
if hasattr(upsample_block, "has_cross_attention") and upsample_block.has_cross_attention:
|
490 |
+
sample = upsample_block(
|
491 |
+
hidden_states=sample,
|
492 |
+
temb=emb,
|
493 |
+
res_hidden_states_tuple=res_samples,
|
494 |
+
encoder_hidden_states=encoder_hidden_states,
|
495 |
+
upsample_size=upsample_size,
|
496 |
+
attention_mask=attention_mask,
|
497 |
+
use_image_num=use_image_num,
|
498 |
+
)
|
499 |
+
else:
|
500 |
+
sample = upsample_block(
|
501 |
+
hidden_states=sample, temb=emb, res_hidden_states_tuple=res_samples, upsample_size=upsample_size
|
502 |
+
)
|
503 |
+
# post-process
|
504 |
+
sample = self.conv_norm_out(sample)
|
505 |
+
sample = self.conv_act(sample)
|
506 |
+
sample = self.conv_out(sample)
|
507 |
+
# print(sample.shape)
|
508 |
+
|
509 |
+
if not return_dict:
|
510 |
+
return (sample,)
|
511 |
+
sample = UNet3DConditionOutput(sample=sample)
|
512 |
+
return sample
|
513 |
+
|
514 |
+
def forward_with_cfg(self,
|
515 |
+
x,
|
516 |
+
t,
|
517 |
+
encoder_hidden_states = None,
|
518 |
+
class_labels: Optional[torch.Tensor] = None,
|
519 |
+
cfg_scale=4.0,
|
520 |
+
use_fp16=False):
|
521 |
+
"""
|
522 |
+
Forward, but also batches the unconditional forward pass for classifier-free guidance.
|
523 |
+
"""
|
524 |
+
# https://github.com/openai/glide-text2im/blob/main/notebooks/text2im.ipynb
|
525 |
+
half = x[: len(x) // 2]
|
526 |
+
combined = torch.cat([half, half], dim=0)
|
527 |
+
if use_fp16:
|
528 |
+
combined = combined.to(dtype=torch.float16)
|
529 |
+
model_out = self.forward(combined, t, encoder_hidden_states, class_labels).sample
|
530 |
+
# For exact reproducibility reasons, we apply classifier-free guidance on only
|
531 |
+
# three channels by default. The standard approach to cfg applies it to all channels.
|
532 |
+
# This can be done by uncommenting the following line and commenting-out the line following that.
|
533 |
+
eps, rest = model_out[:, :4], model_out[:, 4:]
|
534 |
+
# eps, rest = model_out[:, :3], model_out[:, 3:] # b c f h w
|
535 |
+
cond_eps, uncond_eps = torch.split(eps, len(eps) // 2, dim=0)
|
536 |
+
half_eps = uncond_eps + cfg_scale * (cond_eps - uncond_eps)
|
537 |
+
eps = torch.cat([half_eps, half_eps], dim=0)
|
538 |
+
return torch.cat([eps, rest], dim=1)
|
539 |
+
|
540 |
+
@classmethod
|
541 |
+
def from_pretrained_2d(cls, pretrained_model_path, subfolder=None):
|
542 |
+
if subfolder is not None:
|
543 |
+
pretrained_model_path = os.path.join(pretrained_model_path, subfolder)
|
544 |
+
|
545 |
+
|
546 |
+
config_file = os.path.join(pretrained_model_path, 'config.json')
|
547 |
+
if not os.path.isfile(config_file):
|
548 |
+
raise RuntimeError(f"{config_file} does not exist")
|
549 |
+
with open(config_file, "r") as f:
|
550 |
+
config = json.load(f)
|
551 |
+
config["_class_name"] = cls.__name__
|
552 |
+
config["down_block_types"] = [
|
553 |
+
"CrossAttnDownBlock3D",
|
554 |
+
"CrossAttnDownBlock3D",
|
555 |
+
"CrossAttnDownBlock3D",
|
556 |
+
"DownBlock3D"
|
557 |
+
]
|
558 |
+
config["up_block_types"] = [
|
559 |
+
"UpBlock3D",
|
560 |
+
"CrossAttnUpBlock3D",
|
561 |
+
"CrossAttnUpBlock3D",
|
562 |
+
"CrossAttnUpBlock3D"
|
563 |
+
]
|
564 |
+
|
565 |
+
config["use_first_frame"] = False
|
566 |
+
|
567 |
+
from diffusers.utils import WEIGHTS_NAME # diffusion_pytorch_model.bin
|
568 |
+
|
569 |
+
|
570 |
+
model = cls.from_config(config)
|
571 |
+
model_file = os.path.join(pretrained_model_path, WEIGHTS_NAME)
|
572 |
+
if not os.path.isfile(model_file):
|
573 |
+
raise RuntimeError(f"{model_file} does not exist")
|
574 |
+
state_dict = torch.load(model_file, map_location="cpu")
|
575 |
+
for k, v in model.state_dict().items():
|
576 |
+
# print(k)
|
577 |
+
if '_temp' in k:
|
578 |
+
state_dict.update({k: v})
|
579 |
+
if 'attn_fcross' in k: # conpy parms of attn1 to attn_fcross
|
580 |
+
k = k.replace('attn_fcross', 'attn1')
|
581 |
+
state_dict.update({k: state_dict[k]})
|
582 |
+
if 'norm_fcross' in k:
|
583 |
+
k = k.replace('norm_fcross', 'norm1')
|
584 |
+
state_dict.update({k: state_dict[k]})
|
585 |
+
|
586 |
+
model.load_state_dict(state_dict)
|
587 |
+
|
588 |
+
return model
|
589 |
+
|
590 |
+
if __name__ == '__main__':
|
591 |
+
import torch
|
592 |
+
# from xformers.ops import MemoryEfficientAttentionFlashAttentionOp
|
593 |
+
|
594 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
595 |
+
|
596 |
+
pretrained_model_path = "/mnt/petrelfs/maxin/work/pretrained/stable-diffusion-v1-4/" # p cluster
|
597 |
+
unet = UNet3DConditionModel.from_pretrained_2d(pretrained_model_path, subfolder="unet").to(device)
|
598 |
+
# unet.enable_xformers_memory_efficient_attention(attention_op=MemoryEfficientAttentionFlashAttentionOp)
|
599 |
+
unet.enable_xformers_memory_efficient_attention()
|
600 |
+
unet.enable_gradient_checkpointing()
|
601 |
+
|
602 |
+
unet.train()
|
603 |
+
|
604 |
+
use_image_num = 5
|
605 |
+
noisy_latents = torch.randn((2, 4, 16 + use_image_num, 32, 32)).to(device)
|
606 |
+
bsz = noisy_latents.shape[0]
|
607 |
+
timesteps = torch.randint(0, 1000, (bsz,)).to(device)
|
608 |
+
timesteps = timesteps.long()
|
609 |
+
encoder_hidden_states = torch.randn((bsz, 1 + use_image_num, 77, 768)).to(device)
|
610 |
+
# class_labels = torch.randn((bsz, )).to(device)
|
611 |
+
|
612 |
+
|
613 |
+
model_pred = unet(sample=noisy_latents, timestep=timesteps,
|
614 |
+
encoder_hidden_states=encoder_hidden_states,
|
615 |
+
class_labels=None,
|
616 |
+
use_image_num=use_image_num).sample
|
617 |
+
print(model_pred.shape)
|
base/models/unet_blocks.py
ADDED
@@ -0,0 +1,648 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unet_2d_blocks.py
|
2 |
+
import os
|
3 |
+
import sys
|
4 |
+
sys.path.append(os.path.split(sys.path[0])[0])
|
5 |
+
|
6 |
+
import torch
|
7 |
+
from torch import nn
|
8 |
+
|
9 |
+
try:
|
10 |
+
from .attention import Transformer3DModel
|
11 |
+
from .resnet import Downsample3D, ResnetBlock3D, Upsample3D
|
12 |
+
except:
|
13 |
+
from attention import Transformer3DModel
|
14 |
+
from resnet import Downsample3D, ResnetBlock3D, Upsample3D
|
15 |
+
|
16 |
+
|
17 |
+
def get_down_block(
|
18 |
+
down_block_type,
|
19 |
+
num_layers,
|
20 |
+
in_channels,
|
21 |
+
out_channels,
|
22 |
+
temb_channels,
|
23 |
+
add_downsample,
|
24 |
+
resnet_eps,
|
25 |
+
resnet_act_fn,
|
26 |
+
attn_num_head_channels,
|
27 |
+
resnet_groups=None,
|
28 |
+
cross_attention_dim=None,
|
29 |
+
downsample_padding=None,
|
30 |
+
dual_cross_attention=False,
|
31 |
+
use_linear_projection=False,
|
32 |
+
only_cross_attention=False,
|
33 |
+
upcast_attention=False,
|
34 |
+
resnet_time_scale_shift="default",
|
35 |
+
use_first_frame=False,
|
36 |
+
use_relative_position=False,
|
37 |
+
rotary_emb=False,
|
38 |
+
):
|
39 |
+
# print(down_block_type)
|
40 |
+
# print(use_first_frame)
|
41 |
+
down_block_type = down_block_type[7:] if down_block_type.startswith("UNetRes") else down_block_type
|
42 |
+
if down_block_type == "DownBlock3D":
|
43 |
+
return DownBlock3D(
|
44 |
+
num_layers=num_layers,
|
45 |
+
in_channels=in_channels,
|
46 |
+
out_channels=out_channels,
|
47 |
+
temb_channels=temb_channels,
|
48 |
+
add_downsample=add_downsample,
|
49 |
+
resnet_eps=resnet_eps,
|
50 |
+
resnet_act_fn=resnet_act_fn,
|
51 |
+
resnet_groups=resnet_groups,
|
52 |
+
downsample_padding=downsample_padding,
|
53 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
54 |
+
)
|
55 |
+
elif down_block_type == "CrossAttnDownBlock3D":
|
56 |
+
if cross_attention_dim is None:
|
57 |
+
raise ValueError("cross_attention_dim must be specified for CrossAttnDownBlock3D")
|
58 |
+
return CrossAttnDownBlock3D(
|
59 |
+
num_layers=num_layers,
|
60 |
+
in_channels=in_channels,
|
61 |
+
out_channels=out_channels,
|
62 |
+
temb_channels=temb_channels,
|
63 |
+
add_downsample=add_downsample,
|
64 |
+
resnet_eps=resnet_eps,
|
65 |
+
resnet_act_fn=resnet_act_fn,
|
66 |
+
resnet_groups=resnet_groups,
|
67 |
+
downsample_padding=downsample_padding,
|
68 |
+
cross_attention_dim=cross_attention_dim,
|
69 |
+
attn_num_head_channels=attn_num_head_channels,
|
70 |
+
dual_cross_attention=dual_cross_attention,
|
71 |
+
use_linear_projection=use_linear_projection,
|
72 |
+
only_cross_attention=only_cross_attention,
|
73 |
+
upcast_attention=upcast_attention,
|
74 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
75 |
+
use_first_frame=use_first_frame,
|
76 |
+
use_relative_position=use_relative_position,
|
77 |
+
rotary_emb=rotary_emb,
|
78 |
+
)
|
79 |
+
raise ValueError(f"{down_block_type} does not exist.")
|
80 |
+
|
81 |
+
|
82 |
+
def get_up_block(
|
83 |
+
up_block_type,
|
84 |
+
num_layers,
|
85 |
+
in_channels,
|
86 |
+
out_channels,
|
87 |
+
prev_output_channel,
|
88 |
+
temb_channels,
|
89 |
+
add_upsample,
|
90 |
+
resnet_eps,
|
91 |
+
resnet_act_fn,
|
92 |
+
attn_num_head_channels,
|
93 |
+
resnet_groups=None,
|
94 |
+
cross_attention_dim=None,
|
95 |
+
dual_cross_attention=False,
|
96 |
+
use_linear_projection=False,
|
97 |
+
only_cross_attention=False,
|
98 |
+
upcast_attention=False,
|
99 |
+
resnet_time_scale_shift="default",
|
100 |
+
use_first_frame=False,
|
101 |
+
use_relative_position=False,
|
102 |
+
rotary_emb=False,
|
103 |
+
):
|
104 |
+
up_block_type = up_block_type[7:] if up_block_type.startswith("UNetRes") else up_block_type
|
105 |
+
if up_block_type == "UpBlock3D":
|
106 |
+
return UpBlock3D(
|
107 |
+
num_layers=num_layers,
|
108 |
+
in_channels=in_channels,
|
109 |
+
out_channels=out_channels,
|
110 |
+
prev_output_channel=prev_output_channel,
|
111 |
+
temb_channels=temb_channels,
|
112 |
+
add_upsample=add_upsample,
|
113 |
+
resnet_eps=resnet_eps,
|
114 |
+
resnet_act_fn=resnet_act_fn,
|
115 |
+
resnet_groups=resnet_groups,
|
116 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
117 |
+
)
|
118 |
+
elif up_block_type == "CrossAttnUpBlock3D":
|
119 |
+
if cross_attention_dim is None:
|
120 |
+
raise ValueError("cross_attention_dim must be specified for CrossAttnUpBlock3D")
|
121 |
+
return CrossAttnUpBlock3D(
|
122 |
+
num_layers=num_layers,
|
123 |
+
in_channels=in_channels,
|
124 |
+
out_channels=out_channels,
|
125 |
+
prev_output_channel=prev_output_channel,
|
126 |
+
temb_channels=temb_channels,
|
127 |
+
add_upsample=add_upsample,
|
128 |
+
resnet_eps=resnet_eps,
|
129 |
+
resnet_act_fn=resnet_act_fn,
|
130 |
+
resnet_groups=resnet_groups,
|
131 |
+
cross_attention_dim=cross_attention_dim,
|
132 |
+
attn_num_head_channels=attn_num_head_channels,
|
133 |
+
dual_cross_attention=dual_cross_attention,
|
134 |
+
use_linear_projection=use_linear_projection,
|
135 |
+
only_cross_attention=only_cross_attention,
|
136 |
+
upcast_attention=upcast_attention,
|
137 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
138 |
+
use_first_frame=use_first_frame,
|
139 |
+
use_relative_position=use_relative_position,
|
140 |
+
rotary_emb=rotary_emb,
|
141 |
+
)
|
142 |
+
raise ValueError(f"{up_block_type} does not exist.")
|
143 |
+
|
144 |
+
|
145 |
+
class UNetMidBlock3DCrossAttn(nn.Module):
|
146 |
+
def __init__(
|
147 |
+
self,
|
148 |
+
in_channels: int,
|
149 |
+
temb_channels: int,
|
150 |
+
dropout: float = 0.0,
|
151 |
+
num_layers: int = 1,
|
152 |
+
resnet_eps: float = 1e-6,
|
153 |
+
resnet_time_scale_shift: str = "default",
|
154 |
+
resnet_act_fn: str = "swish",
|
155 |
+
resnet_groups: int = 32,
|
156 |
+
resnet_pre_norm: bool = True,
|
157 |
+
attn_num_head_channels=1,
|
158 |
+
output_scale_factor=1.0,
|
159 |
+
cross_attention_dim=1280,
|
160 |
+
dual_cross_attention=False,
|
161 |
+
use_linear_projection=False,
|
162 |
+
upcast_attention=False,
|
163 |
+
use_first_frame=False,
|
164 |
+
use_relative_position=False,
|
165 |
+
rotary_emb=False,
|
166 |
+
):
|
167 |
+
super().__init__()
|
168 |
+
|
169 |
+
self.has_cross_attention = True
|
170 |
+
self.attn_num_head_channels = attn_num_head_channels
|
171 |
+
resnet_groups = resnet_groups if resnet_groups is not None else min(in_channels // 4, 32)
|
172 |
+
|
173 |
+
# there is always at least one resnet
|
174 |
+
resnets = [
|
175 |
+
ResnetBlock3D(
|
176 |
+
in_channels=in_channels,
|
177 |
+
out_channels=in_channels,
|
178 |
+
temb_channels=temb_channels,
|
179 |
+
eps=resnet_eps,
|
180 |
+
groups=resnet_groups,
|
181 |
+
dropout=dropout,
|
182 |
+
time_embedding_norm=resnet_time_scale_shift,
|
183 |
+
non_linearity=resnet_act_fn,
|
184 |
+
output_scale_factor=output_scale_factor,
|
185 |
+
pre_norm=resnet_pre_norm,
|
186 |
+
)
|
187 |
+
]
|
188 |
+
attentions = []
|
189 |
+
|
190 |
+
for _ in range(num_layers):
|
191 |
+
if dual_cross_attention:
|
192 |
+
raise NotImplementedError
|
193 |
+
attentions.append(
|
194 |
+
Transformer3DModel(
|
195 |
+
attn_num_head_channels,
|
196 |
+
in_channels // attn_num_head_channels,
|
197 |
+
in_channels=in_channels,
|
198 |
+
num_layers=1,
|
199 |
+
cross_attention_dim=cross_attention_dim,
|
200 |
+
norm_num_groups=resnet_groups,
|
201 |
+
use_linear_projection=use_linear_projection,
|
202 |
+
upcast_attention=upcast_attention,
|
203 |
+
use_first_frame=use_first_frame,
|
204 |
+
use_relative_position=use_relative_position,
|
205 |
+
rotary_emb=rotary_emb,
|
206 |
+
)
|
207 |
+
)
|
208 |
+
resnets.append(
|
209 |
+
ResnetBlock3D(
|
210 |
+
in_channels=in_channels,
|
211 |
+
out_channels=in_channels,
|
212 |
+
temb_channels=temb_channels,
|
213 |
+
eps=resnet_eps,
|
214 |
+
groups=resnet_groups,
|
215 |
+
dropout=dropout,
|
216 |
+
time_embedding_norm=resnet_time_scale_shift,
|
217 |
+
non_linearity=resnet_act_fn,
|
218 |
+
output_scale_factor=output_scale_factor,
|
219 |
+
pre_norm=resnet_pre_norm,
|
220 |
+
)
|
221 |
+
)
|
222 |
+
|
223 |
+
self.attentions = nn.ModuleList(attentions)
|
224 |
+
self.resnets = nn.ModuleList(resnets)
|
225 |
+
|
226 |
+
def forward(self, hidden_states, temb=None, encoder_hidden_states=None, attention_mask=None, use_image_num=None):
|
227 |
+
hidden_states = self.resnets[0](hidden_states, temb)
|
228 |
+
for attn, resnet in zip(self.attentions, self.resnets[1:]):
|
229 |
+
hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states, use_image_num=use_image_num).sample
|
230 |
+
hidden_states = resnet(hidden_states, temb)
|
231 |
+
|
232 |
+
return hidden_states
|
233 |
+
|
234 |
+
|
235 |
+
class CrossAttnDownBlock3D(nn.Module):
|
236 |
+
def __init__(
|
237 |
+
self,
|
238 |
+
in_channels: int,
|
239 |
+
out_channels: int,
|
240 |
+
temb_channels: int,
|
241 |
+
dropout: float = 0.0,
|
242 |
+
num_layers: int = 1,
|
243 |
+
resnet_eps: float = 1e-6,
|
244 |
+
resnet_time_scale_shift: str = "default",
|
245 |
+
resnet_act_fn: str = "swish",
|
246 |
+
resnet_groups: int = 32,
|
247 |
+
resnet_pre_norm: bool = True,
|
248 |
+
attn_num_head_channels=1,
|
249 |
+
cross_attention_dim=1280,
|
250 |
+
output_scale_factor=1.0,
|
251 |
+
downsample_padding=1,
|
252 |
+
add_downsample=True,
|
253 |
+
dual_cross_attention=False,
|
254 |
+
use_linear_projection=False,
|
255 |
+
only_cross_attention=False,
|
256 |
+
upcast_attention=False,
|
257 |
+
use_first_frame=False,
|
258 |
+
use_relative_position=False,
|
259 |
+
rotary_emb=False,
|
260 |
+
):
|
261 |
+
super().__init__()
|
262 |
+
resnets = []
|
263 |
+
attentions = []
|
264 |
+
|
265 |
+
# print(use_first_frame)
|
266 |
+
|
267 |
+
self.has_cross_attention = True
|
268 |
+
self.attn_num_head_channels = attn_num_head_channels
|
269 |
+
|
270 |
+
for i in range(num_layers):
|
271 |
+
in_channels = in_channels if i == 0 else out_channels
|
272 |
+
resnets.append(
|
273 |
+
ResnetBlock3D(
|
274 |
+
in_channels=in_channels,
|
275 |
+
out_channels=out_channels,
|
276 |
+
temb_channels=temb_channels,
|
277 |
+
eps=resnet_eps,
|
278 |
+
groups=resnet_groups,
|
279 |
+
dropout=dropout,
|
280 |
+
time_embedding_norm=resnet_time_scale_shift,
|
281 |
+
non_linearity=resnet_act_fn,
|
282 |
+
output_scale_factor=output_scale_factor,
|
283 |
+
pre_norm=resnet_pre_norm,
|
284 |
+
)
|
285 |
+
)
|
286 |
+
if dual_cross_attention:
|
287 |
+
raise NotImplementedError
|
288 |
+
attentions.append(
|
289 |
+
Transformer3DModel(
|
290 |
+
attn_num_head_channels,
|
291 |
+
out_channels // attn_num_head_channels,
|
292 |
+
in_channels=out_channels,
|
293 |
+
num_layers=1,
|
294 |
+
cross_attention_dim=cross_attention_dim,
|
295 |
+
norm_num_groups=resnet_groups,
|
296 |
+
use_linear_projection=use_linear_projection,
|
297 |
+
only_cross_attention=only_cross_attention,
|
298 |
+
upcast_attention=upcast_attention,
|
299 |
+
use_first_frame=use_first_frame,
|
300 |
+
use_relative_position=use_relative_position,
|
301 |
+
rotary_emb=rotary_emb,
|
302 |
+
)
|
303 |
+
)
|
304 |
+
self.attentions = nn.ModuleList(attentions)
|
305 |
+
self.resnets = nn.ModuleList(resnets)
|
306 |
+
|
307 |
+
if add_downsample:
|
308 |
+
self.downsamplers = nn.ModuleList(
|
309 |
+
[
|
310 |
+
Downsample3D(
|
311 |
+
out_channels, use_conv=True, out_channels=out_channels, padding=downsample_padding, name="op"
|
312 |
+
)
|
313 |
+
]
|
314 |
+
)
|
315 |
+
else:
|
316 |
+
self.downsamplers = None
|
317 |
+
|
318 |
+
self.gradient_checkpointing = False
|
319 |
+
|
320 |
+
def forward(self, hidden_states, temb=None, encoder_hidden_states=None, attention_mask=None, use_image_num=None):
|
321 |
+
output_states = ()
|
322 |
+
|
323 |
+
for resnet, attn in zip(self.resnets, self.attentions):
|
324 |
+
if self.training and self.gradient_checkpointing:
|
325 |
+
|
326 |
+
def create_custom_forward(module, return_dict=None):
|
327 |
+
def custom_forward(*inputs):
|
328 |
+
if return_dict is not None:
|
329 |
+
return module(*inputs, return_dict=return_dict)
|
330 |
+
else:
|
331 |
+
return module(*inputs)
|
332 |
+
|
333 |
+
return custom_forward
|
334 |
+
|
335 |
+
def create_custom_forward_attn(module, return_dict=None, use_image_num=None):
|
336 |
+
def custom_forward(*inputs):
|
337 |
+
if return_dict is not None:
|
338 |
+
return module(*inputs, return_dict=return_dict, use_image_num=use_image_num)
|
339 |
+
else:
|
340 |
+
return module(*inputs, use_image_num=use_image_num)
|
341 |
+
|
342 |
+
return custom_forward
|
343 |
+
|
344 |
+
hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
|
345 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
346 |
+
create_custom_forward_attn(attn, return_dict=False, use_image_num=use_image_num),
|
347 |
+
hidden_states,
|
348 |
+
encoder_hidden_states,
|
349 |
+
)[0]
|
350 |
+
else:
|
351 |
+
hidden_states = resnet(hidden_states, temb)
|
352 |
+
hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states, use_image_num=use_image_num).sample
|
353 |
+
|
354 |
+
output_states += (hidden_states,)
|
355 |
+
|
356 |
+
if self.downsamplers is not None:
|
357 |
+
for downsampler in self.downsamplers:
|
358 |
+
hidden_states = downsampler(hidden_states)
|
359 |
+
|
360 |
+
output_states += (hidden_states,)
|
361 |
+
|
362 |
+
return hidden_states, output_states
|
363 |
+
|
364 |
+
|
365 |
+
class DownBlock3D(nn.Module):
|
366 |
+
def __init__(
|
367 |
+
self,
|
368 |
+
in_channels: int,
|
369 |
+
out_channels: int,
|
370 |
+
temb_channels: int,
|
371 |
+
dropout: float = 0.0,
|
372 |
+
num_layers: int = 1,
|
373 |
+
resnet_eps: float = 1e-6,
|
374 |
+
resnet_time_scale_shift: str = "default",
|
375 |
+
resnet_act_fn: str = "swish",
|
376 |
+
resnet_groups: int = 32,
|
377 |
+
resnet_pre_norm: bool = True,
|
378 |
+
output_scale_factor=1.0,
|
379 |
+
add_downsample=True,
|
380 |
+
downsample_padding=1,
|
381 |
+
):
|
382 |
+
super().__init__()
|
383 |
+
resnets = []
|
384 |
+
|
385 |
+
for i in range(num_layers):
|
386 |
+
in_channels = in_channels if i == 0 else out_channels
|
387 |
+
resnets.append(
|
388 |
+
ResnetBlock3D(
|
389 |
+
in_channels=in_channels,
|
390 |
+
out_channels=out_channels,
|
391 |
+
temb_channels=temb_channels,
|
392 |
+
eps=resnet_eps,
|
393 |
+
groups=resnet_groups,
|
394 |
+
dropout=dropout,
|
395 |
+
time_embedding_norm=resnet_time_scale_shift,
|
396 |
+
non_linearity=resnet_act_fn,
|
397 |
+
output_scale_factor=output_scale_factor,
|
398 |
+
pre_norm=resnet_pre_norm,
|
399 |
+
)
|
400 |
+
)
|
401 |
+
|
402 |
+
self.resnets = nn.ModuleList(resnets)
|
403 |
+
|
404 |
+
if add_downsample:
|
405 |
+
self.downsamplers = nn.ModuleList(
|
406 |
+
[
|
407 |
+
Downsample3D(
|
408 |
+
out_channels, use_conv=True, out_channels=out_channels, padding=downsample_padding, name="op"
|
409 |
+
)
|
410 |
+
]
|
411 |
+
)
|
412 |
+
else:
|
413 |
+
self.downsamplers = None
|
414 |
+
|
415 |
+
self.gradient_checkpointing = False
|
416 |
+
|
417 |
+
def forward(self, hidden_states, temb=None):
|
418 |
+
output_states = ()
|
419 |
+
|
420 |
+
for resnet in self.resnets:
|
421 |
+
if self.training and self.gradient_checkpointing:
|
422 |
+
|
423 |
+
def create_custom_forward(module):
|
424 |
+
def custom_forward(*inputs):
|
425 |
+
return module(*inputs)
|
426 |
+
|
427 |
+
return custom_forward
|
428 |
+
|
429 |
+
hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
|
430 |
+
else:
|
431 |
+
hidden_states = resnet(hidden_states, temb)
|
432 |
+
|
433 |
+
output_states += (hidden_states,)
|
434 |
+
|
435 |
+
if self.downsamplers is not None:
|
436 |
+
for downsampler in self.downsamplers:
|
437 |
+
hidden_states = downsampler(hidden_states)
|
438 |
+
|
439 |
+
output_states += (hidden_states,)
|
440 |
+
|
441 |
+
return hidden_states, output_states
|
442 |
+
|
443 |
+
|
444 |
+
class CrossAttnUpBlock3D(nn.Module):
|
445 |
+
def __init__(
|
446 |
+
self,
|
447 |
+
in_channels: int,
|
448 |
+
out_channels: int,
|
449 |
+
prev_output_channel: int,
|
450 |
+
temb_channels: int,
|
451 |
+
dropout: float = 0.0,
|
452 |
+
num_layers: int = 1,
|
453 |
+
resnet_eps: float = 1e-6,
|
454 |
+
resnet_time_scale_shift: str = "default",
|
455 |
+
resnet_act_fn: str = "swish",
|
456 |
+
resnet_groups: int = 32,
|
457 |
+
resnet_pre_norm: bool = True,
|
458 |
+
attn_num_head_channels=1,
|
459 |
+
cross_attention_dim=1280,
|
460 |
+
output_scale_factor=1.0,
|
461 |
+
add_upsample=True,
|
462 |
+
dual_cross_attention=False,
|
463 |
+
use_linear_projection=False,
|
464 |
+
only_cross_attention=False,
|
465 |
+
upcast_attention=False,
|
466 |
+
use_first_frame=False,
|
467 |
+
use_relative_position=False,
|
468 |
+
rotary_emb=False
|
469 |
+
):
|
470 |
+
super().__init__()
|
471 |
+
resnets = []
|
472 |
+
attentions = []
|
473 |
+
|
474 |
+
self.has_cross_attention = True
|
475 |
+
self.attn_num_head_channels = attn_num_head_channels
|
476 |
+
|
477 |
+
for i in range(num_layers):
|
478 |
+
res_skip_channels = in_channels if (i == num_layers - 1) else out_channels
|
479 |
+
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
480 |
+
|
481 |
+
resnets.append(
|
482 |
+
ResnetBlock3D(
|
483 |
+
in_channels=resnet_in_channels + res_skip_channels,
|
484 |
+
out_channels=out_channels,
|
485 |
+
temb_channels=temb_channels,
|
486 |
+
eps=resnet_eps,
|
487 |
+
groups=resnet_groups,
|
488 |
+
dropout=dropout,
|
489 |
+
time_embedding_norm=resnet_time_scale_shift,
|
490 |
+
non_linearity=resnet_act_fn,
|
491 |
+
output_scale_factor=output_scale_factor,
|
492 |
+
pre_norm=resnet_pre_norm,
|
493 |
+
)
|
494 |
+
)
|
495 |
+
if dual_cross_attention:
|
496 |
+
raise NotImplementedError
|
497 |
+
attentions.append(
|
498 |
+
Transformer3DModel(
|
499 |
+
attn_num_head_channels,
|
500 |
+
out_channels // attn_num_head_channels,
|
501 |
+
in_channels=out_channels,
|
502 |
+
num_layers=1,
|
503 |
+
cross_attention_dim=cross_attention_dim,
|
504 |
+
norm_num_groups=resnet_groups,
|
505 |
+
use_linear_projection=use_linear_projection,
|
506 |
+
only_cross_attention=only_cross_attention,
|
507 |
+
upcast_attention=upcast_attention,
|
508 |
+
use_first_frame=use_first_frame,
|
509 |
+
use_relative_position=use_relative_position,
|
510 |
+
rotary_emb=rotary_emb,
|
511 |
+
)
|
512 |
+
)
|
513 |
+
|
514 |
+
self.attentions = nn.ModuleList(attentions)
|
515 |
+
self.resnets = nn.ModuleList(resnets)
|
516 |
+
|
517 |
+
if add_upsample:
|
518 |
+
self.upsamplers = nn.ModuleList([Upsample3D(out_channels, use_conv=True, out_channels=out_channels)])
|
519 |
+
else:
|
520 |
+
self.upsamplers = None
|
521 |
+
|
522 |
+
self.gradient_checkpointing = False
|
523 |
+
|
524 |
+
def forward(
|
525 |
+
self,
|
526 |
+
hidden_states,
|
527 |
+
res_hidden_states_tuple,
|
528 |
+
temb=None,
|
529 |
+
encoder_hidden_states=None,
|
530 |
+
upsample_size=None,
|
531 |
+
attention_mask=None,
|
532 |
+
use_image_num=None,
|
533 |
+
):
|
534 |
+
for resnet, attn in zip(self.resnets, self.attentions):
|
535 |
+
# pop res hidden states
|
536 |
+
res_hidden_states = res_hidden_states_tuple[-1]
|
537 |
+
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
538 |
+
hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
|
539 |
+
|
540 |
+
if self.training and self.gradient_checkpointing:
|
541 |
+
|
542 |
+
def create_custom_forward(module, return_dict=None):
|
543 |
+
def custom_forward(*inputs):
|
544 |
+
if return_dict is not None:
|
545 |
+
return module(*inputs, return_dict=return_dict)
|
546 |
+
else:
|
547 |
+
return module(*inputs)
|
548 |
+
|
549 |
+
return custom_forward
|
550 |
+
|
551 |
+
def create_custom_forward_attn(module, return_dict=None, use_image_num=None):
|
552 |
+
def custom_forward(*inputs):
|
553 |
+
if return_dict is not None:
|
554 |
+
return module(*inputs, return_dict=return_dict, use_image_num=use_image_num)
|
555 |
+
else:
|
556 |
+
return module(*inputs, use_image_num=use_image_num)
|
557 |
+
|
558 |
+
return custom_forward
|
559 |
+
|
560 |
+
hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
|
561 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
562 |
+
create_custom_forward_attn(attn, return_dict=False, use_image_num=use_image_num),
|
563 |
+
hidden_states,
|
564 |
+
encoder_hidden_states,
|
565 |
+
)[0]
|
566 |
+
else:
|
567 |
+
hidden_states = resnet(hidden_states, temb)
|
568 |
+
hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states, use_image_num=use_image_num).sample
|
569 |
+
|
570 |
+
if self.upsamplers is not None:
|
571 |
+
for upsampler in self.upsamplers:
|
572 |
+
hidden_states = upsampler(hidden_states, upsample_size)
|
573 |
+
|
574 |
+
return hidden_states
|
575 |
+
|
576 |
+
|
577 |
+
class UpBlock3D(nn.Module):
|
578 |
+
def __init__(
|
579 |
+
self,
|
580 |
+
in_channels: int,
|
581 |
+
prev_output_channel: int,
|
582 |
+
out_channels: int,
|
583 |
+
temb_channels: int,
|
584 |
+
dropout: float = 0.0,
|
585 |
+
num_layers: int = 1,
|
586 |
+
resnet_eps: float = 1e-6,
|
587 |
+
resnet_time_scale_shift: str = "default",
|
588 |
+
resnet_act_fn: str = "swish",
|
589 |
+
resnet_groups: int = 32,
|
590 |
+
resnet_pre_norm: bool = True,
|
591 |
+
output_scale_factor=1.0,
|
592 |
+
add_upsample=True,
|
593 |
+
):
|
594 |
+
super().__init__()
|
595 |
+
resnets = []
|
596 |
+
|
597 |
+
for i in range(num_layers):
|
598 |
+
res_skip_channels = in_channels if (i == num_layers - 1) else out_channels
|
599 |
+
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
600 |
+
|
601 |
+
resnets.append(
|
602 |
+
ResnetBlock3D(
|
603 |
+
in_channels=resnet_in_channels + res_skip_channels,
|
604 |
+
out_channels=out_channels,
|
605 |
+
temb_channels=temb_channels,
|
606 |
+
eps=resnet_eps,
|
607 |
+
groups=resnet_groups,
|
608 |
+
dropout=dropout,
|
609 |
+
time_embedding_norm=resnet_time_scale_shift,
|
610 |
+
non_linearity=resnet_act_fn,
|
611 |
+
output_scale_factor=output_scale_factor,
|
612 |
+
pre_norm=resnet_pre_norm,
|
613 |
+
)
|
614 |
+
)
|
615 |
+
|
616 |
+
self.resnets = nn.ModuleList(resnets)
|
617 |
+
|
618 |
+
if add_upsample:
|
619 |
+
self.upsamplers = nn.ModuleList([Upsample3D(out_channels, use_conv=True, out_channels=out_channels)])
|
620 |
+
else:
|
621 |
+
self.upsamplers = None
|
622 |
+
|
623 |
+
self.gradient_checkpointing = False
|
624 |
+
|
625 |
+
def forward(self, hidden_states, res_hidden_states_tuple, temb=None, upsample_size=None):
|
626 |
+
for resnet in self.resnets:
|
627 |
+
# pop res hidden states
|
628 |
+
res_hidden_states = res_hidden_states_tuple[-1]
|
629 |
+
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
630 |
+
hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
|
631 |
+
|
632 |
+
if self.training and self.gradient_checkpointing:
|
633 |
+
|
634 |
+
def create_custom_forward(module):
|
635 |
+
def custom_forward(*inputs):
|
636 |
+
return module(*inputs)
|
637 |
+
|
638 |
+
return custom_forward
|
639 |
+
|
640 |
+
hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
|
641 |
+
else:
|
642 |
+
hidden_states = resnet(hidden_states, temb)
|
643 |
+
|
644 |
+
if self.upsamplers is not None:
|
645 |
+
for upsampler in self.upsamplers:
|
646 |
+
hidden_states = upsampler(hidden_states, upsample_size)
|
647 |
+
|
648 |
+
return hidden_states
|
base/models/utils.py
ADDED
@@ -0,0 +1,215 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# adopted from
|
2 |
+
# https://github.com/openai/improved-diffusion/blob/main/improved_diffusion/gaussian_diffusion.py
|
3 |
+
# and
|
4 |
+
# https://github.com/lucidrains/denoising-diffusion-pytorch/blob/7706bdfc6f527f58d33f84b7b522e61e6e3164b3/denoising_diffusion_pytorch/denoising_diffusion_pytorch.py
|
5 |
+
# and
|
6 |
+
# https://github.com/openai/guided-diffusion/blob/0ba878e517b276c45d1195eb29f6f5f72659a05b/guided_diffusion/nn.py
|
7 |
+
#
|
8 |
+
# thanks!
|
9 |
+
|
10 |
+
|
11 |
+
import os
|
12 |
+
import math
|
13 |
+
import torch
|
14 |
+
|
15 |
+
import numpy as np
|
16 |
+
import torch.nn as nn
|
17 |
+
|
18 |
+
from einops import repeat
|
19 |
+
|
20 |
+
|
21 |
+
#################################################################################
|
22 |
+
# Unet Utils #
|
23 |
+
#################################################################################
|
24 |
+
|
25 |
+
def checkpoint(func, inputs, params, flag):
|
26 |
+
"""
|
27 |
+
Evaluate a function without caching intermediate activations, allowing for
|
28 |
+
reduced memory at the expense of extra compute in the backward pass.
|
29 |
+
:param func: the function to evaluate.
|
30 |
+
:param inputs: the argument sequence to pass to `func`.
|
31 |
+
:param params: a sequence of parameters `func` depends on but does not
|
32 |
+
explicitly take as arguments.
|
33 |
+
:param flag: if False, disable gradient checkpointing.
|
34 |
+
"""
|
35 |
+
if flag:
|
36 |
+
args = tuple(inputs) + tuple(params)
|
37 |
+
return CheckpointFunction.apply(func, len(inputs), *args)
|
38 |
+
else:
|
39 |
+
return func(*inputs)
|
40 |
+
|
41 |
+
|
42 |
+
class CheckpointFunction(torch.autograd.Function):
|
43 |
+
@staticmethod
|
44 |
+
def forward(ctx, run_function, length, *args):
|
45 |
+
ctx.run_function = run_function
|
46 |
+
ctx.input_tensors = list(args[:length])
|
47 |
+
ctx.input_params = list(args[length:])
|
48 |
+
|
49 |
+
with torch.no_grad():
|
50 |
+
output_tensors = ctx.run_function(*ctx.input_tensors)
|
51 |
+
return output_tensors
|
52 |
+
|
53 |
+
@staticmethod
|
54 |
+
def backward(ctx, *output_grads):
|
55 |
+
ctx.input_tensors = [x.detach().requires_grad_(True) for x in ctx.input_tensors]
|
56 |
+
with torch.enable_grad():
|
57 |
+
# Fixes a bug where the first op in run_function modifies the
|
58 |
+
# Tensor storage in place, which is not allowed for detach()'d
|
59 |
+
# Tensors.
|
60 |
+
shallow_copies = [x.view_as(x) for x in ctx.input_tensors]
|
61 |
+
output_tensors = ctx.run_function(*shallow_copies)
|
62 |
+
input_grads = torch.autograd.grad(
|
63 |
+
output_tensors,
|
64 |
+
ctx.input_tensors + ctx.input_params,
|
65 |
+
output_grads,
|
66 |
+
allow_unused=True,
|
67 |
+
)
|
68 |
+
del ctx.input_tensors
|
69 |
+
del ctx.input_params
|
70 |
+
del output_tensors
|
71 |
+
return (None, None) + input_grads
|
72 |
+
|
73 |
+
|
74 |
+
def timestep_embedding(timesteps, dim, max_period=10000, repeat_only=False):
|
75 |
+
"""
|
76 |
+
Create sinusoidal timestep embeddings.
|
77 |
+
:param timesteps: a 1-D Tensor of N indices, one per batch element.
|
78 |
+
These may be fractional.
|
79 |
+
:param dim: the dimension of the output.
|
80 |
+
:param max_period: controls the minimum frequency of the embeddings.
|
81 |
+
:return: an [N x dim] Tensor of positional embeddings.
|
82 |
+
"""
|
83 |
+
if not repeat_only:
|
84 |
+
half = dim // 2
|
85 |
+
freqs = torch.exp(
|
86 |
+
-math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half
|
87 |
+
).to(device=timesteps.device)
|
88 |
+
args = timesteps[:, None].float() * freqs[None]
|
89 |
+
embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
|
90 |
+
if dim % 2:
|
91 |
+
embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
|
92 |
+
else:
|
93 |
+
embedding = repeat(timesteps, 'b -> b d', d=dim).contiguous()
|
94 |
+
return embedding
|
95 |
+
|
96 |
+
|
97 |
+
def zero_module(module):
|
98 |
+
"""
|
99 |
+
Zero out the parameters of a module and return it.
|
100 |
+
"""
|
101 |
+
for p in module.parameters():
|
102 |
+
p.detach().zero_()
|
103 |
+
return module
|
104 |
+
|
105 |
+
|
106 |
+
def scale_module(module, scale):
|
107 |
+
"""
|
108 |
+
Scale the parameters of a module and return it.
|
109 |
+
"""
|
110 |
+
for p in module.parameters():
|
111 |
+
p.detach().mul_(scale)
|
112 |
+
return module
|
113 |
+
|
114 |
+
|
115 |
+
def mean_flat(tensor):
|
116 |
+
"""
|
117 |
+
Take the mean over all non-batch dimensions.
|
118 |
+
"""
|
119 |
+
return tensor.mean(dim=list(range(1, len(tensor.shape))))
|
120 |
+
|
121 |
+
|
122 |
+
def normalization(channels):
|
123 |
+
"""
|
124 |
+
Make a standard normalization layer.
|
125 |
+
:param channels: number of input channels.
|
126 |
+
:return: an nn.Module for normalization.
|
127 |
+
"""
|
128 |
+
return GroupNorm32(32, channels)
|
129 |
+
|
130 |
+
|
131 |
+
# PyTorch 1.7 has SiLU, but we support PyTorch 1.5.
|
132 |
+
class SiLU(nn.Module):
|
133 |
+
def forward(self, x):
|
134 |
+
return x * torch.sigmoid(x)
|
135 |
+
|
136 |
+
|
137 |
+
class GroupNorm32(nn.GroupNorm):
|
138 |
+
def forward(self, x):
|
139 |
+
return super().forward(x.float()).type(x.dtype)
|
140 |
+
|
141 |
+
def conv_nd(dims, *args, **kwargs):
|
142 |
+
"""
|
143 |
+
Create a 1D, 2D, or 3D convolution module.
|
144 |
+
"""
|
145 |
+
if dims == 1:
|
146 |
+
return nn.Conv1d(*args, **kwargs)
|
147 |
+
elif dims == 2:
|
148 |
+
return nn.Conv2d(*args, **kwargs)
|
149 |
+
elif dims == 3:
|
150 |
+
return nn.Conv3d(*args, **kwargs)
|
151 |
+
raise ValueError(f"unsupported dimensions: {dims}")
|
152 |
+
|
153 |
+
|
154 |
+
def linear(*args, **kwargs):
|
155 |
+
"""
|
156 |
+
Create a linear module.
|
157 |
+
"""
|
158 |
+
return nn.Linear(*args, **kwargs)
|
159 |
+
|
160 |
+
|
161 |
+
def avg_pool_nd(dims, *args, **kwargs):
|
162 |
+
"""
|
163 |
+
Create a 1D, 2D, or 3D average pooling module.
|
164 |
+
"""
|
165 |
+
if dims == 1:
|
166 |
+
return nn.AvgPool1d(*args, **kwargs)
|
167 |
+
elif dims == 2:
|
168 |
+
return nn.AvgPool2d(*args, **kwargs)
|
169 |
+
elif dims == 3:
|
170 |
+
return nn.AvgPool3d(*args, **kwargs)
|
171 |
+
raise ValueError(f"unsupported dimensions: {dims}")
|
172 |
+
|
173 |
+
|
174 |
+
# class HybridConditioner(nn.Module):
|
175 |
+
|
176 |
+
# def __init__(self, c_concat_config, c_crossattn_config):
|
177 |
+
# super().__init__()
|
178 |
+
# self.concat_conditioner = instantiate_from_config(c_concat_config)
|
179 |
+
# self.crossattn_conditioner = instantiate_from_config(c_crossattn_config)
|
180 |
+
|
181 |
+
# def forward(self, c_concat, c_crossattn):
|
182 |
+
# c_concat = self.concat_conditioner(c_concat)
|
183 |
+
# c_crossattn = self.crossattn_conditioner(c_crossattn)
|
184 |
+
# return {'c_concat': [c_concat], 'c_crossattn': [c_crossattn]}
|
185 |
+
|
186 |
+
|
187 |
+
def noise_like(shape, device, repeat=False):
|
188 |
+
repeat_noise = lambda: torch.randn((1, *shape[1:]), device=device).repeat(shape[0], *((1,) * (len(shape) - 1)))
|
189 |
+
noise = lambda: torch.randn(shape, device=device)
|
190 |
+
return repeat_noise() if repeat else noise()
|
191 |
+
|
192 |
+
def count_flops_attn(model, _x, y):
|
193 |
+
"""
|
194 |
+
A counter for the `thop` package to count the operations in an
|
195 |
+
attention operation.
|
196 |
+
Meant to be used like:
|
197 |
+
macs, params = thop.profile(
|
198 |
+
model,
|
199 |
+
inputs=(inputs, timestamps),
|
200 |
+
custom_ops={QKVAttention: QKVAttention.count_flops},
|
201 |
+
)
|
202 |
+
"""
|
203 |
+
b, c, *spatial = y[0].shape
|
204 |
+
num_spatial = int(np.prod(spatial))
|
205 |
+
# We perform two matmuls with the same number of ops.
|
206 |
+
# The first computes the weight matrix, the second computes
|
207 |
+
# the combination of the value vectors.
|
208 |
+
matmul_ops = 2 * b * (num_spatial ** 2) * c
|
209 |
+
model.total_ops += torch.DoubleTensor([matmul_ops])
|
210 |
+
|
211 |
+
def count_params(model, verbose=False):
|
212 |
+
total_params = sum(p.numel() for p in model.parameters())
|
213 |
+
if verbose:
|
214 |
+
print(f"{model.__class__.__name__} has {total_params * 1.e-6:.2f} M params.")
|
215 |
+
return total_params
|
base/pipelines/__pycache__/pipeline_videogen.cpython-311.pyc
ADDED
Binary file (34.9 kB). View file
|
|
base/pipelines/pipeline_videogen.py
ADDED
@@ -0,0 +1,677 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
3 |
+
# you may not use this file except in compliance with the License.
|
4 |
+
# You may obtain a copy of the License at
|
5 |
+
#
|
6 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
7 |
+
#
|
8 |
+
# Unless required by applicable law or agreed to in writing, software
|
9 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
10 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
11 |
+
# See the License for the specific language governing permissions and
|
12 |
+
# limitations under the License.
|
13 |
+
|
14 |
+
import inspect
|
15 |
+
from typing import Any, Callable, Dict, List, Optional, Union
|
16 |
+
import einops
|
17 |
+
import torch
|
18 |
+
from packaging import version
|
19 |
+
from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
|
20 |
+
|
21 |
+
from diffusers.configuration_utils import FrozenDict
|
22 |
+
from diffusers.models import AutoencoderKL
|
23 |
+
from diffusers.schedulers import KarrasDiffusionSchedulers
|
24 |
+
from diffusers.utils import (
|
25 |
+
deprecate,
|
26 |
+
is_accelerate_available,
|
27 |
+
is_accelerate_version,
|
28 |
+
logging,
|
29 |
+
#randn_tensor,
|
30 |
+
replace_example_docstring,
|
31 |
+
BaseOutput,
|
32 |
+
)
|
33 |
+
|
34 |
+
try:
|
35 |
+
from diffusers.utils import randn_tensor
|
36 |
+
except:
|
37 |
+
from diffusers.utils.torch_utils import randn_tensor
|
38 |
+
|
39 |
+
|
40 |
+
from diffusers.pipeline_utils import DiffusionPipeline
|
41 |
+
from dataclasses import dataclass
|
42 |
+
|
43 |
+
import os, sys
|
44 |
+
sys.path.append(os.path.split(sys.path[0])[0])
|
45 |
+
from models.unet import UNet3DConditionModel
|
46 |
+
|
47 |
+
import numpy as np
|
48 |
+
|
49 |
+
@dataclass
|
50 |
+
class StableDiffusionPipelineOutput(BaseOutput):
|
51 |
+
video: torch.Tensor
|
52 |
+
|
53 |
+
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
|
54 |
+
|
55 |
+
EXAMPLE_DOC_STRING = """
|
56 |
+
Examples:
|
57 |
+
```py
|
58 |
+
>>> import torch
|
59 |
+
>>> from diffusers import StableDiffusionPipeline
|
60 |
+
|
61 |
+
>>> pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16)
|
62 |
+
>>> pipe = pipe.to("cuda")
|
63 |
+
|
64 |
+
>>> prompt = "a photo of an astronaut riding a horse on mars"
|
65 |
+
>>> image = pipe(prompt).images[0]
|
66 |
+
```
|
67 |
+
"""
|
68 |
+
|
69 |
+
|
70 |
+
class VideoGenPipeline(DiffusionPipeline):
|
71 |
+
r"""
|
72 |
+
Pipeline for text-to-image generation using Stable Diffusion.
|
73 |
+
|
74 |
+
This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
|
75 |
+
library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
|
76 |
+
|
77 |
+
Args:
|
78 |
+
vae ([`AutoencoderKL`]):
|
79 |
+
Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
|
80 |
+
text_encoder ([`CLIPTextModel`]):
|
81 |
+
Frozen text-encoder. Stable Diffusion uses the text portion of
|
82 |
+
[CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
|
83 |
+
the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
|
84 |
+
tokenizer (`CLIPTokenizer`):
|
85 |
+
Tokenizer of class
|
86 |
+
[CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
|
87 |
+
unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
|
88 |
+
scheduler ([`SchedulerMixin`]):
|
89 |
+
A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
|
90 |
+
[`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
|
91 |
+
safety_checker ([`StableDiffusionSafetyChecker`]):
|
92 |
+
Classification module that estimates whether generated images could be considered offensive or harmful.
|
93 |
+
Please, refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for details.
|
94 |
+
feature_extractor ([`CLIPFeatureExtractor`]):
|
95 |
+
Model that extracts features from generated images to be used as inputs for the `safety_checker`.
|
96 |
+
"""
|
97 |
+
_optional_components = ["safety_checker", "feature_extractor"]
|
98 |
+
|
99 |
+
def __init__(
|
100 |
+
self,
|
101 |
+
vae: AutoencoderKL,
|
102 |
+
text_encoder: CLIPTextModel,
|
103 |
+
tokenizer: CLIPTokenizer,
|
104 |
+
unet: UNet3DConditionModel,
|
105 |
+
scheduler: KarrasDiffusionSchedulers,
|
106 |
+
):
|
107 |
+
super().__init__()
|
108 |
+
|
109 |
+
if hasattr(scheduler.config, "steps_offset") and scheduler.config.steps_offset != 1:
|
110 |
+
deprecation_message = (
|
111 |
+
f"The configuration file of this scheduler: {scheduler} is outdated. `steps_offset`"
|
112 |
+
f" should be set to 1 instead of {scheduler.config.steps_offset}. Please make sure "
|
113 |
+
"to update the config accordingly as leaving `steps_offset` might led to incorrect results"
|
114 |
+
" in future versions. If you have downloaded this checkpoint from the Hugging Face Hub,"
|
115 |
+
" it would be very nice if you could open a Pull request for the `scheduler/scheduler_config.json`"
|
116 |
+
" file"
|
117 |
+
)
|
118 |
+
deprecate("steps_offset!=1", "1.0.0", deprecation_message, standard_warn=False)
|
119 |
+
new_config = dict(scheduler.config)
|
120 |
+
new_config["steps_offset"] = 1
|
121 |
+
scheduler._internal_dict = FrozenDict(new_config)
|
122 |
+
|
123 |
+
if hasattr(scheduler.config, "clip_sample") and scheduler.config.clip_sample is True:
|
124 |
+
deprecation_message = (
|
125 |
+
f"The configuration file of this scheduler: {scheduler} has not set the configuration `clip_sample`."
|
126 |
+
" `clip_sample` should be set to False in the configuration file. Please make sure to update the"
|
127 |
+
" config accordingly as not setting `clip_sample` in the config might lead to incorrect results in"
|
128 |
+
" future versions. If you have downloaded this checkpoint from the Hugging Face Hub, it would be very"
|
129 |
+
" nice if you could open a Pull request for the `scheduler/scheduler_config.json` file"
|
130 |
+
)
|
131 |
+
deprecate("clip_sample not set", "1.0.0", deprecation_message, standard_warn=False)
|
132 |
+
new_config = dict(scheduler.config)
|
133 |
+
new_config["clip_sample"] = False
|
134 |
+
scheduler._internal_dict = FrozenDict(new_config)
|
135 |
+
|
136 |
+
|
137 |
+
|
138 |
+
is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
|
139 |
+
version.parse(unet.config._diffusers_version).base_version
|
140 |
+
) < version.parse("0.9.0.dev0")
|
141 |
+
is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
|
142 |
+
if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
|
143 |
+
deprecation_message = (
|
144 |
+
"The configuration file of the unet has set the default `sample_size` to smaller than"
|
145 |
+
" 64 which seems highly unlikely. If your checkpoint is a fine-tuned version of any of the"
|
146 |
+
" following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-"
|
147 |
+
" CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5"
|
148 |
+
" \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the"
|
149 |
+
" configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`"
|
150 |
+
" in the config might lead to incorrect results in future versions. If you have downloaded this"
|
151 |
+
" checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for"
|
152 |
+
" the `unet/config.json` file"
|
153 |
+
)
|
154 |
+
deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False)
|
155 |
+
new_config = dict(unet.config)
|
156 |
+
new_config["sample_size"] = 64
|
157 |
+
unet._internal_dict = FrozenDict(new_config)
|
158 |
+
|
159 |
+
self.register_modules(
|
160 |
+
vae=vae,
|
161 |
+
text_encoder=text_encoder,
|
162 |
+
tokenizer=tokenizer,
|
163 |
+
unet=unet,
|
164 |
+
scheduler=scheduler,
|
165 |
+
)
|
166 |
+
self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
|
167 |
+
# self.register_to_config(requires_safety_checker=requires_safety_checker)
|
168 |
+
|
169 |
+
def enable_vae_slicing(self):
|
170 |
+
r"""
|
171 |
+
Enable sliced VAE decoding.
|
172 |
+
|
173 |
+
When this option is enabled, the VAE will split the input tensor in slices to compute decoding in several
|
174 |
+
steps. This is useful to save some memory and allow larger batch sizes.
|
175 |
+
"""
|
176 |
+
self.vae.enable_slicing()
|
177 |
+
|
178 |
+
def disable_vae_slicing(self):
|
179 |
+
r"""
|
180 |
+
Disable sliced VAE decoding. If `enable_vae_slicing` was previously invoked, this method will go back to
|
181 |
+
computing decoding in one step.
|
182 |
+
"""
|
183 |
+
self.vae.disable_slicing()
|
184 |
+
|
185 |
+
def enable_vae_tiling(self):
|
186 |
+
r"""
|
187 |
+
Enable tiled VAE decoding.
|
188 |
+
|
189 |
+
When this option is enabled, the VAE will split the input tensor into tiles to compute decoding and encoding in
|
190 |
+
several steps. This is useful to save a large amount of memory and to allow the processing of larger images.
|
191 |
+
"""
|
192 |
+
self.vae.enable_tiling()
|
193 |
+
|
194 |
+
def disable_vae_tiling(self):
|
195 |
+
r"""
|
196 |
+
Disable tiled VAE decoding. If `enable_vae_tiling` was previously invoked, this method will go back to
|
197 |
+
computing decoding in one step.
|
198 |
+
"""
|
199 |
+
self.vae.disable_tiling()
|
200 |
+
|
201 |
+
def enable_sequential_cpu_offload(self, gpu_id=0):
|
202 |
+
r"""
|
203 |
+
Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
|
204 |
+
text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
|
205 |
+
`torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
|
206 |
+
Note that offloading happens on a submodule basis. Memory savings are higher than with
|
207 |
+
`enable_model_cpu_offload`, but performance is lower.
|
208 |
+
"""
|
209 |
+
if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
|
210 |
+
from accelerate import cpu_offload
|
211 |
+
else:
|
212 |
+
raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
|
213 |
+
|
214 |
+
device = torch.device(f"cuda:{gpu_id}")
|
215 |
+
|
216 |
+
if self.device.type != "cpu":
|
217 |
+
self.to("cpu", silence_dtype_warnings=True)
|
218 |
+
torch.cuda.empty_cache() # otherwise we don't see the memory savings (but they probably exist)
|
219 |
+
|
220 |
+
for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
|
221 |
+
cpu_offload(cpu_offloaded_model, device)
|
222 |
+
|
223 |
+
# if self.safety_checker is not None:
|
224 |
+
# cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
|
225 |
+
|
226 |
+
def enable_model_cpu_offload(self, gpu_id=0):
|
227 |
+
r"""
|
228 |
+
Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
|
229 |
+
to `enable_sequential_cpu_offload`, this method moves one whole model at a time to the GPU when its `forward`
|
230 |
+
method is called, and the model remains in GPU until the next model runs. Memory savings are lower than with
|
231 |
+
`enable_sequential_cpu_offload`, but performance is much better due to the iterative execution of the `unet`.
|
232 |
+
"""
|
233 |
+
if is_accelerate_available() and is_accelerate_version(">=", "0.17.0.dev0"):
|
234 |
+
from accelerate import cpu_offload_with_hook
|
235 |
+
else:
|
236 |
+
raise ImportError("`enable_model_offload` requires `accelerate v0.17.0` or higher.")
|
237 |
+
|
238 |
+
device = torch.device(f"cuda:{gpu_id}")
|
239 |
+
|
240 |
+
if self.device.type != "cpu":
|
241 |
+
self.to("cpu", silence_dtype_warnings=True)
|
242 |
+
torch.cuda.empty_cache() # otherwise we don't see the memory savings (but they probably exist)
|
243 |
+
|
244 |
+
hook = None
|
245 |
+
for cpu_offloaded_model in [self.text_encoder, self.unet, self.vae]:
|
246 |
+
_, hook = cpu_offload_with_hook(cpu_offloaded_model, device, prev_module_hook=hook)
|
247 |
+
|
248 |
+
self.final_offload_hook = hook
|
249 |
+
|
250 |
+
@property
|
251 |
+
def _execution_device(self):
|
252 |
+
r"""
|
253 |
+
Returns the device on which the pipeline's models will be executed. After calling
|
254 |
+
`pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
|
255 |
+
hooks.
|
256 |
+
"""
|
257 |
+
if not hasattr(self.unet, "_hf_hook"):
|
258 |
+
return self.device
|
259 |
+
for module in self.unet.modules():
|
260 |
+
if (
|
261 |
+
hasattr(module, "_hf_hook")
|
262 |
+
and hasattr(module._hf_hook, "execution_device")
|
263 |
+
and module._hf_hook.execution_device is not None
|
264 |
+
):
|
265 |
+
return torch.device(module._hf_hook.execution_device)
|
266 |
+
return self.device
|
267 |
+
|
268 |
+
def _encode_prompt(
|
269 |
+
self,
|
270 |
+
prompt,
|
271 |
+
device,
|
272 |
+
num_images_per_prompt,
|
273 |
+
do_classifier_free_guidance,
|
274 |
+
negative_prompt=None,
|
275 |
+
prompt_embeds: Optional[torch.FloatTensor] = None,
|
276 |
+
negative_prompt_embeds: Optional[torch.FloatTensor] = None,
|
277 |
+
):
|
278 |
+
r"""
|
279 |
+
Encodes the prompt into text encoder hidden states.
|
280 |
+
|
281 |
+
Args:
|
282 |
+
prompt (`str` or `List[str]`, *optional*):
|
283 |
+
prompt to be encoded
|
284 |
+
device: (`torch.device`):
|
285 |
+
torch device
|
286 |
+
num_images_per_prompt (`int`):
|
287 |
+
number of images that should be generated per prompt
|
288 |
+
do_classifier_free_guidance (`bool`):
|
289 |
+
whether to use classifier free guidance or not
|
290 |
+
negative_prompt (`str` or `List[str]`, *optional*):
|
291 |
+
The prompt or prompts not to guide the image generation. If not defined, one has to pass
|
292 |
+
`negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead.
|
293 |
+
Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
|
294 |
+
prompt_embeds (`torch.FloatTensor`, *optional*):
|
295 |
+
Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
|
296 |
+
provided, text embeddings will be generated from `prompt` input argument.
|
297 |
+
negative_prompt_embeds (`torch.FloatTensor`, *optional*):
|
298 |
+
Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
|
299 |
+
weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
|
300 |
+
argument.
|
301 |
+
"""
|
302 |
+
if prompt is not None and isinstance(prompt, str):
|
303 |
+
batch_size = 1
|
304 |
+
elif prompt is not None and isinstance(prompt, list):
|
305 |
+
batch_size = len(prompt)
|
306 |
+
else:
|
307 |
+
batch_size = prompt_embeds.shape[0]
|
308 |
+
|
309 |
+
if prompt_embeds is None:
|
310 |
+
text_inputs = self.tokenizer(
|
311 |
+
prompt,
|
312 |
+
padding="max_length",
|
313 |
+
max_length=self.tokenizer.model_max_length,
|
314 |
+
truncation=True,
|
315 |
+
return_tensors="pt",
|
316 |
+
)
|
317 |
+
text_input_ids = text_inputs.input_ids
|
318 |
+
untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
|
319 |
+
|
320 |
+
if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
|
321 |
+
text_input_ids, untruncated_ids
|
322 |
+
):
|
323 |
+
removed_text = self.tokenizer.batch_decode(
|
324 |
+
untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
|
325 |
+
)
|
326 |
+
logger.warning(
|
327 |
+
"The following part of your input was truncated because CLIP can only handle sequences up to"
|
328 |
+
f" {self.tokenizer.model_max_length} tokens: {removed_text}"
|
329 |
+
)
|
330 |
+
|
331 |
+
if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
|
332 |
+
attention_mask = text_inputs.attention_mask.to(device)
|
333 |
+
else:
|
334 |
+
attention_mask = None
|
335 |
+
|
336 |
+
prompt_embeds = self.text_encoder(
|
337 |
+
text_input_ids.to(device),
|
338 |
+
attention_mask=attention_mask,
|
339 |
+
)
|
340 |
+
prompt_embeds = prompt_embeds[0]
|
341 |
+
|
342 |
+
prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)
|
343 |
+
|
344 |
+
bs_embed, seq_len, _ = prompt_embeds.shape
|
345 |
+
# duplicate text embeddings for each generation per prompt, using mps friendly method
|
346 |
+
prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
|
347 |
+
prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
|
348 |
+
|
349 |
+
# get unconditional embeddings for classifier free guidance
|
350 |
+
if do_classifier_free_guidance and negative_prompt_embeds is None:
|
351 |
+
uncond_tokens: List[str]
|
352 |
+
if negative_prompt is None:
|
353 |
+
uncond_tokens = [""] * batch_size
|
354 |
+
elif type(prompt) is not type(negative_prompt):
|
355 |
+
raise TypeError(
|
356 |
+
f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
|
357 |
+
f" {type(prompt)}."
|
358 |
+
)
|
359 |
+
elif isinstance(negative_prompt, str):
|
360 |
+
uncond_tokens = [negative_prompt]
|
361 |
+
elif batch_size != len(negative_prompt):
|
362 |
+
raise ValueError(
|
363 |
+
f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
|
364 |
+
f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
|
365 |
+
" the batch size of `prompt`."
|
366 |
+
)
|
367 |
+
else:
|
368 |
+
uncond_tokens = negative_prompt
|
369 |
+
|
370 |
+
max_length = prompt_embeds.shape[1]
|
371 |
+
uncond_input = self.tokenizer(
|
372 |
+
uncond_tokens,
|
373 |
+
padding="max_length",
|
374 |
+
max_length=max_length,
|
375 |
+
truncation=True,
|
376 |
+
return_tensors="pt",
|
377 |
+
)
|
378 |
+
|
379 |
+
if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
|
380 |
+
attention_mask = uncond_input.attention_mask.to(device)
|
381 |
+
else:
|
382 |
+
attention_mask = None
|
383 |
+
|
384 |
+
negative_prompt_embeds = self.text_encoder(
|
385 |
+
uncond_input.input_ids.to(device),
|
386 |
+
attention_mask=attention_mask,
|
387 |
+
)
|
388 |
+
negative_prompt_embeds = negative_prompt_embeds[0]
|
389 |
+
|
390 |
+
if do_classifier_free_guidance:
|
391 |
+
# duplicate unconditional embeddings for each generation per prompt, using mps friendly method
|
392 |
+
seq_len = negative_prompt_embeds.shape[1]
|
393 |
+
|
394 |
+
negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)
|
395 |
+
|
396 |
+
negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
|
397 |
+
negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
|
398 |
+
|
399 |
+
# For classifier free guidance, we need to do two forward passes.
|
400 |
+
# Here we concatenate the unconditional and text embeddings into a single batch
|
401 |
+
# to avoid doing two forward passes
|
402 |
+
prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
|
403 |
+
|
404 |
+
return prompt_embeds
|
405 |
+
|
406 |
+
def decode_latents(self, latents):
|
407 |
+
video_length = latents.shape[2]
|
408 |
+
latents = 1 / 0.18215 * latents
|
409 |
+
latents = einops.rearrange(latents, "b c f h w -> (b f) c h w")
|
410 |
+
video = self.vae.decode(latents).sample
|
411 |
+
video = einops.rearrange(video, "(b f) c h w -> b f h w c", f=video_length)
|
412 |
+
video = ((video / 2 + 0.5) * 255).add_(0.5).clamp_(0, 255).to(dtype=torch.uint8).cpu().contiguous()
|
413 |
+
return video
|
414 |
+
|
415 |
+
def prepare_extra_step_kwargs(self, generator, eta):
|
416 |
+
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
|
417 |
+
# eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
|
418 |
+
# eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
|
419 |
+
# and should be between [0, 1]
|
420 |
+
|
421 |
+
accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
|
422 |
+
extra_step_kwargs = {}
|
423 |
+
if accepts_eta:
|
424 |
+
extra_step_kwargs["eta"] = eta
|
425 |
+
|
426 |
+
# check if the scheduler accepts generator
|
427 |
+
accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
|
428 |
+
if accepts_generator:
|
429 |
+
extra_step_kwargs["generator"] = generator
|
430 |
+
return extra_step_kwargs
|
431 |
+
|
432 |
+
def check_inputs(
|
433 |
+
self,
|
434 |
+
prompt,
|
435 |
+
height,
|
436 |
+
width,
|
437 |
+
callback_steps,
|
438 |
+
negative_prompt=None,
|
439 |
+
prompt_embeds=None,
|
440 |
+
negative_prompt_embeds=None,
|
441 |
+
):
|
442 |
+
if height % 8 != 0 or width % 8 != 0:
|
443 |
+
raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
|
444 |
+
|
445 |
+
if (callback_steps is None) or (
|
446 |
+
callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
|
447 |
+
):
|
448 |
+
raise ValueError(
|
449 |
+
f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
|
450 |
+
f" {type(callback_steps)}."
|
451 |
+
)
|
452 |
+
|
453 |
+
if prompt is not None and prompt_embeds is not None:
|
454 |
+
raise ValueError(
|
455 |
+
f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
|
456 |
+
" only forward one of the two."
|
457 |
+
)
|
458 |
+
elif prompt is None and prompt_embeds is None:
|
459 |
+
raise ValueError(
|
460 |
+
"Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
|
461 |
+
)
|
462 |
+
elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
|
463 |
+
raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
|
464 |
+
|
465 |
+
if negative_prompt is not None and negative_prompt_embeds is not None:
|
466 |
+
raise ValueError(
|
467 |
+
f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
|
468 |
+
f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
|
469 |
+
)
|
470 |
+
|
471 |
+
if prompt_embeds is not None and negative_prompt_embeds is not None:
|
472 |
+
if prompt_embeds.shape != negative_prompt_embeds.shape:
|
473 |
+
raise ValueError(
|
474 |
+
"`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
|
475 |
+
f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
|
476 |
+
f" {negative_prompt_embeds.shape}."
|
477 |
+
)
|
478 |
+
|
479 |
+
def prepare_latents(self, batch_size, num_channels_latents, video_length, height, width, dtype, device, generator, latents=None):
|
480 |
+
shape = (batch_size, num_channels_latents, video_length, height // self.vae_scale_factor, width // self.vae_scale_factor)
|
481 |
+
if isinstance(generator, list) and len(generator) != batch_size:
|
482 |
+
raise ValueError(
|
483 |
+
f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
|
484 |
+
f" size of {batch_size}. Make sure the batch size matches the length of the generators."
|
485 |
+
)
|
486 |
+
|
487 |
+
if latents is None:
|
488 |
+
latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
|
489 |
+
else:
|
490 |
+
latents = latents.to(device)
|
491 |
+
|
492 |
+
# scale the initial noise by the standard deviation required by the scheduler
|
493 |
+
latents = latents * self.scheduler.init_noise_sigma
|
494 |
+
return latents
|
495 |
+
|
496 |
+
@torch.no_grad()
|
497 |
+
@replace_example_docstring(EXAMPLE_DOC_STRING)
|
498 |
+
def __call__(
|
499 |
+
self,
|
500 |
+
prompt: Union[str, List[str]] = None,
|
501 |
+
height: Optional[int] = None,
|
502 |
+
width: Optional[int] = None,
|
503 |
+
video_length: int = 16,
|
504 |
+
num_inference_steps: int = 50,
|
505 |
+
guidance_scale: float = 7.5,
|
506 |
+
negative_prompt: Optional[Union[str, List[str]]] = None,
|
507 |
+
num_images_per_prompt: Optional[int] = 1,
|
508 |
+
eta: float = 0.0,
|
509 |
+
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
|
510 |
+
latents: Optional[torch.FloatTensor] = None,
|
511 |
+
prompt_embeds: Optional[torch.FloatTensor] = None,
|
512 |
+
negative_prompt_embeds: Optional[torch.FloatTensor] = None,
|
513 |
+
output_type: Optional[str] = "pil",
|
514 |
+
return_dict: bool = True,
|
515 |
+
callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
|
516 |
+
callback_steps: int = 1,
|
517 |
+
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
|
518 |
+
):
|
519 |
+
r"""
|
520 |
+
Function invoked when calling the pipeline for generation.
|
521 |
+
|
522 |
+
Args:
|
523 |
+
prompt (`str` or `List[str]`, *optional*):
|
524 |
+
The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
|
525 |
+
instead.
|
526 |
+
height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
|
527 |
+
The height in pixels of the generated image.
|
528 |
+
width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
|
529 |
+
The width in pixels of the generated image.
|
530 |
+
num_inference_steps (`int`, *optional*, defaults to 50):
|
531 |
+
The number of denoising steps. More denoising steps usually lead to a higher quality image at the
|
532 |
+
expense of slower inference.
|
533 |
+
guidance_scale (`float`, *optional*, defaults to 7.5):
|
534 |
+
Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
|
535 |
+
`guidance_scale` is defined as `w` of equation 2. of [Imagen
|
536 |
+
Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
|
537 |
+
1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
|
538 |
+
usually at the expense of lower image quality.
|
539 |
+
negative_prompt (`str` or `List[str]`, *optional*):
|
540 |
+
The prompt or prompts not to guide the image generation. If not defined, one has to pass
|
541 |
+
`negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead.
|
542 |
+
Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
|
543 |
+
num_images_per_prompt (`int`, *optional*, defaults to 1):
|
544 |
+
The number of images to generate per prompt.
|
545 |
+
eta (`float`, *optional*, defaults to 0.0):
|
546 |
+
Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
|
547 |
+
[`schedulers.DDIMScheduler`], will be ignored for others.
|
548 |
+
generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
|
549 |
+
One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
|
550 |
+
to make generation deterministic.
|
551 |
+
latents (`torch.FloatTensor`, *optional*):
|
552 |
+
Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
|
553 |
+
generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
|
554 |
+
tensor will ge generated by sampling using the supplied random `generator`.
|
555 |
+
prompt_embeds (`torch.FloatTensor`, *optional*):
|
556 |
+
Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
|
557 |
+
provided, text embeddings will be generated from `prompt` input argument.
|
558 |
+
negative_prompt_embeds (`torch.FloatTensor`, *optional*):
|
559 |
+
Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
|
560 |
+
weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
|
561 |
+
argument.
|
562 |
+
output_type (`str`, *optional*, defaults to `"pil"`):
|
563 |
+
The output format of the generate image. Choose between
|
564 |
+
[PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
|
565 |
+
return_dict (`bool`, *optional*, defaults to `True`):
|
566 |
+
Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
|
567 |
+
plain tuple.
|
568 |
+
callback (`Callable`, *optional*):
|
569 |
+
A function that will be called every `callback_steps` steps during inference. The function will be
|
570 |
+
called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
|
571 |
+
callback_steps (`int`, *optional*, defaults to 1):
|
572 |
+
The frequency at which the `callback` function will be called. If not specified, the callback will be
|
573 |
+
called at every step.
|
574 |
+
cross_attention_kwargs (`dict`, *optional*):
|
575 |
+
A kwargs dictionary that if specified is passed along to the `AttnProcessor` as defined under
|
576 |
+
`self.processor` in
|
577 |
+
[diffusers.cross_attention](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/cross_attention.py).
|
578 |
+
|
579 |
+
Examples:
|
580 |
+
|
581 |
+
Returns:
|
582 |
+
[`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
|
583 |
+
[`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
|
584 |
+
When returning a tuple, the first element is a list with the generated images, and the second element is a
|
585 |
+
list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
|
586 |
+
(nsfw) content, according to the `safety_checker`.
|
587 |
+
"""
|
588 |
+
# 0. Default height and width to unet
|
589 |
+
height = height or self.unet.config.sample_size * self.vae_scale_factor
|
590 |
+
width = width or self.unet.config.sample_size * self.vae_scale_factor
|
591 |
+
|
592 |
+
# 1. Check inputs. Raise error if not correct
|
593 |
+
self.check_inputs(
|
594 |
+
prompt, height, width, callback_steps, negative_prompt, prompt_embeds, negative_prompt_embeds
|
595 |
+
)
|
596 |
+
|
597 |
+
# 2. Define call parameters
|
598 |
+
if prompt is not None and isinstance(prompt, str):
|
599 |
+
batch_size = 1
|
600 |
+
elif prompt is not None and isinstance(prompt, list):
|
601 |
+
batch_size = len(prompt)
|
602 |
+
else:
|
603 |
+
batch_size = prompt_embeds.shape[0]
|
604 |
+
|
605 |
+
device = self._execution_device
|
606 |
+
# here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
|
607 |
+
# of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
|
608 |
+
# corresponds to doing no classifier free guidance.
|
609 |
+
do_classifier_free_guidance = guidance_scale > 1.0
|
610 |
+
|
611 |
+
# 3. Encode input prompt
|
612 |
+
prompt_embeds = self._encode_prompt(
|
613 |
+
prompt,
|
614 |
+
device,
|
615 |
+
num_images_per_prompt,
|
616 |
+
do_classifier_free_guidance,
|
617 |
+
negative_prompt,
|
618 |
+
prompt_embeds=prompt_embeds,
|
619 |
+
negative_prompt_embeds=negative_prompt_embeds,
|
620 |
+
)
|
621 |
+
|
622 |
+
# 4. Prepare timesteps
|
623 |
+
self.scheduler.set_timesteps(num_inference_steps, device=device)
|
624 |
+
timesteps = self.scheduler.timesteps
|
625 |
+
|
626 |
+
# 5. Prepare latent variables
|
627 |
+
num_channels_latents = self.unet.config.in_channels
|
628 |
+
latents = self.prepare_latents(
|
629 |
+
batch_size * num_images_per_prompt,
|
630 |
+
num_channels_latents,
|
631 |
+
video_length,
|
632 |
+
height,
|
633 |
+
width,
|
634 |
+
prompt_embeds.dtype,
|
635 |
+
device,
|
636 |
+
generator,
|
637 |
+
latents,
|
638 |
+
)
|
639 |
+
|
640 |
+
# 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
|
641 |
+
extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
|
642 |
+
|
643 |
+
# 7. Denoising loop
|
644 |
+
num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
|
645 |
+
with self.progress_bar(total=num_inference_steps) as progress_bar:
|
646 |
+
for i, t in enumerate(timesteps):
|
647 |
+
# expand the latents if we are doing classifier free guidance
|
648 |
+
latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
|
649 |
+
latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
|
650 |
+
|
651 |
+
# predict the noise residual
|
652 |
+
noise_pred = self.unet(
|
653 |
+
latent_model_input,
|
654 |
+
t,
|
655 |
+
encoder_hidden_states=prompt_embeds,
|
656 |
+
# cross_attention_kwargs=cross_attention_kwargs,
|
657 |
+
).sample
|
658 |
+
|
659 |
+
# perform guidance
|
660 |
+
if do_classifier_free_guidance:
|
661 |
+
noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
|
662 |
+
noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
|
663 |
+
|
664 |
+
# compute the previous noisy sample x_t -> x_t-1
|
665 |
+
latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
|
666 |
+
|
667 |
+
# call the callback, if provided
|
668 |
+
if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
|
669 |
+
progress_bar.update()
|
670 |
+
if callback is not None and i % callback_steps == 0:
|
671 |
+
callback(i, t, latents)
|
672 |
+
|
673 |
+
|
674 |
+
# 8. Post-processing
|
675 |
+
video = self.decode_latents(latents)
|
676 |
+
|
677 |
+
return StableDiffusionPipelineOutput(video=video)
|
base/pipelines/sample.py
ADDED
@@ -0,0 +1,88 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import torch
|
3 |
+
import argparse
|
4 |
+
import torchvision
|
5 |
+
|
6 |
+
from pipeline_videogen import VideoGenPipeline
|
7 |
+
|
8 |
+
from download import find_model
|
9 |
+
from diffusers.schedulers import DDIMScheduler, DDPMScheduler, PNDMScheduler, EulerDiscreteScheduler
|
10 |
+
from diffusers.models import AutoencoderKL
|
11 |
+
from transformers import CLIPTokenizer, CLIPTextModel, CLIPTextModelWithProjection
|
12 |
+
from omegaconf import OmegaConf
|
13 |
+
|
14 |
+
import os, sys
|
15 |
+
sys.path.append(os.path.split(sys.path[0])[0])
|
16 |
+
from models import get_models
|
17 |
+
import imageio
|
18 |
+
|
19 |
+
def main(args):
|
20 |
+
#torch.manual_seed(args.seed)
|
21 |
+
torch.set_grad_enabled(False)
|
22 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
23 |
+
|
24 |
+
sd_path = args.pretrained_path + "/stable-diffusion-v1-4"
|
25 |
+
unet = get_models(args, sd_path).to(device, dtype=torch.float16)
|
26 |
+
state_dict = find_model(args.pretrained_path + "/lavie_base.pt")
|
27 |
+
unet.load_state_dict(state_dict)
|
28 |
+
|
29 |
+
vae = AutoencoderKL.from_pretrained(sd_path, subfolder="vae", torch_dtype=torch.float16).to(device)
|
30 |
+
tokenizer_one = CLIPTokenizer.from_pretrained(sd_path, subfolder="tokenizer")
|
31 |
+
text_encoder_one = CLIPTextModel.from_pretrained(sd_path, subfolder="text_encoder", torch_dtype=torch.float16).to(device) # huge
|
32 |
+
|
33 |
+
# set eval mode
|
34 |
+
unet.eval()
|
35 |
+
vae.eval()
|
36 |
+
text_encoder_one.eval()
|
37 |
+
|
38 |
+
if args.sample_method == 'ddim':
|
39 |
+
scheduler = DDIMScheduler.from_pretrained(sd_path,
|
40 |
+
subfolder="scheduler",
|
41 |
+
beta_start=args.beta_start,
|
42 |
+
beta_end=args.beta_end,
|
43 |
+
beta_schedule=args.beta_schedule)
|
44 |
+
elif args.sample_method == 'eulerdiscrete':
|
45 |
+
scheduler = EulerDiscreteScheduler.from_pretrained(sd_path,
|
46 |
+
subfolder="scheduler",
|
47 |
+
beta_start=args.beta_start,
|
48 |
+
beta_end=args.beta_end,
|
49 |
+
beta_schedule=args.beta_schedule)
|
50 |
+
elif args.sample_method == 'ddpm':
|
51 |
+
scheduler = DDPMScheduler.from_pretrained(sd_path,
|
52 |
+
subfolder="scheduler",
|
53 |
+
beta_start=args.beta_start,
|
54 |
+
beta_end=args.beta_end,
|
55 |
+
beta_schedule=args.beta_schedule)
|
56 |
+
else:
|
57 |
+
raise NotImplementedError
|
58 |
+
|
59 |
+
videogen_pipeline = VideoGenPipeline(vae=vae,
|
60 |
+
text_encoder=text_encoder_one,
|
61 |
+
tokenizer=tokenizer_one,
|
62 |
+
scheduler=scheduler,
|
63 |
+
unet=unet).to(device)
|
64 |
+
videogen_pipeline.enable_xformers_memory_efficient_attention()
|
65 |
+
|
66 |
+
if not os.path.exists(args.output_folder):
|
67 |
+
os.makedirs(args.output_folder)
|
68 |
+
|
69 |
+
video_grids = []
|
70 |
+
for prompt in args.text_prompt:
|
71 |
+
print('Processing the ({}) prompt'.format(prompt))
|
72 |
+
videos = videogen_pipeline(prompt,
|
73 |
+
video_length=args.video_length,
|
74 |
+
height=args.image_size[0],
|
75 |
+
width=args.image_size[1],
|
76 |
+
num_inference_steps=args.num_sampling_steps,
|
77 |
+
guidance_scale=args.guidance_scale).video
|
78 |
+
imageio.mimwrite(args.output_folder + prompt.replace(' ', '_') + '.mp4', videos[0], fps=8, quality=9) # highest quality is 10, lowest is 0
|
79 |
+
|
80 |
+
print('save path {}'.format(args.output_folder))
|
81 |
+
|
82 |
+
if __name__ == "__main__":
|
83 |
+
parser = argparse.ArgumentParser()
|
84 |
+
parser.add_argument("--config", type=str, default="")
|
85 |
+
args = parser.parse_args()
|
86 |
+
|
87 |
+
main(OmegaConf.load(args.config))
|
88 |
+
|
base/pipelines/sample.sh
ADDED
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
1 |
+
export CUDA_VISIBLE_DEVICES=6
|
2 |
+
python pipelines/sample.py --config configs/sample.yaml
|
base/text_to_video/__init__.py
ADDED
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import torch
|
3 |
+
import argparse
|
4 |
+
import torchvision
|
5 |
+
|
6 |
+
from pipelines.pipeline_videogen import VideoGenPipeline
|
7 |
+
|
8 |
+
from download import find_model
|
9 |
+
from diffusers.schedulers import DDIMScheduler, DDPMScheduler, PNDMScheduler, EulerDiscreteScheduler
|
10 |
+
from diffusers.models import AutoencoderKL
|
11 |
+
from transformers import CLIPTokenizer, CLIPTextModel, CLIPTextModelWithProjection
|
12 |
+
from omegaconf import OmegaConf
|
13 |
+
|
14 |
+
import os, sys
|
15 |
+
sys.path.append(os.path.split(sys.path[0])[0])
|
16 |
+
from models import get_models
|
17 |
+
import imageio
|
18 |
+
|
19 |
+
config_path = "/mnt/petrelfs/zhouyan/project/lavie-release/base/configs/sample.yaml"
|
20 |
+
args = OmegaConf.load("/mnt/petrelfs/zhouyan/project/lavie-release/base/configs/sample.yaml")
|
21 |
+
|
22 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
23 |
+
|
24 |
+
def model_t2v_fun(args):
|
25 |
+
sd_path = args.pretrained_path + "/stable-diffusion-v1-4"
|
26 |
+
unet = get_models(args, sd_path).to(device, dtype=torch.float16)
|
27 |
+
# state_dict = find_model(args.pretrained_path + "/lavie_base.pt")
|
28 |
+
state_dict = find_model("/mnt/petrelfs/share_data/wangyaohui/lavie/pretrained_models/lavie_base.pt")
|
29 |
+
unet.load_state_dict(state_dict)
|
30 |
+
|
31 |
+
vae = AutoencoderKL.from_pretrained(sd_path, subfolder="vae", torch_dtype=torch.float16).to(device)
|
32 |
+
tokenizer_one = CLIPTokenizer.from_pretrained(sd_path, subfolder="tokenizer")
|
33 |
+
text_encoder_one = CLIPTextModel.from_pretrained(sd_path, subfolder="text_encoder", torch_dtype=torch.float16).to(device) # huge
|
34 |
+
unet.eval()
|
35 |
+
vae.eval()
|
36 |
+
text_encoder_one.eval()
|
37 |
+
scheduler = DDIMScheduler.from_pretrained(sd_path, subfolder="scheduler", beta_start=args.beta_start, beta_end=args.beta_end, beta_schedule=args.beta_schedule)
|
38 |
+
return VideoGenPipeline(vae=vae, text_encoder=text_encoder_one, tokenizer=tokenizer_one, scheduler=scheduler, unet=unet)
|
39 |
+
|
40 |
+
def setup_seed(seed):
|
41 |
+
torch.manual_seed(seed)
|
42 |
+
torch.cuda.manual_seed_all(seed)
|
43 |
+
|
44 |
+
|
base/text_to_video/__pycache__/__init__.cpython-311.pyc
ADDED
Binary file (3.38 kB). View file
|
|
base/try.py
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
|
3 |
+
with gr.Blocks() as demo:
|
4 |
+
prompt = gr.Textbox(label="Prompt", placeholder="enter prompt", show_label=True, elem_id="prompt-in")
|
5 |
+
demo.launch(server_name="0.0.0.0")
|
environment.yml
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
name: lavie
|
2 |
+
channels:
|
3 |
+
- pytorch
|
4 |
+
- nvidia
|
5 |
+
dependencies:
|
6 |
+
- python=3.11.3
|
7 |
+
- pytorch=2.0.1
|
8 |
+
- pytorch-cuda=11.7
|
9 |
+
- torchvision=0.15.2
|
10 |
+
- pip:
|
11 |
+
- accelerate==0.19.0
|
12 |
+
- av==10.0.0
|
13 |
+
- decord==0.6.0
|
14 |
+
- diffusers[torch]==0.16.0
|
15 |
+
- einops==0.6.1
|
16 |
+
- ffmpeg==1.4
|
17 |
+
- imageio==2.31.1
|
18 |
+
- imageio-ffmpeg==0.4.9
|
19 |
+
- pandas==2.0.1
|
20 |
+
- timm==0.6.13
|
21 |
+
- tqdm==4.65.0
|
22 |
+
- transformers==4.28.1
|
23 |
+
- xformers==0.0.20
|
24 |
+
- omegaconf==2.3.0
|
25 |
+
- natsort==8.4.0
|
26 |
+
- rotary_embedding_torch
|
27 |
+
- gradio==4.3.0
|
interpolation/configs/sample.yaml
ADDED
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
args:
|
2 |
+
input_folder: "../res/base/"
|
3 |
+
pretrained_path: "../pretrained_models"
|
4 |
+
output_folder: "../res/interpolation/"
|
5 |
+
seed_list:
|
6 |
+
- 3418
|
7 |
+
|
8 |
+
fps_list:
|
9 |
+
- 24
|
10 |
+
|
11 |
+
# model config:
|
12 |
+
model: TSR
|
13 |
+
num_frames: 61
|
14 |
+
image_size: [320, 512]
|
15 |
+
num_sampling_steps: 50
|
16 |
+
vae: mse
|
17 |
+
use_timecross_transformer: False
|
18 |
+
frame_interval: 1
|
19 |
+
|
20 |
+
# sample config:
|
21 |
+
seed: 0
|
22 |
+
cfg_scale: 4.0
|
23 |
+
run_time: 12
|
24 |
+
use_compile: False
|
25 |
+
enable_xformers_memory_efficient_attention: True
|
26 |
+
num_sample: 1
|
27 |
+
|
28 |
+
additional_prompt: ", 4k."
|
29 |
+
negative_prompt: "None"
|
30 |
+
do_classifier_free_guidance: True
|
31 |
+
use_ddim_sample_loop: True
|
32 |
+
|
33 |
+
researve_frame: 3
|
34 |
+
mask_type: "tsr"
|
35 |
+
use_concat: True
|
36 |
+
copy_no_mask: True
|
interpolation/datasets/__init__.py
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
from datasets import video_transforms
|
interpolation/datasets/video_transforms.py
ADDED
@@ -0,0 +1,109 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import random
|
3 |
+
import numbers
|
4 |
+
from torchvision.transforms import RandomCrop, RandomResizedCrop
|
5 |
+
|
6 |
+
def _is_tensor_video_clip(clip):
|
7 |
+
if not torch.is_tensor(clip):
|
8 |
+
raise TypeError("clip should be Tensor. Got %s" % type(clip))
|
9 |
+
|
10 |
+
if not clip.ndimension() == 4:
|
11 |
+
raise ValueError("clip should be 4D. Got %dD" % clip.dim())
|
12 |
+
|
13 |
+
return True
|
14 |
+
|
15 |
+
|
16 |
+
def to_tensor(clip):
|
17 |
+
"""
|
18 |
+
Convert tensor data type from uint8 to float, divide value by 255.0 and
|
19 |
+
permute the dimensions of clip tensor
|
20 |
+
Args:
|
21 |
+
clip (torch.tensor, dtype=torch.uint8): Size is (T, C, H, W)
|
22 |
+
Return:
|
23 |
+
clip (torch.tensor, dtype=torch.float): Size is (T, C, H, W)
|
24 |
+
"""
|
25 |
+
_is_tensor_video_clip(clip)
|
26 |
+
if not clip.dtype == torch.uint8:
|
27 |
+
raise TypeError("clip tensor should have data type uint8. Got %s" % str(clip.dtype))
|
28 |
+
# return clip.float().permute(3, 0, 1, 2) / 255.0
|
29 |
+
return clip.float() / 255.0
|
30 |
+
|
31 |
+
|
32 |
+
def resize(clip, target_size, interpolation_mode):
|
33 |
+
if len(target_size) != 2:
|
34 |
+
raise ValueError(f"target size should be tuple (height, width), instead got {target_size}")
|
35 |
+
return torch.nn.functional.interpolate(clip, size=target_size, mode=interpolation_mode, align_corners=False)
|
36 |
+
|
37 |
+
|
38 |
+
class ToTensorVideo:
|
39 |
+
"""
|
40 |
+
Convert tensor data type from uint8 to float, divide value by 255.0 and
|
41 |
+
permute the dimensions of clip tensor
|
42 |
+
"""
|
43 |
+
|
44 |
+
def __init__(self):
|
45 |
+
pass
|
46 |
+
|
47 |
+
def __call__(self, clip):
|
48 |
+
"""
|
49 |
+
Args:
|
50 |
+
clip (torch.tensor, dtype=torch.uint8): Size is (T, C, H, W)
|
51 |
+
Return:
|
52 |
+
clip (torch.tensor, dtype=torch.float): Size is (T, C, H, W)
|
53 |
+
"""
|
54 |
+
return to_tensor(clip)
|
55 |
+
|
56 |
+
def __repr__(self) -> str:
|
57 |
+
return self.__class__.__name__
|
58 |
+
|
59 |
+
|
60 |
+
class ResizeVideo:
|
61 |
+
'''
|
62 |
+
Resize to the specified size
|
63 |
+
'''
|
64 |
+
def __init__(
|
65 |
+
self,
|
66 |
+
size,
|
67 |
+
interpolation_mode="bilinear",
|
68 |
+
):
|
69 |
+
if isinstance(size, tuple):
|
70 |
+
if len(size) != 2:
|
71 |
+
raise ValueError(f"size should be tuple (height, width), instead got {size}")
|
72 |
+
self.size = size
|
73 |
+
else:
|
74 |
+
self.size = (size, size)
|
75 |
+
|
76 |
+
self.interpolation_mode = interpolation_mode
|
77 |
+
|
78 |
+
|
79 |
+
def __call__(self, clip):
|
80 |
+
"""
|
81 |
+
Args:
|
82 |
+
clip (torch.tensor): Video clip to be cropped. Size is (T, C, H, W)
|
83 |
+
Returns:
|
84 |
+
torch.tensor: scale resized video clip.
|
85 |
+
size is (T, C, h, w)
|
86 |
+
"""
|
87 |
+
clip_resize = resize(clip, target_size=self.size, interpolation_mode=self.interpolation_mode)
|
88 |
+
return clip_resize
|
89 |
+
|
90 |
+
def __repr__(self) -> str:
|
91 |
+
return f"{self.__class__.__name__}(size={self.size}, interpolation_mode={self.interpolation_mode}"
|
92 |
+
|
93 |
+
|
94 |
+
class TemporalRandomCrop(object):
|
95 |
+
"""Temporally crop the given frame indices at a random location.
|
96 |
+
|
97 |
+
Args:
|
98 |
+
size (int): Desired length of frames will be seen in the model.
|
99 |
+
"""
|
100 |
+
|
101 |
+
def __init__(self, size):
|
102 |
+
self.size = size
|
103 |
+
|
104 |
+
def __call__(self, total_frames):
|
105 |
+
rand_end = max(0, total_frames - self.size - 1)
|
106 |
+
begin_index = random.randint(0, rand_end)
|
107 |
+
end_index = min(begin_index + self.size, total_frames)
|
108 |
+
return begin_index, end_index
|
109 |
+
|
interpolation/diffusion/__init__.py
ADDED
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Modified from OpenAI's diffusion repos
|
2 |
+
# GLIDE: https://github.com/openai/glide-text2im/blob/main/glide_text2im/gaussian_diffusion.py
|
3 |
+
# ADM: https://github.com/openai/guided-diffusion/blob/main/guided_diffusion
|
4 |
+
# IDDPM: https://github.com/openai/improved-diffusion/blob/main/improved_diffusion/gaussian_diffusion.py
|
5 |
+
|
6 |
+
from . import gaussian_diffusion as gd
|
7 |
+
from .respace import SpacedDiffusion, space_timesteps
|
8 |
+
|
9 |
+
|
10 |
+
def create_diffusion(
|
11 |
+
timestep_respacing,
|
12 |
+
noise_schedule="linear",
|
13 |
+
use_kl=False,
|
14 |
+
sigma_small=False,
|
15 |
+
predict_xstart=False,
|
16 |
+
# learn_sigma=True,
|
17 |
+
learn_sigma=False, # for unet
|
18 |
+
rescale_learned_sigmas=False,
|
19 |
+
diffusion_steps=1000
|
20 |
+
):
|
21 |
+
betas = gd.get_named_beta_schedule(noise_schedule, diffusion_steps)
|
22 |
+
if use_kl:
|
23 |
+
loss_type = gd.LossType.RESCALED_KL
|
24 |
+
elif rescale_learned_sigmas:
|
25 |
+
loss_type = gd.LossType.RESCALED_MSE
|
26 |
+
else:
|
27 |
+
loss_type = gd.LossType.MSE
|
28 |
+
if timestep_respacing is None or timestep_respacing == "":
|
29 |
+
timestep_respacing = [diffusion_steps]
|
30 |
+
return SpacedDiffusion(
|
31 |
+
use_timesteps=space_timesteps(diffusion_steps, timestep_respacing),
|
32 |
+
betas=betas,
|
33 |
+
model_mean_type=(
|
34 |
+
gd.ModelMeanType.EPSILON if not predict_xstart else gd.ModelMeanType.START_X
|
35 |
+
),
|
36 |
+
model_var_type=(
|
37 |
+
(
|
38 |
+
gd.ModelVarType.FIXED_LARGE
|
39 |
+
if not sigma_small
|
40 |
+
else gd.ModelVarType.FIXED_SMALL
|
41 |
+
)
|
42 |
+
if not learn_sigma
|
43 |
+
else gd.ModelVarType.LEARNED_RANGE
|
44 |
+
),
|
45 |
+
loss_type=loss_type
|
46 |
+
# rescale_timesteps=rescale_timesteps,
|
47 |
+
)
|
interpolation/diffusion/diffusion_utils.py
ADDED
@@ -0,0 +1,88 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Modified from OpenAI's diffusion repos
|
2 |
+
# GLIDE: https://github.com/openai/glide-text2im/blob/main/glide_text2im/gaussian_diffusion.py
|
3 |
+
# ADM: https://github.com/openai/guided-diffusion/blob/main/guided_diffusion
|
4 |
+
# IDDPM: https://github.com/openai/improved-diffusion/blob/main/improved_diffusion/gaussian_diffusion.py
|
5 |
+
|
6 |
+
import torch as th
|
7 |
+
import numpy as np
|
8 |
+
|
9 |
+
|
10 |
+
def normal_kl(mean1, logvar1, mean2, logvar2):
|
11 |
+
"""
|
12 |
+
Compute the KL divergence between two gaussians.
|
13 |
+
Shapes are automatically broadcasted, so batches can be compared to
|
14 |
+
scalars, among other use cases.
|
15 |
+
"""
|
16 |
+
tensor = None
|
17 |
+
for obj in (mean1, logvar1, mean2, logvar2):
|
18 |
+
if isinstance(obj, th.Tensor):
|
19 |
+
tensor = obj
|
20 |
+
break
|
21 |
+
assert tensor is not None, "at least one argument must be a Tensor"
|
22 |
+
|
23 |
+
# Force variances to be Tensors. Broadcasting helps convert scalars to
|
24 |
+
# Tensors, but it does not work for th.exp().
|
25 |
+
logvar1, logvar2 = [
|
26 |
+
x if isinstance(x, th.Tensor) else th.tensor(x).to(tensor)
|
27 |
+
for x in (logvar1, logvar2)
|
28 |
+
]
|
29 |
+
|
30 |
+
return 0.5 * (
|
31 |
+
-1.0
|
32 |
+
+ logvar2
|
33 |
+
- logvar1
|
34 |
+
+ th.exp(logvar1 - logvar2)
|
35 |
+
+ ((mean1 - mean2) ** 2) * th.exp(-logvar2)
|
36 |
+
)
|
37 |
+
|
38 |
+
|
39 |
+
def approx_standard_normal_cdf(x):
|
40 |
+
"""
|
41 |
+
A fast approximation of the cumulative distribution function of the
|
42 |
+
standard normal.
|
43 |
+
"""
|
44 |
+
return 0.5 * (1.0 + th.tanh(np.sqrt(2.0 / np.pi) * (x + 0.044715 * th.pow(x, 3))))
|
45 |
+
|
46 |
+
|
47 |
+
def continuous_gaussian_log_likelihood(x, *, means, log_scales):
|
48 |
+
"""
|
49 |
+
Compute the log-likelihood of a continuous Gaussian distribution.
|
50 |
+
:param x: the targets
|
51 |
+
:param means: the Gaussian mean Tensor.
|
52 |
+
:param log_scales: the Gaussian log stddev Tensor.
|
53 |
+
:return: a tensor like x of log probabilities (in nats).
|
54 |
+
"""
|
55 |
+
centered_x = x - means
|
56 |
+
inv_stdv = th.exp(-log_scales)
|
57 |
+
normalized_x = centered_x * inv_stdv
|
58 |
+
log_probs = th.distributions.Normal(th.zeros_like(x), th.ones_like(x)).log_prob(normalized_x)
|
59 |
+
return log_probs
|
60 |
+
|
61 |
+
|
62 |
+
def discretized_gaussian_log_likelihood(x, *, means, log_scales):
|
63 |
+
"""
|
64 |
+
Compute the log-likelihood of a Gaussian distribution discretizing to a
|
65 |
+
given image.
|
66 |
+
:param x: the target images. It is assumed that this was uint8 values,
|
67 |
+
rescaled to the range [-1, 1].
|
68 |
+
:param means: the Gaussian mean Tensor.
|
69 |
+
:param log_scales: the Gaussian log stddev Tensor.
|
70 |
+
:return: a tensor like x of log probabilities (in nats).
|
71 |
+
"""
|
72 |
+
assert x.shape == means.shape == log_scales.shape
|
73 |
+
centered_x = x - means
|
74 |
+
inv_stdv = th.exp(-log_scales)
|
75 |
+
plus_in = inv_stdv * (centered_x + 1.0 / 255.0)
|
76 |
+
cdf_plus = approx_standard_normal_cdf(plus_in)
|
77 |
+
min_in = inv_stdv * (centered_x - 1.0 / 255.0)
|
78 |
+
cdf_min = approx_standard_normal_cdf(min_in)
|
79 |
+
log_cdf_plus = th.log(cdf_plus.clamp(min=1e-12))
|
80 |
+
log_one_minus_cdf_min = th.log((1.0 - cdf_min).clamp(min=1e-12))
|
81 |
+
cdf_delta = cdf_plus - cdf_min
|
82 |
+
log_probs = th.where(
|
83 |
+
x < -0.999,
|
84 |
+
log_cdf_plus,
|
85 |
+
th.where(x > 0.999, log_one_minus_cdf_min, th.log(cdf_delta.clamp(min=1e-12))),
|
86 |
+
)
|
87 |
+
assert log_probs.shape == x.shape
|
88 |
+
return log_probs
|