Roopansh commited on
Commit
73c83cf
β€’
1 Parent(s): b26af83

Initial Commit

Browse files
This view is limited to 50 files because it contains too many changes. Β  See raw diff
Files changed (50) hide show
  1. .gitattributes +1 -0
  2. .gitignore +180 -0
  3. README.md +160 -10
  4. app.py +313 -0
  5. app_VTON.py +328 -0
  6. apply_net.py +359 -0
  7. ckpt/densepose/model_final_162be9.pkl +3 -0
  8. ckpt/humanparsing/parsing_atr.onnx +3 -0
  9. ckpt/humanparsing/parsing_lip.onnx +3 -0
  10. ckpt/openpose/.DS_Store +3 -0
  11. ckpt/openpose/ckpts/body_pose_model.pth +3 -0
  12. configs/Base-DensePose-RCNN-FPN.yaml +48 -0
  13. configs/HRNet/densepose_rcnn_HRFPN_HRNet_w32_s1x.yaml +16 -0
  14. configs/HRNet/densepose_rcnn_HRFPN_HRNet_w40_s1x.yaml +23 -0
  15. configs/HRNet/densepose_rcnn_HRFPN_HRNet_w48_s1x.yaml +23 -0
  16. configs/cse/Base-DensePose-RCNN-FPN-Human.yaml +20 -0
  17. configs/cse/Base-DensePose-RCNN-FPN.yaml +60 -0
  18. configs/cse/densepose_rcnn_R_101_FPN_DL_s1x.yaml +12 -0
  19. configs/cse/densepose_rcnn_R_101_FPN_DL_soft_s1x.yaml +12 -0
  20. configs/cse/densepose_rcnn_R_101_FPN_s1x.yaml +12 -0
  21. configs/cse/densepose_rcnn_R_101_FPN_soft_s1x.yaml +12 -0
  22. configs/cse/densepose_rcnn_R_50_FPN_DL_s1x.yaml +12 -0
  23. configs/cse/densepose_rcnn_R_50_FPN_DL_soft_s1x.yaml +12 -0
  24. configs/cse/densepose_rcnn_R_50_FPN_s1x.yaml +12 -0
  25. configs/cse/densepose_rcnn_R_50_FPN_soft_animals_CA_finetune_16k.yaml +133 -0
  26. configs/cse/densepose_rcnn_R_50_FPN_soft_animals_CA_finetune_4k.yaml +133 -0
  27. configs/cse/densepose_rcnn_R_50_FPN_soft_animals_I0_finetune_16k.yaml +119 -0
  28. configs/cse/densepose_rcnn_R_50_FPN_soft_animals_I0_finetune_i2m_16k.yaml +121 -0
  29. configs/cse/densepose_rcnn_R_50_FPN_soft_animals_I0_finetune_m2m_16k.yaml +138 -0
  30. configs/cse/densepose_rcnn_R_50_FPN_soft_animals_finetune_16k.yaml +119 -0
  31. configs/cse/densepose_rcnn_R_50_FPN_soft_animals_finetune_4k.yaml +119 -0
  32. configs/cse/densepose_rcnn_R_50_FPN_soft_animals_finetune_maskonly_24k.yaml +118 -0
  33. configs/cse/densepose_rcnn_R_50_FPN_soft_chimps_finetune_4k.yaml +29 -0
  34. configs/cse/densepose_rcnn_R_50_FPN_soft_s1x.yaml +12 -0
  35. configs/densepose_rcnn_R_101_FPN_DL_WC1M_s1x.yaml +18 -0
  36. configs/densepose_rcnn_R_101_FPN_DL_WC1_s1x.yaml +16 -0
  37. configs/densepose_rcnn_R_101_FPN_DL_WC2M_s1x.yaml +18 -0
  38. configs/densepose_rcnn_R_101_FPN_DL_WC2_s1x.yaml +16 -0
  39. configs/densepose_rcnn_R_101_FPN_DL_s1x.yaml +10 -0
  40. configs/densepose_rcnn_R_101_FPN_WC1M_s1x.yaml +18 -0
  41. configs/densepose_rcnn_R_101_FPN_WC1_s1x.yaml +16 -0
  42. configs/densepose_rcnn_R_101_FPN_WC2M_s1x.yaml +18 -0
  43. configs/densepose_rcnn_R_101_FPN_WC2_s1x.yaml +16 -0
  44. configs/densepose_rcnn_R_101_FPN_s1x.yaml +8 -0
  45. configs/densepose_rcnn_R_101_FPN_s1x_legacy.yaml +17 -0
  46. configs/densepose_rcnn_R_50_FPN_DL_WC1M_s1x.yaml +18 -0
  47. configs/densepose_rcnn_R_50_FPN_DL_WC1_s1x.yaml +16 -0
  48. configs/densepose_rcnn_R_50_FPN_DL_WC2M_s1x.yaml +18 -0
  49. configs/densepose_rcnn_R_50_FPN_DL_WC2_s1x.yaml +16 -0
  50. configs/densepose_rcnn_R_50_FPN_DL_s1x.yaml +10 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ ckpt/** filter=lfs diff=lfs merge=lfs -text
.gitignore ADDED
@@ -0,0 +1,180 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Byte-compiled / optimized / DLL files
2
+ __pycache__/
3
+ *.py[cod]
4
+ *$py.class
5
+
6
+ # C extensions
7
+ *.so
8
+
9
+ # Distribution / packaging
10
+ .Python
11
+ build/
12
+ develop-eggs/
13
+ dist/
14
+ downloads/
15
+ eggs/
16
+ .eggs/
17
+ lib/
18
+ lib64/
19
+ parts/
20
+ sdist/
21
+ var/
22
+ wheels/
23
+ share/python-wheels/
24
+ *.egg-info/
25
+ .installed.cfg
26
+ *.egg
27
+ MANIFEST
28
+ venv
29
+ models
30
+ .vs/
31
+ outputs
32
+ temp_faces
33
+ loras
34
+
35
+ # PyInstaller
36
+ # Usually these files are written by a python script from a template
37
+ # before PyInstaller builds the exe, so as to inject date/other infos into it.
38
+ *.manifest
39
+ *.spec
40
+
41
+ # Installer logs
42
+ pip-log.txt
43
+ pip-delete-this-directory.txt
44
+
45
+ # Unit test / coverage reports
46
+ htmlcov/
47
+ .tox/
48
+ .nox/
49
+ .coverage
50
+ .coverage.*
51
+ .cache
52
+ nosetests.xml
53
+ coverage.xml
54
+ *.cover
55
+ *.py,cover
56
+ .hypothesis/
57
+ .pytest_cache/
58
+ cover/
59
+
60
+ # Translations
61
+ *.mo
62
+ *.pot
63
+
64
+ # Django stuff:
65
+ *.log
66
+ local_settings.py
67
+ db.sqlite3
68
+ db.sqlite3-journal
69
+
70
+ # Flask stuff:
71
+ instance/
72
+ .webassets-cache
73
+
74
+ # Scrapy stuff:
75
+ .scrapy
76
+
77
+ # Sphinx documentation
78
+ docs/_build/
79
+
80
+ # PyBuilder
81
+ .pybuilder/
82
+ target/
83
+
84
+ # Jupyter Notebook
85
+ .ipynb_checkpoints
86
+
87
+ # IPython
88
+ profile_default/
89
+ ipython_config.py
90
+
91
+ # pyenv
92
+ # For a library or package, you might want to ignore these files since the code is
93
+ # intended to run in multiple environments; otherwise, check them in:
94
+ # .python-version
95
+
96
+ # pipenv
97
+ # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
98
+ # However, in case of collaboration, if having platform-specific dependencies or dependencies
99
+ # having no cross-platform support, pipenv may install dependencies that don't work, or not
100
+ # install all needed dependencies.
101
+ #Pipfile.lock
102
+
103
+ # poetry
104
+ # Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
105
+ # This is especially recommended for binary packages to ensure reproducibility, and is more
106
+ # commonly ignored for libraries.
107
+ # https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
108
+ #poetry.lock
109
+
110
+ # pdm
111
+ # Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
112
+ #pdm.lock
113
+ # pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
114
+ # in version control.
115
+ # https://pdm.fming.dev/#use-with-ide
116
+ .pdm.toml
117
+
118
+ # PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
119
+ __pypackages__/
120
+
121
+ # Celery stuff
122
+ celerybeat-schedule
123
+ celerybeat.pid
124
+
125
+ # SageMath parsed files
126
+ *.sage.py
127
+
128
+ # Environments
129
+ .env
130
+ .venv
131
+ env/
132
+ venv/
133
+ ENV/
134
+ env.bak/
135
+ venv.bak/
136
+
137
+ # Spyder project settings
138
+ .spyderproject
139
+ .spyproject
140
+
141
+ # Rope project settings
142
+ .ropeproject
143
+
144
+ # mkdocs documentation
145
+ /site
146
+
147
+ # mypy
148
+ .mypy_cache/
149
+ .dmypy.json
150
+ dmypy.json
151
+
152
+ # Pyre type checker
153
+ .pyre/
154
+
155
+ # pytype static type analyzer
156
+ .pytype/
157
+
158
+ # Cython debug symbols
159
+ cython_debug/
160
+
161
+ # PyCharm
162
+ # JetBrains specific template is maintained in a separate JetBrains.gitignore that can
163
+ # be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
164
+ # and can be added to the global gitignore or merged into this file. For a more nuclear
165
+ # option (not recommended) you can uncomment the following to ignore the entire idea folder.
166
+ #.idea/
167
+
168
+ .vscode/
169
+ *.pyc
170
+
171
+ #others
172
+ outputs/
173
+ __pycache__/
174
+ densepose/__pycache__/
175
+ detectron2/__pycache__/
176
+ ip_adapter/__pycache__/
177
+ preprocess/humanparsing/__pycache__/
178
+ preprocess/openpose/__pycache__/
179
+ util/__pycache__/
180
+ src/__pycache__/
README.md CHANGED
@@ -1,12 +1,162 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
- title: Ailusion VTON DEMO V1
3
- emoji: πŸ‘
4
- colorFrom: red
5
- colorTo: indigo
6
- sdk: gradio
7
- sdk_version: 4.31.2
8
- app_file: app.py
9
- pinned: false
10
- ---
11
 
12
- Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ <div align="center">
3
+ <h1>IDM-VTON: Improving Diffusion Models for Authentic Virtual Try-on in the Wild</h1>
4
+
5
+ <a href='https://idm-vton.github.io'><img src='https://img.shields.io/badge/Project-Page-green'></a>
6
+ <a href='https://arxiv.org/abs/2403.05139'><img src='https://img.shields.io/badge/Paper-Arxiv-red'></a>
7
+ <a href='https://huggingface.co/spaces/yisol/IDM-VTON'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Demo-blue'></a>
8
+ <a href='https://huggingface.co/yisol/IDM-VTON'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-blue'></a>
9
+
10
+
11
+ </div>
12
+
13
+ This is the official implementation of the paper ["Improving Diffusion Models for Authentic Virtual Try-on in the Wild"](https://arxiv.org/abs/2403.05139).
14
+
15
+ Star ⭐ us if you like it!
16
+
17
  ---
 
 
 
 
 
 
 
 
 
18
 
19
+
20
+ ![teaser2](assets/teaser2.png)&nbsp;
21
+ ![teaser](assets/teaser.png)&nbsp;
22
+
23
+
24
+ ## TODO LIST
25
+
26
+
27
+ - [x] demo model
28
+ - [x] inference code
29
+ - [ ] training code
30
+
31
+
32
+
33
+ ## Requirements
34
+
35
+ ```
36
+ git clone https://github.com/yisol/IDM-VTON.git
37
+ cd IDM-VTON
38
+
39
+ conda env create -f environment.yaml
40
+ conda activate idm
41
+ ```
42
+
43
+ ## Data preparation
44
+
45
+ ### VITON-HD
46
+ You can download VITON-HD dataset from [VITON-HD](https://github.com/shadow2496/VITON-HD).
47
+
48
+ After download VITON-HD dataset, move vitonhd_test_tagged.json into the test folder.
49
+
50
+ Structure of the Dataset directory should be as follows.
51
+
52
+ ```
53
+
54
+ train
55
+ |-- ...
56
+
57
+ test
58
+ |-- image
59
+ |-- image-densepose
60
+ |-- agnostic-mask
61
+ |-- cloth
62
+ |-- vitonhd_test_tagged.json
63
+
64
+ ```
65
+
66
+ ### DressCode
67
+ You can download DressCode dataset from [DressCode](https://github.com/aimagelab/dress-code).
68
+
69
+ We provide pre-computed densepose images and captions for garments [here](https://kaistackr-my.sharepoint.com/:u:/g/personal/cpis7_kaist_ac_kr/EaIPRG-aiRRIopz9i002FOwBDa-0-BHUKVZ7Ia5yAVVG3A?e=YxkAip).
70
+
71
+ We used [detectron2](https://github.com/facebookresearch/detectron2) for obtaining densepose images, refer [here](https://github.com/sangyun884/HR-VITON/issues/45) for more details.
72
+
73
+ After download the DressCode dataset, place image-densepose directories and caption text files as follows.
74
+
75
+ ```
76
+ DressCode
77
+ |-- dresses
78
+ |-- images
79
+ |-- image-densepose
80
+ |-- dc_caption.txt
81
+ |-- ...
82
+ |-- lower_body
83
+ |-- images
84
+ |-- image-densepose
85
+ |-- dc_caption.txt
86
+ |-- ...
87
+ |-- upper_body
88
+ |-- images
89
+ |-- image-densepose
90
+ |-- dc_caption.txt
91
+ |-- ...
92
+ ```
93
+
94
+
95
+ ## Inference
96
+
97
+
98
+ ### VITON-HD
99
+
100
+ Inference using python file with arguments,
101
+
102
+ ```
103
+ accelerate launch inference.py \
104
+ --width 768 --height 1024 --num_inference_steps 30 \
105
+ --output_dir "result" \
106
+ --unpaired \
107
+ --data_dir "DATA_DIR" \
108
+ --seed 42 \
109
+ --test_batch_size 2 \
110
+ --guidance_scale 2.0
111
+ ```
112
+
113
+ or, you can simply run with the script file.
114
+
115
+ ```
116
+ sh inference.sh
117
+ ```
118
+
119
+ ### DressCode
120
+
121
+ For DressCode dataset, put the category you want to generate images via category argument,
122
+ ```
123
+ accelerate launch inference_dc.py \
124
+ --width 768 --height 1024 --num_inference_steps 30 \
125
+ --output_dir "result" \
126
+ --unpaired \
127
+ --data_dir "DATA_DIR" \
128
+ --seed 42
129
+ --test_batch_size 2
130
+ --guidance_scale 2.0
131
+ --category "upper_body"
132
+ ```
133
+
134
+ or, you can simply run with the script file.
135
+ ```
136
+ sh inference.sh
137
+ ```
138
+
139
+
140
+ ## Acknowledgements
141
+
142
+ For the [demo](https://huggingface.co/spaces/yisol/IDM-VTON), GPUs are supported from [ZeroGPU](https://huggingface.co/zero-gpu-explorers), and masking generation codes are based on [OOTDiffusion](https://github.com/levihsu/OOTDiffusion) and [DCI-VTON](https://github.com/bcmi/DCI-VTON-Virtual-Try-On).
143
+
144
+ Parts of our code are based on [IP-Adapter](https://github.com/tencent-ailab/IP-Adapter).
145
+
146
+
147
+
148
+ ## Citation
149
+ ```
150
+ @article{choi2024improving,
151
+ title={Improving Diffusion Models for Virtual Try-on},
152
+ author={Choi, Yisol and Kwak, Sangkyung and Lee, Kyungmin and Choi, Hyungwon and Shin, Jinwoo},
153
+ journal={arXiv preprint arXiv:2403.05139},
154
+ year={2024}
155
+ }
156
+ ```
157
+
158
+ ## License
159
+ The codes and checkpoints in this repository are under the [CC BY-NC-SA 4.0 license](https://creativecommons.org/licenses/by-nc-sa/4.0/legalcode).
160
+
161
+
162
+
app.py ADDED
@@ -0,0 +1,313 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+ from PIL import Image
3
+ from src.tryon_pipeline import StableDiffusionXLInpaintPipeline as TryonPipeline
4
+ from src.unet_hacked_garmnet import UNet2DConditionModel as UNet2DConditionModel_ref
5
+ from src.unet_hacked_tryon import UNet2DConditionModel
6
+ from transformers import (
7
+ CLIPImageProcessor,
8
+ CLIPVisionModelWithProjection,
9
+ CLIPTextModel,
10
+ CLIPTextModelWithProjection,
11
+ )
12
+ from diffusers import DDPMScheduler,AutoencoderKL
13
+ from typing import List
14
+
15
+ import torch
16
+ import os
17
+ from transformers import AutoTokenizer
18
+ import spaces
19
+ import numpy as np
20
+ from utils_mask import get_mask_location
21
+ from torchvision import transforms
22
+ import apply_net
23
+ from preprocess.humanparsing.run_parsing import Parsing
24
+ from preprocess.openpose.run_openpose import OpenPose
25
+ from detectron2.data.detection_utils import convert_PIL_to_numpy,_apply_exif_orientation
26
+ from torchvision.transforms.functional import to_pil_image
27
+
28
+
29
+ def pil_to_binary_mask(pil_image, threshold=0):
30
+ np_image = np.array(pil_image)
31
+ grayscale_image = Image.fromarray(np_image).convert("L")
32
+ binary_mask = np.array(grayscale_image) > threshold
33
+ mask = np.zeros(binary_mask.shape, dtype=np.uint8)
34
+ for i in range(binary_mask.shape[0]):
35
+ for j in range(binary_mask.shape[1]):
36
+ if binary_mask[i,j] == True :
37
+ mask[i,j] = 1
38
+ mask = (mask*255).astype(np.uint8)
39
+ output_mask = Image.fromarray(mask)
40
+ return output_mask
41
+
42
+
43
+ base_path = 'yisol/IDM-VTON'
44
+ example_path = os.path.join(os.path.dirname(__file__), 'example')
45
+
46
+ unet = UNet2DConditionModel.from_pretrained(
47
+ base_path,
48
+ subfolder="unet",
49
+ torch_dtype=torch.float16,
50
+ )
51
+ unet.requires_grad_(False)
52
+ tokenizer_one = AutoTokenizer.from_pretrained(
53
+ base_path,
54
+ subfolder="tokenizer",
55
+ revision=None,
56
+ use_fast=False,
57
+ )
58
+ tokenizer_two = AutoTokenizer.from_pretrained(
59
+ base_path,
60
+ subfolder="tokenizer_2",
61
+ revision=None,
62
+ use_fast=False,
63
+ )
64
+ noise_scheduler = DDPMScheduler.from_pretrained(base_path, subfolder="scheduler")
65
+
66
+ text_encoder_one = CLIPTextModel.from_pretrained(
67
+ base_path,
68
+ subfolder="text_encoder",
69
+ torch_dtype=torch.float16,
70
+ )
71
+ text_encoder_two = CLIPTextModelWithProjection.from_pretrained(
72
+ base_path,
73
+ subfolder="text_encoder_2",
74
+ torch_dtype=torch.float16,
75
+ )
76
+ image_encoder = CLIPVisionModelWithProjection.from_pretrained(
77
+ base_path,
78
+ subfolder="image_encoder",
79
+ torch_dtype=torch.float16,
80
+ )
81
+ vae = AutoencoderKL.from_pretrained(base_path,
82
+ subfolder="vae",
83
+ torch_dtype=torch.float16,
84
+ )
85
+
86
+ # "stabilityai/stable-diffusion-xl-base-1.0",
87
+ UNet_Encoder = UNet2DConditionModel_ref.from_pretrained(
88
+ base_path,
89
+ subfolder="unet_encoder",
90
+ torch_dtype=torch.float16,
91
+ )
92
+
93
+ parsing_model = Parsing(0)
94
+ openpose_model = OpenPose(0)
95
+
96
+ UNet_Encoder.requires_grad_(False)
97
+ image_encoder.requires_grad_(False)
98
+ vae.requires_grad_(False)
99
+ unet.requires_grad_(False)
100
+ text_encoder_one.requires_grad_(False)
101
+ text_encoder_two.requires_grad_(False)
102
+ tensor_transfrom = transforms.Compose(
103
+ [
104
+ transforms.ToTensor(),
105
+ transforms.Normalize([0.5], [0.5]),
106
+ ]
107
+ )
108
+
109
+ pipe = TryonPipeline.from_pretrained(
110
+ base_path,
111
+ unet=unet,
112
+ vae=vae,
113
+ feature_extractor= CLIPImageProcessor(),
114
+ text_encoder = text_encoder_one,
115
+ text_encoder_2 = text_encoder_two,
116
+ tokenizer = tokenizer_one,
117
+ tokenizer_2 = tokenizer_two,
118
+ scheduler = noise_scheduler,
119
+ image_encoder=image_encoder,
120
+ torch_dtype=torch.float16,
121
+ )
122
+ pipe.unet_encoder = UNet_Encoder
123
+
124
+ @spaces.GPU
125
+ def start_tryon(dict,garm_img,garment_des,is_checked,is_checked_crop,denoise_steps,seed):
126
+ device = "cuda"
127
+
128
+ openpose_model.preprocessor.body_estimation.model.to(device)
129
+ pipe.to(device)
130
+ pipe.unet_encoder.to(device)
131
+
132
+ garm_img= garm_img.convert("RGB").resize((768,1024))
133
+ human_img_orig = dict["background"].convert("RGB")
134
+
135
+ if is_checked_crop:
136
+ width, height = human_img_orig.size
137
+ target_width = int(min(width, height * (3 / 4)))
138
+ target_height = int(min(height, width * (4 / 3)))
139
+ left = (width - target_width) / 2
140
+ top = (height - target_height) / 2
141
+ right = (width + target_width) / 2
142
+ bottom = (height + target_height) / 2
143
+ cropped_img = human_img_orig.crop((left, top, right, bottom))
144
+ crop_size = cropped_img.size
145
+ human_img = cropped_img.resize((768,1024))
146
+ else:
147
+ human_img = human_img_orig.resize((768,1024))
148
+
149
+
150
+ if is_checked:
151
+ keypoints = openpose_model(human_img.resize((384,512)))
152
+ model_parse, _ = parsing_model(human_img.resize((384,512)))
153
+ mask, mask_gray = get_mask_location('hd', "upper_body", model_parse, keypoints)
154
+ mask = mask.resize((768,1024))
155
+ else:
156
+ mask = pil_to_binary_mask(dict['layers'][0].convert("RGB").resize((768, 1024)))
157
+ # mask = transforms.ToTensor()(mask)
158
+ # mask = mask.unsqueeze(0)
159
+ mask_gray = (1-transforms.ToTensor()(mask)) * tensor_transfrom(human_img)
160
+ mask_gray = to_pil_image((mask_gray+1.0)/2.0)
161
+
162
+
163
+ human_img_arg = _apply_exif_orientation(human_img.resize((384,512)))
164
+ human_img_arg = convert_PIL_to_numpy(human_img_arg, format="BGR")
165
+
166
+
167
+
168
+ args = apply_net.create_argument_parser().parse_args(('show', './configs/densepose_rcnn_R_50_FPN_s1x.yaml', './ckpt/densepose/model_final_162be9.pkl', 'dp_segm', '-v', '--opts', 'MODEL.DEVICE', 'cuda'))
169
+ # verbosity = getattr(args, "verbosity", None)
170
+ pose_img = args.func(args,human_img_arg)
171
+ pose_img = pose_img[:,:,::-1]
172
+ pose_img = Image.fromarray(pose_img).resize((768,1024))
173
+
174
+ with torch.no_grad():
175
+ # Extract the images
176
+ with torch.cuda.amp.autocast():
177
+ with torch.no_grad():
178
+ prompt = "model is wearing " + garment_des
179
+ negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
180
+ with torch.inference_mode():
181
+ (
182
+ prompt_embeds,
183
+ negative_prompt_embeds,
184
+ pooled_prompt_embeds,
185
+ negative_pooled_prompt_embeds,
186
+ ) = pipe.encode_prompt(
187
+ prompt,
188
+ num_images_per_prompt=1,
189
+ do_classifier_free_guidance=True,
190
+ negative_prompt=negative_prompt,
191
+ )
192
+
193
+ prompt = "a photo of " + garment_des
194
+ negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
195
+ if not isinstance(prompt, List):
196
+ prompt = [prompt] * 1
197
+ if not isinstance(negative_prompt, List):
198
+ negative_prompt = [negative_prompt] * 1
199
+ with torch.inference_mode():
200
+ (
201
+ prompt_embeds_c,
202
+ _,
203
+ _,
204
+ _,
205
+ ) = pipe.encode_prompt(
206
+ prompt,
207
+ num_images_per_prompt=1,
208
+ do_classifier_free_guidance=False,
209
+ negative_prompt=negative_prompt,
210
+ )
211
+
212
+
213
+
214
+ pose_img = tensor_transfrom(pose_img).unsqueeze(0).to(device,torch.float16)
215
+ garm_tensor = tensor_transfrom(garm_img).unsqueeze(0).to(device,torch.float16)
216
+ generator = torch.Generator(device).manual_seed(seed) if seed is not None else None
217
+ images = pipe(
218
+ prompt_embeds=prompt_embeds.to(device,torch.float16),
219
+ negative_prompt_embeds=negative_prompt_embeds.to(device,torch.float16),
220
+ pooled_prompt_embeds=pooled_prompt_embeds.to(device,torch.float16),
221
+ negative_pooled_prompt_embeds=negative_pooled_prompt_embeds.to(device,torch.float16),
222
+ num_inference_steps=denoise_steps,
223
+ generator=generator,
224
+ strength = 1.0,
225
+ pose_img = pose_img.to(device,torch.float16),
226
+ text_embeds_cloth=prompt_embeds_c.to(device,torch.float16),
227
+ cloth = garm_tensor.to(device,torch.float16),
228
+ mask_image=mask,
229
+ image=human_img,
230
+ height=1024,
231
+ width=768,
232
+ ip_adapter_image = garm_img.resize((768,1024)),
233
+ guidance_scale=2.0,
234
+ )[0]
235
+
236
+ if is_checked_crop:
237
+ out_img = images[0].resize(crop_size)
238
+ human_img_orig.paste(out_img, (int(left), int(top)))
239
+ return human_img_orig, mask_gray
240
+ else:
241
+ return images[0], mask_gray
242
+ # return images[0], mask_gray
243
+
244
+ garm_list = os.listdir(os.path.join(example_path,"cloth"))
245
+ garm_list_path = [os.path.join(example_path,"cloth",garm) for garm in garm_list]
246
+
247
+ human_list = os.listdir(os.path.join(example_path,"human"))
248
+ human_list_path = [os.path.join(example_path,"human",human) for human in human_list]
249
+
250
+ human_ex_list = []
251
+ for ex_human in human_list_path:
252
+ ex_dict= {}
253
+ ex_dict['background'] = ex_human
254
+ ex_dict['layers'] = None
255
+ ex_dict['composite'] = None
256
+ human_ex_list.append(ex_dict)
257
+
258
+ ##default human
259
+
260
+
261
+ image_blocks = gr.Blocks().queue()
262
+ with image_blocks as demo:
263
+ gr.Markdown("## IDM-VTON πŸ‘•πŸ‘”πŸ‘š")
264
+ gr.Markdown("Virtual Try-on with your image and garment image. Check out the [source codes](https://github.com/yisol/IDM-VTON) and the [model](https://huggingface.co/yisol/IDM-VTON)")
265
+ with gr.Row():
266
+ with gr.Column():
267
+ imgs = gr.ImageEditor(sources='upload', type="pil", label='Human. Mask with pen or use auto-masking', interactive=True)
268
+ with gr.Row():
269
+ is_checked = gr.Checkbox(label="Yes", info="Use auto-generated mask (Takes 5 seconds)",value=True)
270
+ with gr.Row():
271
+ is_checked_crop = gr.Checkbox(label="Yes", info="Use auto-crop & resizing",value=False)
272
+
273
+ example = gr.Examples(
274
+ inputs=imgs,
275
+ examples_per_page=10,
276
+ examples=human_ex_list
277
+ )
278
+
279
+ with gr.Column():
280
+ garm_img = gr.Image(label="Garment", sources='upload', type="pil")
281
+ with gr.Row(elem_id="prompt-container"):
282
+ with gr.Row():
283
+ prompt = gr.Textbox(placeholder="Description of garment ex) Short Sleeve Round Neck T-shirts", show_label=False, elem_id="prompt")
284
+ example = gr.Examples(
285
+ inputs=garm_img,
286
+ examples_per_page=8,
287
+ examples=garm_list_path)
288
+ with gr.Column():
289
+ # image_out = gr.Image(label="Output", elem_id="output-img", height=400)
290
+ masked_img = gr.Image(label="Masked image output", elem_id="masked-img",show_share_button=False)
291
+ with gr.Column():
292
+ # image_out = gr.Image(label="Output", elem_id="output-img", height=400)
293
+ image_out = gr.Image(label="Output", elem_id="output-img",show_share_button=False)
294
+
295
+
296
+
297
+
298
+ with gr.Column():
299
+ try_button = gr.Button(value="Try-on")
300
+ with gr.Accordion(label="Advanced Settings", open=False):
301
+ with gr.Row():
302
+ denoise_steps = gr.Number(label="Denoising Steps", minimum=20, maximum=40, value=30, step=1)
303
+ seed = gr.Number(label="Seed", minimum=-1, maximum=2147483647, step=1, value=42)
304
+
305
+
306
+
307
+ try_button.click(fn=start_tryon, inputs=[imgs, garm_img, prompt, is_checked,is_checked_crop, denoise_steps, seed], outputs=[image_out,masked_img], api_name='tryon')
308
+
309
+
310
+
311
+
312
+ image_blocks.launch()
313
+
app_VTON.py ADDED
@@ -0,0 +1,328 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+ import argparse, torch, os
3
+ from PIL import Image
4
+ from src.tryon_pipeline import StableDiffusionXLInpaintPipeline as TryonPipeline
5
+ from src.unet_hacked_garmnet import UNet2DConditionModel as UNet2DConditionModel_ref
6
+ from src.unet_hacked_tryon import UNet2DConditionModel
7
+ from transformers import (
8
+ CLIPImageProcessor,
9
+ CLIPVisionModelWithProjection,
10
+ )
11
+ from diffusers import AutoencoderKL
12
+ from typing import List
13
+ from util.common import open_folder
14
+ from util.image import pil_to_binary_mask, save_output_image
15
+ from utils_mask import get_mask_location
16
+ from torchvision import transforms
17
+ import apply_net
18
+ from preprocess.humanparsing.run_parsing import Parsing
19
+ from preprocess.openpose.run_openpose import OpenPose
20
+ from detectron2.data.detection_utils import convert_PIL_to_numpy,_apply_exif_orientation
21
+ from torchvision.transforms.functional import to_pil_image
22
+ from util.pipeline import quantize_4bit, restart_cpu_offload, torch_gc
23
+
24
+ parser = argparse.ArgumentParser()
25
+ parser.add_argument("--share", type=str, default=False, help="Set to True to share the app publicly.")
26
+ parser.add_argument("--lowvram", action="store_true", help="Enable CPU offload for model operations.")
27
+ parser.add_argument("--load_mode", default=None, type=str, choices=["4bit", "8bit"], help="Quantization mode for optimization memory consumption")
28
+ parser.add_argument("--fixed_vae", action="store_true", default=True, help="Use fixed vae for FP16.")
29
+ args = parser.parse_args()
30
+
31
+ load_mode = args.load_mode
32
+ fixed_vae = args.fixed_vae
33
+
34
+ dtype = torch.float16
35
+ device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
36
+ model_id = 'yisol/IDM-VTON'
37
+ vae_model_id = 'madebyollin/sdxl-vae-fp16-fix'
38
+
39
+ dtypeQuantize = dtype
40
+
41
+ if(load_mode in ('4bit','8bit')):
42
+ dtypeQuantize = torch.float8_e4m3fn
43
+
44
+ ENABLE_CPU_OFFLOAD = args.lowvram
45
+ torch.backends.cudnn.allow_tf32 = False
46
+ torch.backends.cuda.allow_tf32 = False
47
+ need_restart_cpu_offloading = False
48
+
49
+ unet = None
50
+ pipe = None
51
+ UNet_Encoder = None
52
+ example_path = os.path.join(os.path.dirname(__file__), 'example')
53
+
54
+ def start_tryon(dict, garm_img, garment_des, category, is_checked, is_checked_crop, denoise_steps, is_randomize_seed, seed, number_of_images):
55
+ global pipe, unet, UNet_Encoder, need_restart_cpu_offloading
56
+
57
+ if pipe == None:
58
+ unet = UNet2DConditionModel.from_pretrained(
59
+ model_id,
60
+ subfolder="unet",
61
+ torch_dtype=dtypeQuantize,
62
+ )
63
+ if load_mode == '4bit':
64
+ quantize_4bit(unet)
65
+
66
+ unet.requires_grad_(False)
67
+
68
+ image_encoder = CLIPVisionModelWithProjection.from_pretrained(
69
+ model_id,
70
+ subfolder="image_encoder",
71
+ torch_dtype=torch.float16,
72
+ )
73
+ if load_mode == '4bit':
74
+ quantize_4bit(image_encoder)
75
+
76
+ if fixed_vae:
77
+ vae = AutoencoderKL.from_pretrained(vae_model_id, torch_dtype=dtype)
78
+ else:
79
+ vae = AutoencoderKL.from_pretrained(model_id,
80
+ subfolder="vae",
81
+ torch_dtype=dtype,
82
+ )
83
+
84
+ # "stabilityai/stable-diffusion-xl-base-1.0",
85
+ UNet_Encoder = UNet2DConditionModel_ref.from_pretrained(
86
+ model_id,
87
+ subfolder="unet_encoder",
88
+ torch_dtype=dtypeQuantize,
89
+ )
90
+
91
+ if load_mode == '4bit':
92
+ quantize_4bit(UNet_Encoder)
93
+
94
+ UNet_Encoder.requires_grad_(False)
95
+ image_encoder.requires_grad_(False)
96
+ vae.requires_grad_(False)
97
+ unet.requires_grad_(False)
98
+
99
+ pipe_param = {
100
+ 'pretrained_model_name_or_path': model_id,
101
+ 'unet': unet,
102
+ 'torch_dtype': dtype,
103
+ 'vae': vae,
104
+ 'image_encoder': image_encoder,
105
+ 'feature_extractor': CLIPImageProcessor(),
106
+ }
107
+
108
+ pipe = TryonPipeline.from_pretrained(**pipe_param).to(device)
109
+ pipe.unet_encoder = UNet_Encoder
110
+ pipe.unet_encoder.to(pipe.unet.device)
111
+
112
+ if load_mode == '4bit':
113
+ if pipe.text_encoder is not None:
114
+ quantize_4bit(pipe.text_encoder)
115
+ if pipe.text_encoder_2 is not None:
116
+ quantize_4bit(pipe.text_encoder_2)
117
+
118
+ else:
119
+ if ENABLE_CPU_OFFLOAD:
120
+ need_restart_cpu_offloading =True
121
+
122
+ torch_gc()
123
+ parsing_model = Parsing(0)
124
+ openpose_model = OpenPose(0)
125
+ openpose_model.preprocessor.body_estimation.model.to(device)
126
+ tensor_transfrom = transforms.Compose(
127
+ [
128
+ transforms.ToTensor(),
129
+ transforms.Normalize([0.5], [0.5]),
130
+ ]
131
+ )
132
+
133
+ if need_restart_cpu_offloading:
134
+ restart_cpu_offload(pipe, load_mode)
135
+ elif ENABLE_CPU_OFFLOAD:
136
+ pipe.enable_model_cpu_offload()
137
+
138
+ #if load_mode != '4bit' :
139
+ # pipe.enable_xformers_memory_efficient_attention()
140
+
141
+ garm_img= garm_img.convert("RGB").resize((768,1024))
142
+ human_img_orig = dict["background"].convert("RGB")
143
+
144
+ if is_checked_crop:
145
+ width, height = human_img_orig.size
146
+ target_width = int(min(width, height * (3 / 4)))
147
+ target_height = int(min(height, width * (4 / 3)))
148
+ left = (width - target_width) / 2
149
+ top = (height - target_height) / 2
150
+ right = (width + target_width) / 2
151
+ bottom = (height + target_height) / 2
152
+ cropped_img = human_img_orig.crop((left, top, right, bottom))
153
+ crop_size = cropped_img.size
154
+ human_img = cropped_img.resize((768,1024))
155
+ else:
156
+ human_img = human_img_orig.resize((768,1024))
157
+
158
+ if is_checked:
159
+ keypoints = openpose_model(human_img.resize((384,512)))
160
+ model_parse, _ = parsing_model(human_img.resize((384,512)))
161
+ mask, mask_gray = get_mask_location('hd', category, model_parse, keypoints)
162
+ mask = mask.resize((768,1024))
163
+ else:
164
+ mask = pil_to_binary_mask(dict['layers'][0].convert("RGB").resize((768, 1024)))
165
+ # mask = transforms.ToTensor()(mask)
166
+ # mask = mask.unsqueeze(0)
167
+
168
+ mask_gray = (1-transforms.ToTensor()(mask)) * tensor_transfrom(human_img)
169
+ mask_gray = to_pil_image((mask_gray+1.0)/2.0)
170
+
171
+ human_img_arg = _apply_exif_orientation(human_img.resize((384,512)))
172
+ human_img_arg = convert_PIL_to_numpy(human_img_arg, format="BGR")
173
+
174
+ args = apply_net.create_argument_parser().parse_args(('show', './configs/densepose_rcnn_R_50_FPN_s1x.yaml', './ckpt/densepose/model_final_162be9.pkl', 'dp_segm', '-v', '--opts', 'MODEL.DEVICE', 'cuda'))
175
+ # verbosity = getattr(args, "verbosity", None)
176
+ pose_img = args.func(args,human_img_arg)
177
+ pose_img = pose_img[:,:,::-1]
178
+ pose_img = Image.fromarray(pose_img).resize((768,1024))
179
+
180
+ if pipe.text_encoder is not None:
181
+ pipe.text_encoder.to(device)
182
+
183
+ if pipe.text_encoder_2 is not None:
184
+ pipe.text_encoder_2.to(device)
185
+
186
+ with torch.no_grad():
187
+ # Extract the images
188
+ with torch.cuda.amp.autocast(dtype=dtype):
189
+ with torch.no_grad():
190
+ prompt = "model is wearing " + garment_des
191
+ negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
192
+ with torch.inference_mode():
193
+ (
194
+ prompt_embeds,
195
+ negative_prompt_embeds,
196
+ pooled_prompt_embeds,
197
+ negative_pooled_prompt_embeds,
198
+ ) = pipe.encode_prompt(
199
+ prompt,
200
+ num_images_per_prompt=1,
201
+ do_classifier_free_guidance=True,
202
+ negative_prompt=negative_prompt,
203
+ )
204
+
205
+ prompt = "a photo of " + garment_des
206
+ negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
207
+ if not isinstance(prompt, List):
208
+ prompt = [prompt] * 1
209
+ if not isinstance(negative_prompt, List):
210
+ negative_prompt = [negative_prompt] * 1
211
+ with torch.inference_mode():
212
+ (
213
+ prompt_embeds_c,
214
+ _,
215
+ _,
216
+ _,
217
+ ) = pipe.encode_prompt(
218
+ prompt,
219
+ num_images_per_prompt=1,
220
+ do_classifier_free_guidance=False,
221
+ negative_prompt=negative_prompt,
222
+ )
223
+
224
+ pose_img = tensor_transfrom(pose_img).unsqueeze(0).to(device,dtype)
225
+ garm_tensor = tensor_transfrom(garm_img).unsqueeze(0).to(device,dtype)
226
+ results = []
227
+ current_seed = seed
228
+ for i in range(number_of_images):
229
+ if is_randomize_seed:
230
+ current_seed = torch.randint(0, 2**32, size=(1,)).item()
231
+ generator = torch.Generator(device).manual_seed(current_seed) if seed != -1 else None
232
+ current_seed = current_seed + i
233
+
234
+ images = pipe(
235
+ prompt_embeds=prompt_embeds.to(device,dtype),
236
+ negative_prompt_embeds=negative_prompt_embeds.to(device,dtype),
237
+ pooled_prompt_embeds=pooled_prompt_embeds.to(device,dtype),
238
+ negative_pooled_prompt_embeds=negative_pooled_prompt_embeds.to(device,dtype),
239
+ num_inference_steps=denoise_steps,
240
+ generator=generator,
241
+ strength = 1.0,
242
+ pose_img = pose_img.to(device,dtype),
243
+ text_embeds_cloth=prompt_embeds_c.to(device,dtype),
244
+ cloth = garm_tensor.to(device,dtype),
245
+ mask_image=mask,
246
+ image=human_img,
247
+ height=1024,
248
+ width=768,
249
+ ip_adapter_image = garm_img.resize((768,1024)),
250
+ guidance_scale=2.0,
251
+ dtype=dtype,
252
+ device=device,
253
+ )[0]
254
+ if is_checked_crop:
255
+ out_img = images[0].resize(crop_size)
256
+ human_img_orig.paste(out_img, (int(left), int(top)))
257
+ img_path = save_output_image(human_img_orig, base_path="outputs", base_filename='img', seed=current_seed)
258
+ results.append(img_path)
259
+ else:
260
+ img_path = save_output_image(images[0], base_path="outputs", base_filename='img')
261
+ results.append(img_path)
262
+ return results, mask_gray
263
+
264
+ garm_list = os.listdir(os.path.join(example_path,"cloth"))
265
+ garm_list_path = [os.path.join(example_path,"cloth",garm) for garm in garm_list]
266
+
267
+ human_list = os.listdir(os.path.join(example_path,"human"))
268
+ human_list_path = [os.path.join(example_path,"human",human) for human in human_list]
269
+
270
+ human_ex_list = []
271
+ for ex_human in human_list_path:
272
+ if "Jensen" in ex_human or "sam1 (1)" in ex_human:
273
+ ex_dict = {}
274
+ ex_dict['background'] = ex_human
275
+ ex_dict['layers'] = None
276
+ ex_dict['composite'] = None
277
+ human_ex_list.append(ex_dict)
278
+
279
+ image_blocks = gr.Blocks().queue()
280
+ with image_blocks as demo:
281
+ gr.Markdown("## V7 - IDM-VTON πŸ‘•πŸ‘”πŸ‘š improved by SECourses and DEVAIEXP: 1-Click Installers Latest Version On : https://www.patreon.com/posts/103022942")
282
+ gr.Markdown("Virtual Try-on with your image and garment image. Check out the [source codes](https://github.com/yisol/IDM-VTON) and the [model](https://huggingface.co/yisol/IDM-VTON)")
283
+ with gr.Row():
284
+ with gr.Column():
285
+ imgs = gr.ImageEditor(sources='upload', type="pil", label='Human. Mask with pen or use auto-masking', interactive=True)
286
+ with gr.Row():
287
+ category = gr.Radio(choices=["upper_body", "lower_body", "dresses"], label="Select Garment Category", value="upper_body")
288
+ is_checked = gr.Checkbox(label="Yes", info="Use auto-generated mask (Takes 5 seconds)",value=True)
289
+ with gr.Row():
290
+ is_checked_crop = gr.Checkbox(label="Yes", info="Use auto-crop & resizing",value=True)
291
+
292
+ example = gr.Examples(
293
+ inputs=imgs,
294
+ examples_per_page=2,
295
+ examples=human_ex_list
296
+ )
297
+
298
+ with gr.Column():
299
+ garm_img = gr.Image(label="Garment", sources='upload', type="pil")
300
+ with gr.Row(elem_id="prompt-container"):
301
+ with gr.Row():
302
+ prompt = gr.Textbox(placeholder="Description of garment ex) Short Sleeve Round Neck T-shirts", show_label=False, elem_id="prompt")
303
+ example = gr.Examples(
304
+ inputs=garm_img,
305
+ examples_per_page=8,
306
+ examples=garm_list_path)
307
+ with gr.Column():
308
+ with gr.Row():
309
+ # image_out = gr.Image(label="Output", elem_id="output-img", height=400)
310
+ masked_img = gr.Image(label="Masked image output", elem_id="masked-img",show_share_button=False)
311
+ with gr.Row():
312
+ btn_open_outputs = gr.Button("Open Outputs Folder")
313
+ btn_open_outputs.click(fn=open_folder)
314
+ with gr.Column():
315
+ with gr.Row():
316
+ # image_out = gr.Image(label="Output", elem_id="output-img", height=400)
317
+ image_gallery = gr.Gallery(label="Generated Images", show_label=True)
318
+ with gr.Row():
319
+ try_button = gr.Button(value="Try-on")
320
+ denoise_steps = gr.Number(label="Denoising Steps", minimum=20, maximum=120, value=30, step=1)
321
+ seed = gr.Number(label="Seed", minimum=-1, maximum=2147483647, step=1, value=1)
322
+ is_randomize_seed = gr.Checkbox(label="Randomize seed for each generated image", value=True)
323
+ number_of_images = gr.Number(label="Number Of Images To Generate (it will start from your input seed and increment by 1)", minimum=1, maximum=9999, value=1, step=1)
324
+
325
+
326
+ try_button.click(fn=start_tryon, inputs=[imgs, garm_img, prompt, category, is_checked, is_checked_crop, denoise_steps, is_randomize_seed, seed, number_of_images], outputs=[image_gallery, masked_img],api_name='tryon')
327
+
328
+ image_blocks.launch(inbrowser=True,share=args.share)
apply_net.py ADDED
@@ -0,0 +1,359 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ # Copyright (c) Facebook, Inc. and its affiliates.
3
+
4
+ import argparse
5
+ import glob
6
+ import logging
7
+ import os
8
+ import sys
9
+ from typing import Any, ClassVar, Dict, List
10
+ import torch
11
+
12
+ from detectron2.config import CfgNode, get_cfg
13
+ from detectron2.data.detection_utils import read_image
14
+ from detectron2.engine.defaults import DefaultPredictor
15
+ from detectron2.structures.instances import Instances
16
+ from detectron2.utils.logger import setup_logger
17
+
18
+ from densepose import add_densepose_config
19
+ from densepose.structures import DensePoseChartPredictorOutput, DensePoseEmbeddingPredictorOutput
20
+ from densepose.utils.logger import verbosity_to_level
21
+ from densepose.vis.base import CompoundVisualizer
22
+ from densepose.vis.bounding_box import ScoredBoundingBoxVisualizer
23
+ from densepose.vis.densepose_outputs_vertex import (
24
+ DensePoseOutputsTextureVisualizer,
25
+ DensePoseOutputsVertexVisualizer,
26
+ get_texture_atlases,
27
+ )
28
+ from densepose.vis.densepose_results import (
29
+ DensePoseResultsContourVisualizer,
30
+ DensePoseResultsFineSegmentationVisualizer,
31
+ DensePoseResultsUVisualizer,
32
+ DensePoseResultsVVisualizer,
33
+ )
34
+ from densepose.vis.densepose_results_textures import (
35
+ DensePoseResultsVisualizerWithTexture,
36
+ get_texture_atlas,
37
+ )
38
+ from densepose.vis.extractor import (
39
+ CompoundExtractor,
40
+ DensePoseOutputsExtractor,
41
+ DensePoseResultExtractor,
42
+ create_extractor,
43
+ )
44
+
45
+ DOC = """Apply Net - a tool to print / visualize DensePose results
46
+ """
47
+
48
+ LOGGER_NAME = "apply_net"
49
+ logger = logging.getLogger(LOGGER_NAME)
50
+
51
+ _ACTION_REGISTRY: Dict[str, "Action"] = {}
52
+
53
+
54
+ class Action:
55
+ @classmethod
56
+ def add_arguments(cls: type, parser: argparse.ArgumentParser):
57
+ parser.add_argument(
58
+ "-v",
59
+ "--verbosity",
60
+ action="count",
61
+ help="Verbose mode. Multiple -v options increase the verbosity.",
62
+ )
63
+
64
+
65
+ def register_action(cls: type):
66
+ """
67
+ Decorator for action classes to automate action registration
68
+ """
69
+ global _ACTION_REGISTRY
70
+ _ACTION_REGISTRY[cls.COMMAND] = cls
71
+ return cls
72
+
73
+
74
+ class InferenceAction(Action):
75
+ @classmethod
76
+ def add_arguments(cls: type, parser: argparse.ArgumentParser):
77
+ super(InferenceAction, cls).add_arguments(parser)
78
+ parser.add_argument("cfg", metavar="<config>", help="Config file")
79
+ parser.add_argument("model", metavar="<model>", help="Model file")
80
+ parser.add_argument(
81
+ "--opts",
82
+ help="Modify config options using the command-line 'KEY VALUE' pairs",
83
+ default=[],
84
+ nargs=argparse.REMAINDER,
85
+ )
86
+
87
+ @classmethod
88
+ def execute(cls: type, args: argparse.Namespace, human_img):
89
+ logger.info(f"Loading config from {args.cfg}")
90
+ opts = []
91
+ cfg = cls.setup_config(args.cfg, args.model, args, opts)
92
+ logger.info(f"Loading model from {args.model}")
93
+ predictor = DefaultPredictor(cfg)
94
+ # logger.info(f"Loading data from {args.input}")
95
+ # file_list = cls._get_input_file_list(args.input)
96
+ # if len(file_list) == 0:
97
+ # logger.warning(f"No input images for {args.input}")
98
+ # return
99
+ context = cls.create_context(args, cfg)
100
+ # for file_name in file_list:
101
+ # img = read_image(file_name, format="BGR") # predictor expects BGR image.
102
+ with torch.no_grad():
103
+ outputs = predictor(human_img)["instances"]
104
+ out_pose = cls.execute_on_outputs(context, {"image": human_img}, outputs)
105
+ cls.postexecute(context)
106
+ return out_pose
107
+
108
+ @classmethod
109
+ def setup_config(
110
+ cls: type, config_fpath: str, model_fpath: str, args: argparse.Namespace, opts: List[str]
111
+ ):
112
+ cfg = get_cfg()
113
+ add_densepose_config(cfg)
114
+ cfg.merge_from_file(config_fpath)
115
+ cfg.merge_from_list(args.opts)
116
+ if opts:
117
+ cfg.merge_from_list(opts)
118
+ cfg.MODEL.WEIGHTS = model_fpath
119
+ cfg.freeze()
120
+ return cfg
121
+
122
+ @classmethod
123
+ def _get_input_file_list(cls: type, input_spec: str):
124
+ if os.path.isdir(input_spec):
125
+ file_list = [
126
+ os.path.join(input_spec, fname)
127
+ for fname in os.listdir(input_spec)
128
+ if os.path.isfile(os.path.join(input_spec, fname))
129
+ ]
130
+ elif os.path.isfile(input_spec):
131
+ file_list = [input_spec]
132
+ else:
133
+ file_list = glob.glob(input_spec)
134
+ return file_list
135
+
136
+
137
+ @register_action
138
+ class DumpAction(InferenceAction):
139
+ """
140
+ Dump action that outputs results to a pickle file
141
+ """
142
+
143
+ COMMAND: ClassVar[str] = "dump"
144
+
145
+ @classmethod
146
+ def add_parser(cls: type, subparsers: argparse._SubParsersAction):
147
+ parser = subparsers.add_parser(cls.COMMAND, help="Dump model outputs to a file.")
148
+ cls.add_arguments(parser)
149
+ parser.set_defaults(func=cls.execute)
150
+
151
+ @classmethod
152
+ def add_arguments(cls: type, parser: argparse.ArgumentParser):
153
+ super(DumpAction, cls).add_arguments(parser)
154
+ parser.add_argument(
155
+ "--output",
156
+ metavar="<dump_file>",
157
+ default="results.pkl",
158
+ help="File name to save dump to",
159
+ )
160
+
161
+ @classmethod
162
+ def execute_on_outputs(
163
+ cls: type, context: Dict[str, Any], entry: Dict[str, Any], outputs: Instances
164
+ ):
165
+ image_fpath = entry["file_name"]
166
+ logger.info(f"Processing {image_fpath}")
167
+ result = {"file_name": image_fpath}
168
+ if outputs.has("scores"):
169
+ result["scores"] = outputs.get("scores").cpu()
170
+ if outputs.has("pred_boxes"):
171
+ result["pred_boxes_XYXY"] = outputs.get("pred_boxes").tensor.cpu()
172
+ if outputs.has("pred_densepose"):
173
+ if isinstance(outputs.pred_densepose, DensePoseChartPredictorOutput):
174
+ extractor = DensePoseResultExtractor()
175
+ elif isinstance(outputs.pred_densepose, DensePoseEmbeddingPredictorOutput):
176
+ extractor = DensePoseOutputsExtractor()
177
+ result["pred_densepose"] = extractor(outputs)[0]
178
+ context["results"].append(result)
179
+
180
+ @classmethod
181
+ def create_context(cls: type, args: argparse.Namespace, cfg: CfgNode):
182
+ context = {"results": [], "out_fname": args.output}
183
+ return context
184
+
185
+ @classmethod
186
+ def postexecute(cls: type, context: Dict[str, Any]):
187
+ out_fname = context["out_fname"]
188
+ out_dir = os.path.dirname(out_fname)
189
+ if len(out_dir) > 0 and not os.path.exists(out_dir):
190
+ os.makedirs(out_dir)
191
+ with open(out_fname, "wb") as hFile:
192
+ torch.save(context["results"], hFile)
193
+ logger.info(f"Output saved to {out_fname}")
194
+
195
+
196
+ @register_action
197
+ class ShowAction(InferenceAction):
198
+ """
199
+ Show action that visualizes selected entries on an image
200
+ """
201
+
202
+ COMMAND: ClassVar[str] = "show"
203
+ VISUALIZERS: ClassVar[Dict[str, object]] = {
204
+ "dp_contour": DensePoseResultsContourVisualizer,
205
+ "dp_segm": DensePoseResultsFineSegmentationVisualizer,
206
+ "dp_u": DensePoseResultsUVisualizer,
207
+ "dp_v": DensePoseResultsVVisualizer,
208
+ "dp_iuv_texture": DensePoseResultsVisualizerWithTexture,
209
+ "dp_cse_texture": DensePoseOutputsTextureVisualizer,
210
+ "dp_vertex": DensePoseOutputsVertexVisualizer,
211
+ "bbox": ScoredBoundingBoxVisualizer,
212
+ }
213
+
214
+ @classmethod
215
+ def add_parser(cls: type, subparsers: argparse._SubParsersAction):
216
+ parser = subparsers.add_parser(cls.COMMAND, help="Visualize selected entries")
217
+ cls.add_arguments(parser)
218
+ parser.set_defaults(func=cls.execute)
219
+
220
+ @classmethod
221
+ def add_arguments(cls: type, parser: argparse.ArgumentParser):
222
+ super(ShowAction, cls).add_arguments(parser)
223
+ parser.add_argument(
224
+ "visualizations",
225
+ metavar="<visualizations>",
226
+ help="Comma separated list of visualizations, possible values: "
227
+ "[{}]".format(",".join(sorted(cls.VISUALIZERS.keys()))),
228
+ )
229
+ parser.add_argument(
230
+ "--min_score",
231
+ metavar="<score>",
232
+ default=0.8,
233
+ type=float,
234
+ help="Minimum detection score to visualize",
235
+ )
236
+ parser.add_argument(
237
+ "--nms_thresh", metavar="<threshold>", default=None, type=float, help="NMS threshold"
238
+ )
239
+ parser.add_argument(
240
+ "--texture_atlas",
241
+ metavar="<texture_atlas>",
242
+ default=None,
243
+ help="Texture atlas file (for IUV texture transfer)",
244
+ )
245
+ parser.add_argument(
246
+ "--texture_atlases_map",
247
+ metavar="<texture_atlases_map>",
248
+ default=None,
249
+ help="JSON string of a dict containing texture atlas files for each mesh",
250
+ )
251
+ parser.add_argument(
252
+ "--output",
253
+ metavar="<image_file>",
254
+ default="outputres.png",
255
+ help="File name to save output to",
256
+ )
257
+
258
+ @classmethod
259
+ def setup_config(
260
+ cls: type, config_fpath: str, model_fpath: str, args: argparse.Namespace, opts: List[str]
261
+ ):
262
+ opts.append("MODEL.ROI_HEADS.SCORE_THRESH_TEST")
263
+ opts.append(str(args.min_score))
264
+ if args.nms_thresh is not None:
265
+ opts.append("MODEL.ROI_HEADS.NMS_THRESH_TEST")
266
+ opts.append(str(args.nms_thresh))
267
+ cfg = super(ShowAction, cls).setup_config(config_fpath, model_fpath, args, opts)
268
+ return cfg
269
+
270
+ @classmethod
271
+ def execute_on_outputs(
272
+ cls: type, context: Dict[str, Any], entry: Dict[str, Any], outputs: Instances
273
+ ):
274
+ import cv2
275
+ import numpy as np
276
+ visualizer = context["visualizer"]
277
+ extractor = context["extractor"]
278
+ # image_fpath = entry["file_name"]
279
+ # logger.info(f"Processing {image_fpath}")
280
+ image = cv2.cvtColor(entry["image"], cv2.COLOR_BGR2GRAY)
281
+ image = np.tile(image[:, :, np.newaxis], [1, 1, 3])
282
+ data = extractor(outputs)
283
+ image_vis = visualizer.visualize(image, data)
284
+
285
+ return image_vis
286
+ entry_idx = context["entry_idx"] + 1
287
+ out_fname = './image-densepose/' + image_fpath.split('/')[-1]
288
+ out_dir = './image-densepose'
289
+ out_dir = os.path.dirname(out_fname)
290
+ if len(out_dir) > 0 and not os.path.exists(out_dir):
291
+ os.makedirs(out_dir)
292
+ cv2.imwrite(out_fname, image_vis)
293
+ logger.info(f"Output saved to {out_fname}")
294
+ context["entry_idx"] += 1
295
+
296
+ @classmethod
297
+ def postexecute(cls: type, context: Dict[str, Any]):
298
+ pass
299
+ # python ./apply_net.py show ./configs/densepose_rcnn_R_50_FPN_s1x.yaml https://dl.fbaipublicfiles.com/densepose/densepose_rcnn_R_50_FPN_s1x/165712039/model_final_162be9.pkl /home/alin0222/DressCode/upper_body/images dp_segm -v --opts MODEL.DEVICE cpu
300
+
301
+ @classmethod
302
+ def _get_out_fname(cls: type, entry_idx: int, fname_base: str):
303
+ base, ext = os.path.splitext(fname_base)
304
+ return base + ".{0:04d}".format(entry_idx) + ext
305
+
306
+ @classmethod
307
+ def create_context(cls: type, args: argparse.Namespace, cfg: CfgNode) -> Dict[str, Any]:
308
+ vis_specs = args.visualizations.split(",")
309
+ visualizers = []
310
+ extractors = []
311
+ for vis_spec in vis_specs:
312
+ texture_atlas = get_texture_atlas(args.texture_atlas)
313
+ texture_atlases_dict = get_texture_atlases(args.texture_atlases_map)
314
+ vis = cls.VISUALIZERS[vis_spec](
315
+ cfg=cfg,
316
+ texture_atlas=texture_atlas,
317
+ texture_atlases_dict=texture_atlases_dict,
318
+ )
319
+ visualizers.append(vis)
320
+ extractor = create_extractor(vis)
321
+ extractors.append(extractor)
322
+ visualizer = CompoundVisualizer(visualizers)
323
+ extractor = CompoundExtractor(extractors)
324
+ context = {
325
+ "extractor": extractor,
326
+ "visualizer": visualizer,
327
+ "out_fname": args.output,
328
+ "entry_idx": 0,
329
+ }
330
+ return context
331
+
332
+
333
+ def create_argument_parser() -> argparse.ArgumentParser:
334
+ parser = argparse.ArgumentParser(
335
+ description=DOC,
336
+ formatter_class=lambda prog: argparse.HelpFormatter(prog, max_help_position=120),
337
+ )
338
+ parser.set_defaults(func=lambda _: parser.print_help(sys.stdout))
339
+ subparsers = parser.add_subparsers(title="Actions")
340
+ for _, action in _ACTION_REGISTRY.items():
341
+ action.add_parser(subparsers)
342
+ return parser
343
+
344
+
345
+ def main():
346
+ parser = create_argument_parser()
347
+ args = parser.parse_args()
348
+ verbosity = getattr(args, "verbosity", None)
349
+ global logger
350
+ logger = setup_logger(name=LOGGER_NAME)
351
+ logger.setLevel(verbosity_to_level(verbosity))
352
+ args.func(args)
353
+
354
+
355
+ if __name__ == "__main__":
356
+ main()
357
+
358
+
359
+ # python ./apply_net.py show ./configs/densepose_rcnn_R_50_FPN_s1x.yaml https://dl.fbaipublicfiles.com/densepose/densepose_rcnn_R_50_FPN_s1x/165712039/model_final_162be9.pkl /home/alin0222/Dresscode/dresses/humanonly dp_segm -v --opts MODEL.DEVICE cuda
ckpt/densepose/model_final_162be9.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b8a7382001b16e453bad95ca9dbc68ae8f2b839b304cf90eaf5c27fbdb4dae91
3
+ size 255757821
ckpt/humanparsing/parsing_atr.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:04c7d1d070d0e0ae943d86b18cb5aaaea9e278d97462e9cfb270cbbe4cd977f4
3
+ size 266859305
ckpt/humanparsing/parsing_lip.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8436e1dae96e2601c373d1ace29c8f0978b16357d9038c17a8ba756cca376dbc
3
+ size 266863411
ckpt/openpose/.DS_Store ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e953475b1378e1d0566f8ad8de20077ce8610ae23fb2b5f8bfe57104aca8e911
3
+ size 6148
ckpt/openpose/ckpts/body_pose_model.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:25a948c16078b0f08e236bda51a385d855ef4c153598947c28c0d47ed94bb746
3
+ size 209267595
configs/Base-DensePose-RCNN-FPN.yaml ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ VERSION: 2
2
+ MODEL:
3
+ META_ARCHITECTURE: "GeneralizedRCNN"
4
+ BACKBONE:
5
+ NAME: "build_resnet_fpn_backbone"
6
+ RESNETS:
7
+ OUT_FEATURES: ["res2", "res3", "res4", "res5"]
8
+ FPN:
9
+ IN_FEATURES: ["res2", "res3", "res4", "res5"]
10
+ ANCHOR_GENERATOR:
11
+ SIZES: [[32], [64], [128], [256], [512]] # One size for each in feature map
12
+ ASPECT_RATIOS: [[0.5, 1.0, 2.0]] # Three aspect ratios (same for all in feature maps)
13
+ RPN:
14
+ IN_FEATURES: ["p2", "p3", "p4", "p5", "p6"]
15
+ PRE_NMS_TOPK_TRAIN: 2000 # Per FPN level
16
+ PRE_NMS_TOPK_TEST: 1000 # Per FPN level
17
+ # Detectron1 uses 2000 proposals per-batch,
18
+ # (See "modeling/rpn/rpn_outputs.py" for details of this legacy issue)
19
+ # which is approximately 1000 proposals per-image since the default batch size for FPN is 2.
20
+ POST_NMS_TOPK_TRAIN: 1000
21
+ POST_NMS_TOPK_TEST: 1000
22
+
23
+ DENSEPOSE_ON: True
24
+ ROI_HEADS:
25
+ NAME: "DensePoseROIHeads"
26
+ IN_FEATURES: ["p2", "p3", "p4", "p5"]
27
+ NUM_CLASSES: 1
28
+ ROI_BOX_HEAD:
29
+ NAME: "FastRCNNConvFCHead"
30
+ NUM_FC: 2
31
+ POOLER_RESOLUTION: 7
32
+ POOLER_SAMPLING_RATIO: 2
33
+ POOLER_TYPE: "ROIAlign"
34
+ ROI_DENSEPOSE_HEAD:
35
+ NAME: "DensePoseV1ConvXHead"
36
+ POOLER_TYPE: "ROIAlign"
37
+ NUM_COARSE_SEGM_CHANNELS: 2
38
+ DATASETS:
39
+ TRAIN: ("densepose_coco_2014_train", "densepose_coco_2014_valminusminival")
40
+ TEST: ("densepose_coco_2014_minival",)
41
+ SOLVER:
42
+ IMS_PER_BATCH: 16
43
+ BASE_LR: 0.01
44
+ STEPS: (60000, 80000)
45
+ MAX_ITER: 90000
46
+ WARMUP_FACTOR: 0.1
47
+ INPUT:
48
+ MIN_SIZE_TRAIN: (640, 672, 704, 736, 768, 800)
configs/HRNet/densepose_rcnn_HRFPN_HRNet_w32_s1x.yaml ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "../Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://1drv.ms/u/s!Aus8VCZ_C_33dYBMemi9xOUFR0w"
4
+ BACKBONE:
5
+ NAME: "build_hrfpn_backbone"
6
+ RPN:
7
+ IN_FEATURES: ['p1', 'p2', 'p3', 'p4', 'p5']
8
+ ROI_HEADS:
9
+ IN_FEATURES: ['p1', 'p2', 'p3', 'p4', 'p5']
10
+ SOLVER:
11
+ MAX_ITER: 130000
12
+ STEPS: (100000, 120000)
13
+ CLIP_GRADIENTS:
14
+ ENABLED: True
15
+ CLIP_TYPE: "norm"
16
+ BASE_LR: 0.03
configs/HRNet/densepose_rcnn_HRFPN_HRNet_w40_s1x.yaml ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "../Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://1drv.ms/u/s!Aus8VCZ_C_33ck0gvo5jfoWBOPo"
4
+ BACKBONE:
5
+ NAME: "build_hrfpn_backbone"
6
+ RPN:
7
+ IN_FEATURES: ['p1', 'p2', 'p3', 'p4', 'p5']
8
+ ROI_HEADS:
9
+ IN_FEATURES: ['p1', 'p2', 'p3', 'p4', 'p5']
10
+ HRNET:
11
+ STAGE2:
12
+ NUM_CHANNELS: [40, 80]
13
+ STAGE3:
14
+ NUM_CHANNELS: [40, 80, 160]
15
+ STAGE4:
16
+ NUM_CHANNELS: [40, 80, 160, 320]
17
+ SOLVER:
18
+ MAX_ITER: 130000
19
+ STEPS: (100000, 120000)
20
+ CLIP_GRADIENTS:
21
+ ENABLED: True
22
+ CLIP_TYPE: "norm"
23
+ BASE_LR: 0.03
configs/HRNet/densepose_rcnn_HRFPN_HRNet_w48_s1x.yaml ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "../Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://1drv.ms/u/s!Aus8VCZ_C_33dKvqI6pBZlifgJk"
4
+ BACKBONE:
5
+ NAME: "build_hrfpn_backbone"
6
+ RPN:
7
+ IN_FEATURES: ['p1', 'p2', 'p3', 'p4', 'p5']
8
+ ROI_HEADS:
9
+ IN_FEATURES: ['p1', 'p2', 'p3', 'p4', 'p5']
10
+ HRNET:
11
+ STAGE2:
12
+ NUM_CHANNELS: [48, 96]
13
+ STAGE3:
14
+ NUM_CHANNELS: [48, 96, 192]
15
+ STAGE4:
16
+ NUM_CHANNELS: [48, 96, 192, 384]
17
+ SOLVER:
18
+ MAX_ITER: 130000
19
+ STEPS: (100000, 120000)
20
+ CLIP_GRADIENTS:
21
+ ENABLED: True
22
+ CLIP_TYPE: "norm"
23
+ BASE_LR: 0.03
configs/cse/Base-DensePose-RCNN-FPN-Human.yaml ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ ROI_DENSEPOSE_HEAD:
4
+ CSE:
5
+ EMBEDDERS:
6
+ "smpl_27554":
7
+ TYPE: vertex_feature
8
+ NUM_VERTICES: 27554
9
+ FEATURE_DIM: 256
10
+ FEATURES_TRAINABLE: False
11
+ IS_TRAINABLE: True
12
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_smpl_27554_256.pkl"
13
+ DATASETS:
14
+ TRAIN:
15
+ - "densepose_coco_2014_train_cse"
16
+ - "densepose_coco_2014_valminusminival_cse"
17
+ TEST:
18
+ - "densepose_coco_2014_minival_cse"
19
+ CLASS_TO_MESH_NAME_MAPPING:
20
+ "0": "smpl_27554"
configs/cse/Base-DensePose-RCNN-FPN.yaml ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ VERSION: 2
2
+ MODEL:
3
+ META_ARCHITECTURE: "GeneralizedRCNN"
4
+ BACKBONE:
5
+ NAME: "build_resnet_fpn_backbone"
6
+ RESNETS:
7
+ OUT_FEATURES: ["res2", "res3", "res4", "res5"]
8
+ FPN:
9
+ IN_FEATURES: ["res2", "res3", "res4", "res5"]
10
+ ANCHOR_GENERATOR:
11
+ SIZES: [[32], [64], [128], [256], [512]] # One size for each in feature map
12
+ ASPECT_RATIOS: [[0.5, 1.0, 2.0]] # Three aspect ratios (same for all in feature maps)
13
+ RPN:
14
+ IN_FEATURES: ["p2", "p3", "p4", "p5", "p6"]
15
+ PRE_NMS_TOPK_TRAIN: 2000 # Per FPN level
16
+ PRE_NMS_TOPK_TEST: 1000 # Per FPN level
17
+ # Detectron1 uses 2000 proposals per-batch,
18
+ # (See "modeling/rpn/rpn_outputs.py" for details of this legacy issue)
19
+ # which is approximately 1000 proposals per-image since the default batch size for FPN is 2.
20
+ POST_NMS_TOPK_TRAIN: 1000
21
+ POST_NMS_TOPK_TEST: 1000
22
+
23
+ DENSEPOSE_ON: True
24
+ ROI_HEADS:
25
+ NAME: "DensePoseROIHeads"
26
+ IN_FEATURES: ["p2", "p3", "p4", "p5"]
27
+ NUM_CLASSES: 1
28
+ ROI_BOX_HEAD:
29
+ NAME: "FastRCNNConvFCHead"
30
+ NUM_FC: 2
31
+ POOLER_RESOLUTION: 7
32
+ POOLER_SAMPLING_RATIO: 2
33
+ POOLER_TYPE: "ROIAlign"
34
+ ROI_DENSEPOSE_HEAD:
35
+ NAME: "DensePoseV1ConvXHead"
36
+ POOLER_TYPE: "ROIAlign"
37
+ NUM_COARSE_SEGM_CHANNELS: 2
38
+ PREDICTOR_NAME: "DensePoseEmbeddingPredictor"
39
+ LOSS_NAME: "DensePoseCseLoss"
40
+ CSE:
41
+ # embedding loss, possible values:
42
+ # - "EmbeddingLoss"
43
+ # - "SoftEmbeddingLoss"
44
+ EMBED_LOSS_NAME: "EmbeddingLoss"
45
+ SOLVER:
46
+ IMS_PER_BATCH: 16
47
+ BASE_LR: 0.01
48
+ STEPS: (60000, 80000)
49
+ MAX_ITER: 90000
50
+ WARMUP_FACTOR: 0.1
51
+ CLIP_GRADIENTS:
52
+ CLIP_TYPE: norm
53
+ CLIP_VALUE: 1.0
54
+ ENABLED: true
55
+ NORM_TYPE: 2.0
56
+ INPUT:
57
+ MIN_SIZE_TRAIN: (640, 672, 704, 736, 768, 800)
58
+ DENSEPOSE_EVALUATION:
59
+ TYPE: cse
60
+ STORAGE: file
configs/cse/densepose_rcnn_R_101_FPN_DL_s1x.yaml ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN-Human.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ CSE:
9
+ EMBED_LOSS_NAME: "EmbeddingLoss"
10
+ SOLVER:
11
+ MAX_ITER: 130000
12
+ STEPS: (100000, 120000)
configs/cse/densepose_rcnn_R_101_FPN_DL_soft_s1x.yaml ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN-Human.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ CSE:
9
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
10
+ SOLVER:
11
+ MAX_ITER: 130000
12
+ STEPS: (100000, 120000)
configs/cse/densepose_rcnn_R_101_FPN_s1x.yaml ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN-Human.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseV1ConvXHead"
8
+ CSE:
9
+ EMBED_LOSS_NAME: "EmbeddingLoss"
10
+ SOLVER:
11
+ MAX_ITER: 130000
12
+ STEPS: (100000, 120000)
configs/cse/densepose_rcnn_R_101_FPN_soft_s1x.yaml ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN-Human.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseV1ConvXHead"
8
+ CSE:
9
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
10
+ SOLVER:
11
+ MAX_ITER: 130000
12
+ STEPS: (100000, 120000)
configs/cse/densepose_rcnn_R_50_FPN_DL_s1x.yaml ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN-Human.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-50.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ CSE:
9
+ EMBED_LOSS_NAME: "EmbeddingLoss"
10
+ SOLVER:
11
+ MAX_ITER: 130000
12
+ STEPS: (100000, 120000)
configs/cse/densepose_rcnn_R_50_FPN_DL_soft_s1x.yaml ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN-Human.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-50.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ CSE:
9
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
10
+ SOLVER:
11
+ MAX_ITER: 130000
12
+ STEPS: (100000, 120000)
configs/cse/densepose_rcnn_R_50_FPN_s1x.yaml ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN-Human.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-50.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseV1ConvXHead"
8
+ CSE:
9
+ EMBED_LOSS_NAME: "EmbeddingLoss"
10
+ SOLVER:
11
+ MAX_ITER: 130000
12
+ STEPS: (100000, 120000)
configs/cse/densepose_rcnn_R_50_FPN_soft_animals_CA_finetune_16k.yaml ADDED
@@ -0,0 +1,133 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://dl.fbaipublicfiles.com/densepose/cse/densepose_rcnn_R_50_FPN_soft_s1x/250533982/model_final_2c4512.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_HEADS:
7
+ NUM_CLASSES: 1
8
+ ROI_DENSEPOSE_HEAD:
9
+ NAME: "DensePoseV1ConvXHead"
10
+ COARSE_SEGM_TRAINED_BY_MASKS: True
11
+ CSE:
12
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
13
+ EMBEDDING_DIST_GAUSS_SIGMA: 0.1
14
+ GEODESIC_DIST_GAUSS_SIGMA: 0.1
15
+ EMBEDDERS:
16
+ "cat_7466":
17
+ TYPE: vertex_feature
18
+ NUM_VERTICES: 7466
19
+ FEATURE_DIM: 256
20
+ FEATURES_TRAINABLE: False
21
+ IS_TRAINABLE: True
22
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cat_7466_256.pkl"
23
+ "dog_7466":
24
+ TYPE: vertex_feature
25
+ NUM_VERTICES: 7466
26
+ FEATURE_DIM: 256
27
+ FEATURES_TRAINABLE: False
28
+ IS_TRAINABLE: True
29
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_dog_7466_256.pkl"
30
+ "sheep_5004":
31
+ TYPE: vertex_feature
32
+ NUM_VERTICES: 5004
33
+ FEATURE_DIM: 256
34
+ FEATURES_TRAINABLE: False
35
+ IS_TRAINABLE: True
36
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_sheep_5004_256.pkl"
37
+ "horse_5004":
38
+ TYPE: vertex_feature
39
+ NUM_VERTICES: 5004
40
+ FEATURE_DIM: 256
41
+ FEATURES_TRAINABLE: False
42
+ IS_TRAINABLE: True
43
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_horse_5004_256.pkl"
44
+ "zebra_5002":
45
+ TYPE: vertex_feature
46
+ NUM_VERTICES: 5002
47
+ FEATURE_DIM: 256
48
+ FEATURES_TRAINABLE: False
49
+ IS_TRAINABLE: True
50
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_zebra_5002_256.pkl"
51
+ "giraffe_5002":
52
+ TYPE: vertex_feature
53
+ NUM_VERTICES: 5002
54
+ FEATURE_DIM: 256
55
+ FEATURES_TRAINABLE: False
56
+ IS_TRAINABLE: True
57
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_giraffe_5002_256.pkl"
58
+ "elephant_5002":
59
+ TYPE: vertex_feature
60
+ NUM_VERTICES: 5002
61
+ FEATURE_DIM: 256
62
+ FEATURES_TRAINABLE: False
63
+ IS_TRAINABLE: True
64
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_elephant_5002_256.pkl"
65
+ "cow_5002":
66
+ TYPE: vertex_feature
67
+ NUM_VERTICES: 5002
68
+ FEATURE_DIM: 256
69
+ FEATURES_TRAINABLE: False
70
+ IS_TRAINABLE: True
71
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cow_5002_256.pkl"
72
+ "bear_4936":
73
+ TYPE: vertex_feature
74
+ NUM_VERTICES: 4936
75
+ FEATURE_DIM: 256
76
+ FEATURES_TRAINABLE: False
77
+ IS_TRAINABLE: True
78
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_bear_4936_256.pkl"
79
+ DATASETS:
80
+ TRAIN:
81
+ - "densepose_lvis_v1_ds2_train_v1"
82
+ TEST:
83
+ - "densepose_lvis_v1_ds2_val_v1"
84
+ WHITELISTED_CATEGORIES:
85
+ "densepose_lvis_v1_ds2_train_v1":
86
+ - 943 # sheep
87
+ - 1202 # zebra
88
+ - 569 # horse
89
+ - 496 # giraffe
90
+ - 422 # elephant
91
+ - 80 # cow
92
+ - 76 # bear
93
+ - 225 # cat
94
+ - 378 # dog
95
+ "densepose_lvis_v1_ds2_val_v1":
96
+ - 943 # sheep
97
+ - 1202 # zebra
98
+ - 569 # horse
99
+ - 496 # giraffe
100
+ - 422 # elephant
101
+ - 80 # cow
102
+ - 76 # bear
103
+ - 225 # cat
104
+ - 378 # dog
105
+ CATEGORY_MAPS:
106
+ "densepose_lvis_v1_ds2_train_v1":
107
+ "1202": 943 # zebra -> sheep
108
+ "569": 943 # horse -> sheep
109
+ "496": 943 # giraffe -> sheep
110
+ "422": 943 # elephant -> sheep
111
+ "80": 943 # cow -> sheep
112
+ "76": 943 # bear -> sheep
113
+ "225": 943 # cat -> sheep
114
+ "378": 943 # dog -> sheep
115
+ "densepose_lvis_v1_ds2_val_v1":
116
+ "1202": 943 # zebra -> sheep
117
+ "569": 943 # horse -> sheep
118
+ "496": 943 # giraffe -> sheep
119
+ "422": 943 # elephant -> sheep
120
+ "80": 943 # cow -> sheep
121
+ "76": 943 # bear -> sheep
122
+ "225": 943 # cat -> sheep
123
+ "378": 943 # dog -> sheep
124
+ CLASS_TO_MESH_NAME_MAPPING:
125
+ # Note: different classes are mapped to a single class
126
+ # mesh is chosen based on GT data, so this is just some
127
+ # value which has no particular meaning
128
+ "0": "sheep_5004"
129
+ SOLVER:
130
+ MAX_ITER: 16000
131
+ STEPS: (12000, 14000)
132
+ DENSEPOSE_EVALUATION:
133
+ EVALUATE_MESH_ALIGNMENT: True
configs/cse/densepose_rcnn_R_50_FPN_soft_animals_CA_finetune_4k.yaml ADDED
@@ -0,0 +1,133 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://dl.fbaipublicfiles.com/densepose/cse/densepose_rcnn_R_50_FPN_soft_s1x/250533982/model_final_2c4512.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_HEADS:
7
+ NUM_CLASSES: 1
8
+ ROI_DENSEPOSE_HEAD:
9
+ NAME: "DensePoseV1ConvXHead"
10
+ COARSE_SEGM_TRAINED_BY_MASKS: True
11
+ CSE:
12
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
13
+ EMBEDDING_DIST_GAUSS_SIGMA: 0.1
14
+ GEODESIC_DIST_GAUSS_SIGMA: 0.1
15
+ EMBEDDERS:
16
+ "cat_5001":
17
+ TYPE: vertex_feature
18
+ NUM_VERTICES: 5001
19
+ FEATURE_DIM: 256
20
+ FEATURES_TRAINABLE: False
21
+ IS_TRAINABLE: True
22
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cat_5001_256.pkl"
23
+ "dog_5002":
24
+ TYPE: vertex_feature
25
+ NUM_VERTICES: 5002
26
+ FEATURE_DIM: 256
27
+ FEATURES_TRAINABLE: False
28
+ IS_TRAINABLE: True
29
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_dog_5002_256.pkl"
30
+ "sheep_5004":
31
+ TYPE: vertex_feature
32
+ NUM_VERTICES: 5004
33
+ FEATURE_DIM: 256
34
+ FEATURES_TRAINABLE: False
35
+ IS_TRAINABLE: True
36
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_sheep_5004_256.pkl"
37
+ "horse_5004":
38
+ TYPE: vertex_feature
39
+ NUM_VERTICES: 5004
40
+ FEATURE_DIM: 256
41
+ FEATURES_TRAINABLE: False
42
+ IS_TRAINABLE: True
43
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_horse_5004_256.pkl"
44
+ "zebra_5002":
45
+ TYPE: vertex_feature
46
+ NUM_VERTICES: 5002
47
+ FEATURE_DIM: 256
48
+ FEATURES_TRAINABLE: False
49
+ IS_TRAINABLE: True
50
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_zebra_5002_256.pkl"
51
+ "giraffe_5002":
52
+ TYPE: vertex_feature
53
+ NUM_VERTICES: 5002
54
+ FEATURE_DIM: 256
55
+ FEATURES_TRAINABLE: False
56
+ IS_TRAINABLE: True
57
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_giraffe_5002_256.pkl"
58
+ "elephant_5002":
59
+ TYPE: vertex_feature
60
+ NUM_VERTICES: 5002
61
+ FEATURE_DIM: 256
62
+ FEATURES_TRAINABLE: False
63
+ IS_TRAINABLE: True
64
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_elephant_5002_256.pkl"
65
+ "cow_5002":
66
+ TYPE: vertex_feature
67
+ NUM_VERTICES: 5002
68
+ FEATURE_DIM: 256
69
+ FEATURES_TRAINABLE: False
70
+ IS_TRAINABLE: True
71
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cow_5002_256.pkl"
72
+ "bear_4936":
73
+ TYPE: vertex_feature
74
+ NUM_VERTICES: 4936
75
+ FEATURE_DIM: 256
76
+ FEATURES_TRAINABLE: False
77
+ IS_TRAINABLE: True
78
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_bear_4936_256.pkl"
79
+ DATASETS:
80
+ TRAIN:
81
+ - "densepose_lvis_v1_ds1_train_v1"
82
+ TEST:
83
+ - "densepose_lvis_v1_ds1_val_v1"
84
+ WHITELISTED_CATEGORIES:
85
+ "densepose_lvis_v1_ds1_train_v1":
86
+ - 943 # sheep
87
+ - 1202 # zebra
88
+ - 569 # horse
89
+ - 496 # giraffe
90
+ - 422 # elephant
91
+ - 80 # cow
92
+ - 76 # bear
93
+ - 225 # cat
94
+ - 378 # dog
95
+ "densepose_lvis_v1_ds1_val_v1":
96
+ - 943 # sheep
97
+ - 1202 # zebra
98
+ - 569 # horse
99
+ - 496 # giraffe
100
+ - 422 # elephant
101
+ - 80 # cow
102
+ - 76 # bear
103
+ - 225 # cat
104
+ - 378 # dog
105
+ CATEGORY_MAPS:
106
+ "densepose_lvis_v1_ds1_train_v1":
107
+ "1202": 943 # zebra -> sheep
108
+ "569": 943 # horse -> sheep
109
+ "496": 943 # giraffe -> sheep
110
+ "422": 943 # elephant -> sheep
111
+ "80": 943 # cow -> sheep
112
+ "76": 943 # bear -> sheep
113
+ "225": 943 # cat -> sheep
114
+ "378": 943 # dog -> sheep
115
+ "densepose_lvis_v1_ds1_val_v1":
116
+ "1202": 943 # zebra -> sheep
117
+ "569": 943 # horse -> sheep
118
+ "496": 943 # giraffe -> sheep
119
+ "422": 943 # elephant -> sheep
120
+ "80": 943 # cow -> sheep
121
+ "76": 943 # bear -> sheep
122
+ "225": 943 # cat -> sheep
123
+ "378": 943 # dog -> sheep
124
+ CLASS_TO_MESH_NAME_MAPPING:
125
+ # Note: different classes are mapped to a single class
126
+ # mesh is chosen based on GT data, so this is just some
127
+ # value which has no particular meaning
128
+ "0": "sheep_5004"
129
+ SOLVER:
130
+ MAX_ITER: 4000
131
+ STEPS: (3000, 3500)
132
+ DENSEPOSE_EVALUATION:
133
+ EVALUATE_MESH_ALIGNMENT: True
configs/cse/densepose_rcnn_R_50_FPN_soft_animals_I0_finetune_16k.yaml ADDED
@@ -0,0 +1,119 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://dl.fbaipublicfiles.com/densepose/cse/densepose_rcnn_R_50_FPN_soft_animals_finetune_maskonly_24k/270668502/model_final_21b1d2.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_HEADS:
7
+ NUM_CLASSES: 9
8
+ ROI_DENSEPOSE_HEAD:
9
+ NAME: "DensePoseV1ConvXHead"
10
+ COARSE_SEGM_TRAINED_BY_MASKS: True
11
+ CSE:
12
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
13
+ EMBEDDING_DIST_GAUSS_SIGMA: 0.1
14
+ GEODESIC_DIST_GAUSS_SIGMA: 0.1
15
+ EMBEDDERS:
16
+ "cat_7466":
17
+ TYPE: vertex_feature
18
+ NUM_VERTICES: 7466
19
+ FEATURE_DIM: 256
20
+ FEATURES_TRAINABLE: False
21
+ IS_TRAINABLE: True
22
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cat_7466_256.pkl"
23
+ "dog_7466":
24
+ TYPE: vertex_feature
25
+ NUM_VERTICES: 7466
26
+ FEATURE_DIM: 256
27
+ FEATURES_TRAINABLE: False
28
+ IS_TRAINABLE: True
29
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_dog_7466_256.pkl"
30
+ "sheep_5004":
31
+ TYPE: vertex_feature
32
+ NUM_VERTICES: 5004
33
+ FEATURE_DIM: 256
34
+ FEATURES_TRAINABLE: False
35
+ IS_TRAINABLE: True
36
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_sheep_5004_256.pkl"
37
+ "horse_5004":
38
+ TYPE: vertex_feature
39
+ NUM_VERTICES: 5004
40
+ FEATURE_DIM: 256
41
+ FEATURES_TRAINABLE: False
42
+ IS_TRAINABLE: True
43
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_horse_5004_256.pkl"
44
+ "zebra_5002":
45
+ TYPE: vertex_feature
46
+ NUM_VERTICES: 5002
47
+ FEATURE_DIM: 256
48
+ FEATURES_TRAINABLE: False
49
+ IS_TRAINABLE: True
50
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_zebra_5002_256.pkl"
51
+ "giraffe_5002":
52
+ TYPE: vertex_feature
53
+ NUM_VERTICES: 5002
54
+ FEATURE_DIM: 256
55
+ FEATURES_TRAINABLE: False
56
+ IS_TRAINABLE: True
57
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_giraffe_5002_256.pkl"
58
+ "elephant_5002":
59
+ TYPE: vertex_feature
60
+ NUM_VERTICES: 5002
61
+ FEATURE_DIM: 256
62
+ FEATURES_TRAINABLE: False
63
+ IS_TRAINABLE: True
64
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_elephant_5002_256.pkl"
65
+ "cow_5002":
66
+ TYPE: vertex_feature
67
+ NUM_VERTICES: 5002
68
+ FEATURE_DIM: 256
69
+ FEATURES_TRAINABLE: False
70
+ IS_TRAINABLE: True
71
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cow_5002_256.pkl"
72
+ "bear_4936":
73
+ TYPE: vertex_feature
74
+ NUM_VERTICES: 4936
75
+ FEATURE_DIM: 256
76
+ FEATURES_TRAINABLE: False
77
+ IS_TRAINABLE: True
78
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_bear_4936_256.pkl"
79
+ DATASETS:
80
+ TRAIN:
81
+ - "densepose_lvis_v1_ds2_train_v1"
82
+ TEST:
83
+ - "densepose_lvis_v1_ds2_val_v1"
84
+ WHITELISTED_CATEGORIES:
85
+ "densepose_lvis_v1_ds2_train_v1":
86
+ - 943 # sheep
87
+ - 1202 # zebra
88
+ - 569 # horse
89
+ - 496 # giraffe
90
+ - 422 # elephant
91
+ - 80 # cow
92
+ - 76 # bear
93
+ - 225 # cat
94
+ - 378 # dog
95
+ "densepose_lvis_v1_ds2_val_v1":
96
+ - 943 # sheep
97
+ - 1202 # zebra
98
+ - 569 # horse
99
+ - 496 # giraffe
100
+ - 422 # elephant
101
+ - 80 # cow
102
+ - 76 # bear
103
+ - 225 # cat
104
+ - 378 # dog
105
+ CLASS_TO_MESH_NAME_MAPPING:
106
+ "0": "bear_4936"
107
+ "1": "cow_5002"
108
+ "2": "cat_7466"
109
+ "3": "dog_7466"
110
+ "4": "elephant_5002"
111
+ "5": "giraffe_5002"
112
+ "6": "horse_5004"
113
+ "7": "sheep_5004"
114
+ "8": "zebra_5002"
115
+ SOLVER:
116
+ MAX_ITER: 16000
117
+ STEPS: (12000, 14000)
118
+ DENSEPOSE_EVALUATION:
119
+ EVALUATE_MESH_ALIGNMENT: True
configs/cse/densepose_rcnn_R_50_FPN_soft_animals_I0_finetune_i2m_16k.yaml ADDED
@@ -0,0 +1,121 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://dl.fbaipublicfiles.com/densepose/cse/densepose_rcnn_R_50_FPN_soft_animals_finetune_maskonly_24k/270668502/model_final_21b1d2.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_HEADS:
7
+ NUM_CLASSES: 9
8
+ ROI_DENSEPOSE_HEAD:
9
+ NAME: "DensePoseV1ConvXHead"
10
+ COARSE_SEGM_TRAINED_BY_MASKS: True
11
+ CSE:
12
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
13
+ EMBEDDING_DIST_GAUSS_SIGMA: 0.1
14
+ GEODESIC_DIST_GAUSS_SIGMA: 0.1
15
+ PIX_TO_SHAPE_CYCLE_LOSS:
16
+ ENABLED: True
17
+ EMBEDDERS:
18
+ "cat_7466":
19
+ TYPE: vertex_feature
20
+ NUM_VERTICES: 7466
21
+ FEATURE_DIM: 256
22
+ FEATURES_TRAINABLE: False
23
+ IS_TRAINABLE: True
24
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cat_7466_256.pkl"
25
+ "dog_7466":
26
+ TYPE: vertex_feature
27
+ NUM_VERTICES: 7466
28
+ FEATURE_DIM: 256
29
+ FEATURES_TRAINABLE: False
30
+ IS_TRAINABLE: True
31
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_dog_7466_256.pkl"
32
+ "sheep_5004":
33
+ TYPE: vertex_feature
34
+ NUM_VERTICES: 5004
35
+ FEATURE_DIM: 256
36
+ FEATURES_TRAINABLE: False
37
+ IS_TRAINABLE: True
38
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_sheep_5004_256.pkl"
39
+ "horse_5004":
40
+ TYPE: vertex_feature
41
+ NUM_VERTICES: 5004
42
+ FEATURE_DIM: 256
43
+ FEATURES_TRAINABLE: False
44
+ IS_TRAINABLE: True
45
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_horse_5004_256.pkl"
46
+ "zebra_5002":
47
+ TYPE: vertex_feature
48
+ NUM_VERTICES: 5002
49
+ FEATURE_DIM: 256
50
+ FEATURES_TRAINABLE: False
51
+ IS_TRAINABLE: True
52
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_zebra_5002_256.pkl"
53
+ "giraffe_5002":
54
+ TYPE: vertex_feature
55
+ NUM_VERTICES: 5002
56
+ FEATURE_DIM: 256
57
+ FEATURES_TRAINABLE: False
58
+ IS_TRAINABLE: True
59
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_giraffe_5002_256.pkl"
60
+ "elephant_5002":
61
+ TYPE: vertex_feature
62
+ NUM_VERTICES: 5002
63
+ FEATURE_DIM: 256
64
+ FEATURES_TRAINABLE: False
65
+ IS_TRAINABLE: True
66
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_elephant_5002_256.pkl"
67
+ "cow_5002":
68
+ TYPE: vertex_feature
69
+ NUM_VERTICES: 5002
70
+ FEATURE_DIM: 256
71
+ FEATURES_TRAINABLE: False
72
+ IS_TRAINABLE: True
73
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cow_5002_256.pkl"
74
+ "bear_4936":
75
+ TYPE: vertex_feature
76
+ NUM_VERTICES: 4936
77
+ FEATURE_DIM: 256
78
+ FEATURES_TRAINABLE: False
79
+ IS_TRAINABLE: True
80
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_bear_4936_256.pkl"
81
+ DATASETS:
82
+ TRAIN:
83
+ - "densepose_lvis_v1_ds2_train_v1"
84
+ TEST:
85
+ - "densepose_lvis_v1_ds2_val_v1"
86
+ WHITELISTED_CATEGORIES:
87
+ "densepose_lvis_v1_ds2_train_v1":
88
+ - 943 # sheep
89
+ - 1202 # zebra
90
+ - 569 # horse
91
+ - 496 # giraffe
92
+ - 422 # elephant
93
+ - 80 # cow
94
+ - 76 # bear
95
+ - 225 # cat
96
+ - 378 # dog
97
+ "densepose_lvis_v1_ds2_val_v1":
98
+ - 943 # sheep
99
+ - 1202 # zebra
100
+ - 569 # horse
101
+ - 496 # giraffe
102
+ - 422 # elephant
103
+ - 80 # cow
104
+ - 76 # bear
105
+ - 225 # cat
106
+ - 378 # dog
107
+ CLASS_TO_MESH_NAME_MAPPING:
108
+ "0": "bear_4936"
109
+ "1": "cow_5002"
110
+ "2": "cat_7466"
111
+ "3": "dog_7466"
112
+ "4": "elephant_5002"
113
+ "5": "giraffe_5002"
114
+ "6": "horse_5004"
115
+ "7": "sheep_5004"
116
+ "8": "zebra_5002"
117
+ SOLVER:
118
+ MAX_ITER: 16000
119
+ STEPS: (12000, 14000)
120
+ DENSEPOSE_EVALUATION:
121
+ EVALUATE_MESH_ALIGNMENT: True
configs/cse/densepose_rcnn_R_50_FPN_soft_animals_I0_finetune_m2m_16k.yaml ADDED
@@ -0,0 +1,138 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://dl.fbaipublicfiles.com/densepose/cse/densepose_rcnn_R_50_FPN_soft_animals_finetune_maskonly_24k/267687159/model_final_354e61.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_HEADS:
7
+ NUM_CLASSES: 9
8
+ ROI_DENSEPOSE_HEAD:
9
+ NAME: "DensePoseV1ConvXHead"
10
+ COARSE_SEGM_TRAINED_BY_MASKS: True
11
+ CSE:
12
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
13
+ EMBEDDING_DIST_GAUSS_SIGMA: 0.1
14
+ GEODESIC_DIST_GAUSS_SIGMA: 0.1
15
+ SHAPE_TO_SHAPE_CYCLE_LOSS:
16
+ ENABLED: True
17
+ EMBEDDERS:
18
+ "cat_7466":
19
+ TYPE: vertex_feature
20
+ NUM_VERTICES: 7466
21
+ FEATURE_DIM: 256
22
+ FEATURES_TRAINABLE: False
23
+ IS_TRAINABLE: True
24
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cat_7466_256.pkl"
25
+ "dog_7466":
26
+ TYPE: vertex_feature
27
+ NUM_VERTICES: 7466
28
+ FEATURE_DIM: 256
29
+ FEATURES_TRAINABLE: False
30
+ IS_TRAINABLE: True
31
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_dog_7466_256.pkl"
32
+ "sheep_5004":
33
+ TYPE: vertex_feature
34
+ NUM_VERTICES: 5004
35
+ FEATURE_DIM: 256
36
+ FEATURES_TRAINABLE: False
37
+ IS_TRAINABLE: True
38
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_sheep_5004_256.pkl"
39
+ "horse_5004":
40
+ TYPE: vertex_feature
41
+ NUM_VERTICES: 5004
42
+ FEATURE_DIM: 256
43
+ FEATURES_TRAINABLE: False
44
+ IS_TRAINABLE: True
45
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_horse_5004_256.pkl"
46
+ "zebra_5002":
47
+ TYPE: vertex_feature
48
+ NUM_VERTICES: 5002
49
+ FEATURE_DIM: 256
50
+ FEATURES_TRAINABLE: False
51
+ IS_TRAINABLE: True
52
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_zebra_5002_256.pkl"
53
+ "giraffe_5002":
54
+ TYPE: vertex_feature
55
+ NUM_VERTICES: 5002
56
+ FEATURE_DIM: 256
57
+ FEATURES_TRAINABLE: False
58
+ IS_TRAINABLE: True
59
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_giraffe_5002_256.pkl"
60
+ "elephant_5002":
61
+ TYPE: vertex_feature
62
+ NUM_VERTICES: 5002
63
+ FEATURE_DIM: 256
64
+ FEATURES_TRAINABLE: False
65
+ IS_TRAINABLE: True
66
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_elephant_5002_256.pkl"
67
+ "cow_5002":
68
+ TYPE: vertex_feature
69
+ NUM_VERTICES: 5002
70
+ FEATURE_DIM: 256
71
+ FEATURES_TRAINABLE: False
72
+ IS_TRAINABLE: True
73
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cow_5002_256.pkl"
74
+ "bear_4936":
75
+ TYPE: vertex_feature
76
+ NUM_VERTICES: 4936
77
+ FEATURE_DIM: 256
78
+ FEATURES_TRAINABLE: False
79
+ IS_TRAINABLE: True
80
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_bear_4936_256.pkl"
81
+ "smpl_27554":
82
+ TYPE: vertex_feature
83
+ NUM_VERTICES: 27554
84
+ FEATURE_DIM: 256
85
+ FEATURES_TRAINABLE: False
86
+ IS_TRAINABLE: True
87
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_smpl_27554_256.pkl"
88
+ DATASETS:
89
+ TRAIN:
90
+ - "densepose_lvis_v1_ds2_train_v1"
91
+ TEST:
92
+ - "densepose_lvis_v1_ds2_val_v1"
93
+ WHITELISTED_CATEGORIES:
94
+ "densepose_lvis_v1_ds2_train_v1":
95
+ - 943 # sheep
96
+ - 1202 # zebra
97
+ - 569 # horse
98
+ - 496 # giraffe
99
+ - 422 # elephant
100
+ - 80 # cow
101
+ - 76 # bear
102
+ - 225 # cat
103
+ - 378 # dog
104
+ "densepose_lvis_v1_ds2_val_v1":
105
+ - 943 # sheep
106
+ - 1202 # zebra
107
+ - 569 # horse
108
+ - 496 # giraffe
109
+ - 422 # elephant
110
+ - 80 # cow
111
+ - 76 # bear
112
+ - 225 # cat
113
+ - 378 # dog
114
+ CLASS_TO_MESH_NAME_MAPPING:
115
+ "0": "bear_4936"
116
+ "1": "cow_5002"
117
+ "2": "cat_7466"
118
+ "3": "dog_7466"
119
+ "4": "elephant_5002"
120
+ "5": "giraffe_5002"
121
+ "6": "horse_5004"
122
+ "7": "sheep_5004"
123
+ "8": "zebra_5002"
124
+ SOLVER:
125
+ MAX_ITER: 16000
126
+ STEPS: (12000, 14000)
127
+ DENSEPOSE_EVALUATION:
128
+ EVALUATE_MESH_ALIGNMENT: True
129
+ MESH_ALIGNMENT_MESH_NAMES:
130
+ - bear_4936
131
+ - cow_5002
132
+ - cat_7466
133
+ - dog_7466
134
+ - elephant_5002
135
+ - giraffe_5002
136
+ - horse_5004
137
+ - sheep_5004
138
+ - zebra_5002
configs/cse/densepose_rcnn_R_50_FPN_soft_animals_finetune_16k.yaml ADDED
@@ -0,0 +1,119 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://dl.fbaipublicfiles.com/densepose/cse/densepose_rcnn_R_50_FPN_soft_s1x/250533982/model_final_2c4512.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_HEADS:
7
+ NUM_CLASSES: 9
8
+ ROI_DENSEPOSE_HEAD:
9
+ NAME: "DensePoseV1ConvXHead"
10
+ COARSE_SEGM_TRAINED_BY_MASKS: True
11
+ CSE:
12
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
13
+ EMBEDDING_DIST_GAUSS_SIGMA: 0.1
14
+ GEODESIC_DIST_GAUSS_SIGMA: 0.1
15
+ EMBEDDERS:
16
+ "cat_7466":
17
+ TYPE: vertex_feature
18
+ NUM_VERTICES: 7466
19
+ FEATURE_DIM: 256
20
+ FEATURES_TRAINABLE: False
21
+ IS_TRAINABLE: True
22
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cat_7466_256.pkl"
23
+ "dog_7466":
24
+ TYPE: vertex_feature
25
+ NUM_VERTICES: 7466
26
+ FEATURE_DIM: 256
27
+ FEATURES_TRAINABLE: False
28
+ IS_TRAINABLE: True
29
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_dog_7466_256.pkl"
30
+ "sheep_5004":
31
+ TYPE: vertex_feature
32
+ NUM_VERTICES: 5004
33
+ FEATURE_DIM: 256
34
+ FEATURES_TRAINABLE: False
35
+ IS_TRAINABLE: True
36
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_sheep_5004_256.pkl"
37
+ "horse_5004":
38
+ TYPE: vertex_feature
39
+ NUM_VERTICES: 5004
40
+ FEATURE_DIM: 256
41
+ FEATURES_TRAINABLE: False
42
+ IS_TRAINABLE: True
43
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_horse_5004_256.pkl"
44
+ "zebra_5002":
45
+ TYPE: vertex_feature
46
+ NUM_VERTICES: 5002
47
+ FEATURE_DIM: 256
48
+ FEATURES_TRAINABLE: False
49
+ IS_TRAINABLE: True
50
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_zebra_5002_256.pkl"
51
+ "giraffe_5002":
52
+ TYPE: vertex_feature
53
+ NUM_VERTICES: 5002
54
+ FEATURE_DIM: 256
55
+ FEATURES_TRAINABLE: False
56
+ IS_TRAINABLE: True
57
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_giraffe_5002_256.pkl"
58
+ "elephant_5002":
59
+ TYPE: vertex_feature
60
+ NUM_VERTICES: 5002
61
+ FEATURE_DIM: 256
62
+ FEATURES_TRAINABLE: False
63
+ IS_TRAINABLE: True
64
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_elephant_5002_256.pkl"
65
+ "cow_5002":
66
+ TYPE: vertex_feature
67
+ NUM_VERTICES: 5002
68
+ FEATURE_DIM: 256
69
+ FEATURES_TRAINABLE: False
70
+ IS_TRAINABLE: True
71
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cow_5002_256.pkl"
72
+ "bear_4936":
73
+ TYPE: vertex_feature
74
+ NUM_VERTICES: 4936
75
+ FEATURE_DIM: 256
76
+ FEATURES_TRAINABLE: False
77
+ IS_TRAINABLE: True
78
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_bear_4936_256.pkl"
79
+ DATASETS:
80
+ TRAIN:
81
+ - "densepose_lvis_v1_ds2_train_v1"
82
+ TEST:
83
+ - "densepose_lvis_v1_ds2_val_v1"
84
+ WHITELISTED_CATEGORIES:
85
+ "densepose_lvis_v1_ds2_train_v1":
86
+ - 943 # sheep
87
+ - 1202 # zebra
88
+ - 569 # horse
89
+ - 496 # giraffe
90
+ - 422 # elephant
91
+ - 80 # cow
92
+ - 76 # bear
93
+ - 225 # cat
94
+ - 378 # dog
95
+ "densepose_lvis_v1_ds2_val_v1":
96
+ - 943 # sheep
97
+ - 1202 # zebra
98
+ - 569 # horse
99
+ - 496 # giraffe
100
+ - 422 # elephant
101
+ - 80 # cow
102
+ - 76 # bear
103
+ - 225 # cat
104
+ - 378 # dog
105
+ CLASS_TO_MESH_NAME_MAPPING:
106
+ "0": "bear_4936"
107
+ "1": "cow_5002"
108
+ "2": "cat_7466"
109
+ "3": "dog_7466"
110
+ "4": "elephant_5002"
111
+ "5": "giraffe_5002"
112
+ "6": "horse_5004"
113
+ "7": "sheep_5004"
114
+ "8": "zebra_5002"
115
+ SOLVER:
116
+ MAX_ITER: 16000
117
+ STEPS: (12000, 14000)
118
+ DENSEPOSE_EVALUATION:
119
+ EVALUATE_MESH_ALIGNMENT: True
configs/cse/densepose_rcnn_R_50_FPN_soft_animals_finetune_4k.yaml ADDED
@@ -0,0 +1,119 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://dl.fbaipublicfiles.com/densepose/cse/densepose_rcnn_R_50_FPN_soft_s1x/250533982/model_final_2c4512.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_HEADS:
7
+ NUM_CLASSES: 9
8
+ ROI_DENSEPOSE_HEAD:
9
+ NAME: "DensePoseV1ConvXHead"
10
+ COARSE_SEGM_TRAINED_BY_MASKS: True
11
+ CSE:
12
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
13
+ EMBEDDING_DIST_GAUSS_SIGMA: 0.1
14
+ GEODESIC_DIST_GAUSS_SIGMA: 0.1
15
+ EMBEDDERS:
16
+ "cat_5001":
17
+ TYPE: vertex_feature
18
+ NUM_VERTICES: 5001
19
+ FEATURE_DIM: 256
20
+ FEATURES_TRAINABLE: False
21
+ IS_TRAINABLE: True
22
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cat_5001_256.pkl"
23
+ "dog_5002":
24
+ TYPE: vertex_feature
25
+ NUM_VERTICES: 5002
26
+ FEATURE_DIM: 256
27
+ FEATURES_TRAINABLE: False
28
+ IS_TRAINABLE: True
29
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_dog_5002_256.pkl"
30
+ "sheep_5004":
31
+ TYPE: vertex_feature
32
+ NUM_VERTICES: 5004
33
+ FEATURE_DIM: 256
34
+ FEATURES_TRAINABLE: False
35
+ IS_TRAINABLE: True
36
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_sheep_5004_256.pkl"
37
+ "horse_5004":
38
+ TYPE: vertex_feature
39
+ NUM_VERTICES: 5004
40
+ FEATURE_DIM: 256
41
+ FEATURES_TRAINABLE: False
42
+ IS_TRAINABLE: True
43
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_horse_5004_256.pkl"
44
+ "zebra_5002":
45
+ TYPE: vertex_feature
46
+ NUM_VERTICES: 5002
47
+ FEATURE_DIM: 256
48
+ FEATURES_TRAINABLE: False
49
+ IS_TRAINABLE: True
50
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_zebra_5002_256.pkl"
51
+ "giraffe_5002":
52
+ TYPE: vertex_feature
53
+ NUM_VERTICES: 5002
54
+ FEATURE_DIM: 256
55
+ FEATURES_TRAINABLE: False
56
+ IS_TRAINABLE: True
57
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_giraffe_5002_256.pkl"
58
+ "elephant_5002":
59
+ TYPE: vertex_feature
60
+ NUM_VERTICES: 5002
61
+ FEATURE_DIM: 256
62
+ FEATURES_TRAINABLE: False
63
+ IS_TRAINABLE: True
64
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_elephant_5002_256.pkl"
65
+ "cow_5002":
66
+ TYPE: vertex_feature
67
+ NUM_VERTICES: 5002
68
+ FEATURE_DIM: 256
69
+ FEATURES_TRAINABLE: False
70
+ IS_TRAINABLE: True
71
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cow_5002_256.pkl"
72
+ "bear_4936":
73
+ TYPE: vertex_feature
74
+ NUM_VERTICES: 4936
75
+ FEATURE_DIM: 256
76
+ FEATURES_TRAINABLE: False
77
+ IS_TRAINABLE: True
78
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_bear_4936_256.pkl"
79
+ DATASETS:
80
+ TRAIN:
81
+ - "densepose_lvis_v1_ds1_train_v1"
82
+ TEST:
83
+ - "densepose_lvis_v1_ds1_val_v1"
84
+ WHITELISTED_CATEGORIES:
85
+ "densepose_lvis_v1_ds1_train_v1":
86
+ - 943 # sheep
87
+ - 1202 # zebra
88
+ - 569 # horse
89
+ - 496 # giraffe
90
+ - 422 # elephant
91
+ - 80 # cow
92
+ - 76 # bear
93
+ - 225 # cat
94
+ - 378 # dog
95
+ "densepose_lvis_v1_ds1_val_v1":
96
+ - 943 # sheep
97
+ - 1202 # zebra
98
+ - 569 # horse
99
+ - 496 # giraffe
100
+ - 422 # elephant
101
+ - 80 # cow
102
+ - 76 # bear
103
+ - 225 # cat
104
+ - 378 # dog
105
+ CLASS_TO_MESH_NAME_MAPPING:
106
+ "0": "bear_4936"
107
+ "1": "cow_5002"
108
+ "2": "cat_5001"
109
+ "3": "dog_5002"
110
+ "4": "elephant_5002"
111
+ "5": "giraffe_5002"
112
+ "6": "horse_5004"
113
+ "7": "sheep_5004"
114
+ "8": "zebra_5002"
115
+ SOLVER:
116
+ MAX_ITER: 4000
117
+ STEPS: (3000, 3500)
118
+ DENSEPOSE_EVALUATION:
119
+ EVALUATE_MESH_ALIGNMENT: True
configs/cse/densepose_rcnn_R_50_FPN_soft_animals_finetune_maskonly_24k.yaml ADDED
@@ -0,0 +1,118 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://dl.fbaipublicfiles.com/densepose/cse/densepose_rcnn_R_50_FPN_soft_s1x/250533982/model_final_2c4512.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_HEADS:
7
+ NUM_CLASSES: 9
8
+ ROI_DENSEPOSE_HEAD:
9
+ NAME: "DensePoseV1ConvXHead"
10
+ COARSE_SEGM_TRAINED_BY_MASKS: True
11
+ CSE:
12
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
13
+ EMBED_LOSS_WEIGHT: 0.0
14
+ EMBEDDING_DIST_GAUSS_SIGMA: 0.1
15
+ GEODESIC_DIST_GAUSS_SIGMA: 0.1
16
+ EMBEDDERS:
17
+ "cat_7466":
18
+ TYPE: vertex_feature
19
+ NUM_VERTICES: 7466
20
+ FEATURE_DIM: 256
21
+ FEATURES_TRAINABLE: False
22
+ IS_TRAINABLE: True
23
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cat_7466_256.pkl"
24
+ "dog_7466":
25
+ TYPE: vertex_feature
26
+ NUM_VERTICES: 7466
27
+ FEATURE_DIM: 256
28
+ FEATURES_TRAINABLE: False
29
+ IS_TRAINABLE: True
30
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_dog_7466_256.pkl"
31
+ "sheep_5004":
32
+ TYPE: vertex_feature
33
+ NUM_VERTICES: 5004
34
+ FEATURE_DIM: 256
35
+ FEATURES_TRAINABLE: False
36
+ IS_TRAINABLE: True
37
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_sheep_5004_256.pkl"
38
+ "horse_5004":
39
+ TYPE: vertex_feature
40
+ NUM_VERTICES: 5004
41
+ FEATURE_DIM: 256
42
+ FEATURES_TRAINABLE: False
43
+ IS_TRAINABLE: True
44
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_horse_5004_256.pkl"
45
+ "zebra_5002":
46
+ TYPE: vertex_feature
47
+ NUM_VERTICES: 5002
48
+ FEATURE_DIM: 256
49
+ FEATURES_TRAINABLE: False
50
+ IS_TRAINABLE: True
51
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_zebra_5002_256.pkl"
52
+ "giraffe_5002":
53
+ TYPE: vertex_feature
54
+ NUM_VERTICES: 5002
55
+ FEATURE_DIM: 256
56
+ FEATURES_TRAINABLE: False
57
+ IS_TRAINABLE: True
58
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_giraffe_5002_256.pkl"
59
+ "elephant_5002":
60
+ TYPE: vertex_feature
61
+ NUM_VERTICES: 5002
62
+ FEATURE_DIM: 256
63
+ FEATURES_TRAINABLE: False
64
+ IS_TRAINABLE: True
65
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_elephant_5002_256.pkl"
66
+ "cow_5002":
67
+ TYPE: vertex_feature
68
+ NUM_VERTICES: 5002
69
+ FEATURE_DIM: 256
70
+ FEATURES_TRAINABLE: False
71
+ IS_TRAINABLE: True
72
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_cow_5002_256.pkl"
73
+ "bear_4936":
74
+ TYPE: vertex_feature
75
+ NUM_VERTICES: 4936
76
+ FEATURE_DIM: 256
77
+ FEATURES_TRAINABLE: False
78
+ IS_TRAINABLE: True
79
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_bear_4936_256.pkl"
80
+ DATASETS:
81
+ TRAIN:
82
+ - "densepose_lvis_v1_ds2_train_v1"
83
+ TEST:
84
+ - "densepose_lvis_v1_ds2_val_v1"
85
+ WHITELISTED_CATEGORIES:
86
+ "densepose_lvis_v1_ds2_train_v1":
87
+ - 943 # sheep
88
+ - 1202 # zebra
89
+ - 569 # horse
90
+ - 496 # giraffe
91
+ - 422 # elephant
92
+ - 80 # cow
93
+ - 76 # bear
94
+ - 225 # cat
95
+ - 378 # dog
96
+ "densepose_lvis_v1_ds2_val_v1":
97
+ - 943 # sheep
98
+ - 1202 # zebra
99
+ - 569 # horse
100
+ - 496 # giraffe
101
+ - 422 # elephant
102
+ - 80 # cow
103
+ - 76 # bear
104
+ - 225 # cat
105
+ - 378 # dog
106
+ CLASS_TO_MESH_NAME_MAPPING:
107
+ "0": "bear_4936"
108
+ "1": "cow_5002"
109
+ "2": "cat_7466"
110
+ "3": "dog_7466"
111
+ "4": "elephant_5002"
112
+ "5": "giraffe_5002"
113
+ "6": "horse_5004"
114
+ "7": "sheep_5004"
115
+ "8": "zebra_5002"
116
+ SOLVER:
117
+ MAX_ITER: 24000
118
+ STEPS: (20000, 22000)
configs/cse/densepose_rcnn_R_50_FPN_soft_chimps_finetune_4k.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "https://dl.fbaipublicfiles.com/densepose/cse/densepose_rcnn_R_50_FPN_soft_s1x/250533982/model_final_2c4512.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseV1ConvXHead"
8
+ CSE:
9
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
10
+ EMBEDDING_DIST_GAUSS_SIGMA: 0.1
11
+ GEODESIC_DIST_GAUSS_SIGMA: 0.1
12
+ EMBEDDERS:
13
+ "chimp_5029":
14
+ TYPE: vertex_feature
15
+ NUM_VERTICES: 5029
16
+ FEATURE_DIM: 256
17
+ FEATURES_TRAINABLE: False
18
+ IS_TRAINABLE: True
19
+ INIT_FILE: "https://dl.fbaipublicfiles.com/densepose/data/cse/lbo/phi_chimp_5029_256.pkl"
20
+ DATASETS:
21
+ TRAIN:
22
+ - "densepose_chimps_cse_train"
23
+ TEST:
24
+ - "densepose_chimps_cse_val"
25
+ CLASS_TO_MESH_NAME_MAPPING:
26
+ "0": "chimp_5029"
27
+ SOLVER:
28
+ MAX_ITER: 4000
29
+ STEPS: (3000, 3500)
configs/cse/densepose_rcnn_R_50_FPN_soft_s1x.yaml ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN-Human.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-50.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseV1ConvXHead"
8
+ CSE:
9
+ EMBED_LOSS_NAME: "SoftEmbeddingLoss"
10
+ SOLVER:
11
+ MAX_ITER: 130000
12
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_101_FPN_DL_WC1M_s1x.yaml ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ UV_CONFIDENCE:
9
+ ENABLED: True
10
+ TYPE: "iid_iso"
11
+ SEGM_CONFIDENCE:
12
+ ENABLED: True
13
+ POINT_REGRESSION_WEIGHTS: 0.0005
14
+ SOLVER:
15
+ CLIP_GRADIENTS:
16
+ ENABLED: True
17
+ MAX_ITER: 130000
18
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_101_FPN_DL_WC1_s1x.yaml ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ UV_CONFIDENCE:
9
+ ENABLED: True
10
+ TYPE: "iid_iso"
11
+ POINT_REGRESSION_WEIGHTS: 0.0005
12
+ SOLVER:
13
+ CLIP_GRADIENTS:
14
+ ENABLED: True
15
+ MAX_ITER: 130000
16
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_101_FPN_DL_WC2M_s1x.yaml ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ UV_CONFIDENCE:
9
+ ENABLED: True
10
+ TYPE: "indep_aniso"
11
+ SEGM_CONFIDENCE:
12
+ ENABLED: True
13
+ POINT_REGRESSION_WEIGHTS: 0.0005
14
+ SOLVER:
15
+ CLIP_GRADIENTS:
16
+ ENABLED: True
17
+ MAX_ITER: 130000
18
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_101_FPN_DL_WC2_s1x.yaml ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ UV_CONFIDENCE:
9
+ ENABLED: True
10
+ TYPE: "indep_aniso"
11
+ POINT_REGRESSION_WEIGHTS: 0.0005
12
+ SOLVER:
13
+ CLIP_GRADIENTS:
14
+ ENABLED: True
15
+ MAX_ITER: 130000
16
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_101_FPN_DL_s1x.yaml ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ SOLVER:
9
+ MAX_ITER: 130000
10
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_101_FPN_WC1M_s1x.yaml ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ UV_CONFIDENCE:
8
+ ENABLED: True
9
+ TYPE: "iid_iso"
10
+ SEGM_CONFIDENCE:
11
+ ENABLED: True
12
+ POINT_REGRESSION_WEIGHTS: 0.0005
13
+ SOLVER:
14
+ CLIP_GRADIENTS:
15
+ ENABLED: True
16
+ MAX_ITER: 130000
17
+ STEPS: (100000, 120000)
18
+ WARMUP_FACTOR: 0.025
configs/densepose_rcnn_R_101_FPN_WC1_s1x.yaml ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ UV_CONFIDENCE:
8
+ ENABLED: True
9
+ TYPE: "iid_iso"
10
+ POINT_REGRESSION_WEIGHTS: 0.0005
11
+ SOLVER:
12
+ CLIP_GRADIENTS:
13
+ ENABLED: True
14
+ MAX_ITER: 130000
15
+ STEPS: (100000, 120000)
16
+ WARMUP_FACTOR: 0.025
configs/densepose_rcnn_R_101_FPN_WC2M_s1x.yaml ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ UV_CONFIDENCE:
8
+ ENABLED: True
9
+ TYPE: "indep_aniso"
10
+ SEGM_CONFIDENCE:
11
+ ENABLED: True
12
+ POINT_REGRESSION_WEIGHTS: 0.0005
13
+ SOLVER:
14
+ CLIP_GRADIENTS:
15
+ ENABLED: True
16
+ MAX_ITER: 130000
17
+ STEPS: (100000, 120000)
18
+ WARMUP_FACTOR: 0.025
configs/densepose_rcnn_R_101_FPN_WC2_s1x.yaml ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ UV_CONFIDENCE:
8
+ ENABLED: True
9
+ TYPE: "indep_aniso"
10
+ POINT_REGRESSION_WEIGHTS: 0.0005
11
+ SOLVER:
12
+ CLIP_GRADIENTS:
13
+ ENABLED: True
14
+ MAX_ITER: 130000
15
+ STEPS: (100000, 120000)
16
+ WARMUP_FACTOR: 0.025
configs/densepose_rcnn_R_101_FPN_s1x.yaml ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ SOLVER:
7
+ MAX_ITER: 130000
8
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_101_FPN_s1x_legacy.yaml ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
4
+ RESNETS:
5
+ DEPTH: 101
6
+ ROI_DENSEPOSE_HEAD:
7
+ NUM_COARSE_SEGM_CHANNELS: 15
8
+ POOLER_RESOLUTION: 14
9
+ HEATMAP_SIZE: 56
10
+ INDEX_WEIGHTS: 2.0
11
+ PART_WEIGHTS: 0.3
12
+ POINT_REGRESSION_WEIGHTS: 0.1
13
+ DECODER_ON: False
14
+ SOLVER:
15
+ BASE_LR: 0.002
16
+ MAX_ITER: 130000
17
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_50_FPN_DL_WC1M_s1x.yaml ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-50.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ UV_CONFIDENCE:
9
+ ENABLED: True
10
+ TYPE: "iid_iso"
11
+ SEGM_CONFIDENCE:
12
+ ENABLED: True
13
+ POINT_REGRESSION_WEIGHTS: 0.0005
14
+ SOLVER:
15
+ CLIP_GRADIENTS:
16
+ ENABLED: True
17
+ MAX_ITER: 130000
18
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_50_FPN_DL_WC1_s1x.yaml ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-50.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ UV_CONFIDENCE:
9
+ ENABLED: True
10
+ TYPE: "iid_iso"
11
+ POINT_REGRESSION_WEIGHTS: 0.0005
12
+ SOLVER:
13
+ CLIP_GRADIENTS:
14
+ ENABLED: True
15
+ MAX_ITER: 130000
16
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_50_FPN_DL_WC2M_s1x.yaml ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-50.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ UV_CONFIDENCE:
9
+ ENABLED: True
10
+ TYPE: "indep_aniso"
11
+ SEGM_CONFIDENCE:
12
+ ENABLED: True
13
+ POINT_REGRESSION_WEIGHTS: 0.0005
14
+ SOLVER:
15
+ CLIP_GRADIENTS:
16
+ ENABLED: True
17
+ MAX_ITER: 130000
18
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_50_FPN_DL_WC2_s1x.yaml ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-50.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ UV_CONFIDENCE:
9
+ ENABLED: True
10
+ TYPE: "indep_aniso"
11
+ POINT_REGRESSION_WEIGHTS: 0.0005
12
+ SOLVER:
13
+ CLIP_GRADIENTS:
14
+ ENABLED: True
15
+ MAX_ITER: 130000
16
+ STEPS: (100000, 120000)
configs/densepose_rcnn_R_50_FPN_DL_s1x.yaml ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ _BASE_: "Base-DensePose-RCNN-FPN.yaml"
2
+ MODEL:
3
+ WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-50.pkl"
4
+ RESNETS:
5
+ DEPTH: 50
6
+ ROI_DENSEPOSE_HEAD:
7
+ NAME: "DensePoseDeepLabHead"
8
+ SOLVER:
9
+ MAX_ITER: 130000
10
+ STEPS: (100000, 120000)