liyang/hand-motion-pipeline
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Initial snapshot: hand motion pipeline (Dyn-HaMR + dex-retargeting + SPIDER)

Browse Source 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
liyang
2026-09-15 14:13:49 +08:00
commit 7e4ef6f98b
7002 changed files with 1452249 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/README.md
Vendored
+19
View File
@@ -0,0 +1,19 @@
# Image-based Human Body 2D Pose Estimation
Multi-person human pose estimation is defined as the task of detecting the poses (or keypoints) of all people from an input image.
Existing approaches can be categorized into top-down and bottom-up approaches.
Top-down methods (e.g. deeppose) divide the task into two stages: human detection and pose estimation. They perform human detection first, followed by single-person pose estimation given human bounding boxes.
Bottom-up approaches (e.g. AE) first detect all the keypoints and then group/associate them into person instances.
## Data preparation
Please follow [DATA Preparation](/docs/en/tasks/2d_body_keypoint.md) to prepare data.
## Demo
Please follow [Demo](/demo/docs/2d_human_pose_demo.md#2d-human-pose-demo) to run demos.
<img src="demo/resources/demo_coco.gif" width="600px" alt>
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/README.md
Vendored
+25
View File
@@ -0,0 +1,25 @@
# Associative embedding: End-to-end learning for joint detection and grouping (AE)
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
AE is one of the most popular 2D bottom-up pose estimation approaches, that first detect all the keypoints and
then group/associate them into person instances.
In order to group all the predicted keypoints to individuals, a tag is also predicted for each detected keypoint.
Tags of the same person are similar, while tags of different people are different. Thus the keypoints can be grouped
according to the tags.
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/higherhrnet_aic.md
Vendored
+61
View File
@@ -0,0 +1,61 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2020/html/Cheng_HigherHRNet_Scale-Aware_Representation_Learning_for_Bottom-Up_Human_Pose_Estimation_CVPR_2020_paper.html">HigherHRNet (CVPR'2020)</a></summary>
```bibtex
@inproceedings{cheng2020higherhrnet,
title={HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation},
author={Cheng, Bowen and Xiao, Bin and Wang, Jingdong and Shi, Honghui and Huang, Thomas S and Zhang, Lei},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={5386--5395},
year={2020}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1711.06475">AI Challenger (ArXiv'2017)</a></summary>
```bibtex
@article{wu2017ai,
title={Ai challenger: A large-scale dataset for going deeper in image understanding},
author={Wu, Jiahong and Zheng, He and Zhao, Bo and Li, Yixin and Yan, Baoming and Liang, Rui and Wang, Wenjia and Zhou, Shipei and Lin, Guosen and Fu, Yanwei and others},
journal={arXiv preprint arXiv:1711.06475},
year={2017}
}
```
</details>
Results on AIC validation set without multi-scale test
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HigherHRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/higherhrnet_w32_aic_512x512.py) | 512x512 | 0.315 | 0.710 | 0.243 | 0.379 | 0.757 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_aic_512x512-9a674c33_20210130.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_aic_512x512_20210130.log.json) |
Results on AIC validation set with multi-scale test. 3 default scales (\[2, 1, 0.5\]) are used
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HigherHRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/higherhrnet_w32_aic_512x512.py) | 512x512 | 0.323 | 0.718 | 0.254 | 0.379 | 0.758 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_aic_512x512-9a674c33_20210130.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_aic_512x512_20210130.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/higherhrnet_aic.yml
Vendored
+42
View File
@@ -0,0 +1,42 @@
Collections:
- Name: HigherHRNet
Paper:
Title: 'HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose
Estimation'
URL: http://openaccess.thecvf.com/content_CVPR_2020/html/Cheng_HigherHRNet_Scale-Aware_Representation_Learning_for_Bottom-Up_Human_Pose_Estimation_CVPR_2020_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/backbones/higherhrnet.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/higherhrnet_w32_aic_512x512.py
In Collection: HigherHRNet
Metadata:
Architecture: &id001
- Associative Embedding
- HigherHRNet
Training Data: AI Challenger
Name: associative_embedding_higherhrnet_w32_aic_512x512
Results:
- Dataset: AI Challenger
Metrics:
AP: 0.315
AP@0.5: 0.71
AP@0.75: 0.243
AR: 0.379
AR@0.5: 0.757
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_aic_512x512-9a674c33_20210130.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/higherhrnet_w32_aic_512x512.py
In Collection: HigherHRNet
Metadata:
Architecture: *id001
Training Data: AI Challenger
Name: associative_embedding_higherhrnet_w32_aic_512x512
Results:
- Dataset: AI Challenger
Metrics:
AP: 0.323
AP@0.5: 0.718
AP@0.75: 0.254
AR: 0.379
AR@0.5: 0.758
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_aic_512x512-9a674c33_20210130.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/higherhrnet_w32_aic_512x512.py
Vendored
+195
View File
@@ -0,0 +1,195 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128, 256],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=32,
num_joints=14,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[32],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.01, 0.01],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/higherhrnet_w32_aic_512x512_udp.py
Vendored
+198
View File
@@ -0,0 +1,198 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128, 256],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=32,
num_joints=14,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[32],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.01, 0.01],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True,
use_udp=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40,
use_udp=True),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
use_udp=True,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1], use_udp=True),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
],
use_udp=True),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/hrnet_aic.md
Vendored
+61
View File
@@ -0,0 +1,61 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2019/html/Sun_Deep_High-Resolution_Representation_Learning_for_Human_Pose_Estimation_CVPR_2019_paper.html">HRNet (CVPR'2019)</a></summary>
```bibtex
@inproceedings{sun2019deep,
title={Deep high-resolution representation learning for human pose estimation},
author={Sun, Ke and Xiao, Bin and Liu, Dong and Wang, Jingdong},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={5693--5703},
year={2019}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1711.06475">AI Challenger (ArXiv'2017)</a></summary>
```bibtex
@article{wu2017ai,
title={Ai challenger: A large-scale dataset for going deeper in image understanding},
author={Wu, Jiahong and Zheng, He and Zhao, Bo and Li, Yixin and Yan, Baoming and Liang, Rui and Wang, Wenjia and Zhou, Shipei and Lin, Guosen and Fu, Yanwei and others},
journal={arXiv preprint arXiv:1711.06475},
year={2017}
}
```
</details>
Results on AIC validation set without multi-scale test
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/hrnet_w32_aic_512x512.py) | 512x512 | 0.303 | 0.697 | 0.225 | 0.373 | 0.755 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_aic_512x512-77e2a98a_20210131.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_aic_512x512_20210131.log.json) |
Results on AIC validation set with multi-scale test. 3 default scales (\[2, 1, 0.5\]) are used
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/hrnet_w32_aic_512x512.py) | 512x512 | 0.318 | 0.717 | 0.246 | 0.379 | 0.764 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_aic_512x512-77e2a98a_20210131.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_aic_512x512_20210131.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/hrnet_aic.yml
Vendored
+41
View File
@@ -0,0 +1,41 @@
Collections:
- Name: HRNet
Paper:
Title: Deep high-resolution representation learning for human pose estimation
URL: http://openaccess.thecvf.com/content_CVPR_2019/html/Sun_Deep_High-Resolution_Representation_Learning_for_Human_Pose_Estimation_CVPR_2019_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/backbones/hrnet.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/hrnet_w32_aic_512x512.py
In Collection: HRNet
Metadata:
Architecture: &id001
- Associative Embedding
- HRNet
Training Data: AI Challenger
Name: associative_embedding_hrnet_w32_aic_512x512
Results:
- Dataset: AI Challenger
Metrics:
AP: 0.303
AP@0.5: 0.697
AP@0.75: 0.225
AR: 0.373
AR@0.5: 0.755
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_aic_512x512-77e2a98a_20210131.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/hrnet_w32_aic_512x512.py
In Collection: HRNet
Metadata:
Architecture: *id001
Training Data: AI Challenger
Name: associative_embedding_hrnet_w32_aic_512x512
Results:
- Dataset: AI Challenger
Metrics:
AP: 0.318
AP@0.5: 0.717
AP@0.75: 0.246
AR: 0.379
AR@0.5: 0.764
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_aic_512x512-77e2a98a_20210131.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/aic/hrnet_w32_aic_512x512.py
Vendored
+191
View File
@@ -0,0 +1,191 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AESimpleHead',
in_channels=32,
num_joints=14,
num_deconv_layers=0,
tag_per_joint=True,
with_ae_loss=[True],
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.01],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_coco.md
Vendored
+67
View File
@@ -0,0 +1,67 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2020/html/Cheng_HigherHRNet_Scale-Aware_Representation_Learning_for_Bottom-Up_Human_Pose_Estimation_CVPR_2020_paper.html">HigherHRNet (CVPR'2020)</a></summary>
```bibtex
@inproceedings{cheng2020higherhrnet,
title={HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation},
author={Cheng, Bowen and Xiao, Bin and Wang, Jingdong and Shi, Honghui and Huang, Thomas S and Zhang, Lei},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={5386--5395},
year={2020}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://link.springer.com/chapter/10.1007/978-3-319-10602-1_48">COCO (ECCV'2014)</a></summary>
```bibtex
@inproceedings{lin2014microsoft,
title={Microsoft coco: Common objects in context},
author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence},
booktitle={European conference on computer vision},
pages={740--755},
year={2014},
organization={Springer}
}
```
</details>
Results on COCO val2017 without multi-scale test
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HigherHRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_512x512.py) | 512x512 | 0.677 | 0.870 | 0.738 | 0.723 | 0.890 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_512x512-8ae85183_20200713.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_512x512_20200713.log.json) |
| [HigherHRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_640x640.py) | 640x640 | 0.686 | 0.871 | 0.747 | 0.733 | 0.898 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_640x640-a22fe938_20200712.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_640x640_20200712.log.json) |
| [HigherHRNet-w48](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w48_coco_512x512.py) | 512x512 | 0.686 | 0.873 | 0.741 | 0.731 | 0.892 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet48_coco_512x512-60fedcbc_20200712.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet48_coco_512x512_20200712.log.json) |
Results on COCO val2017 with multi-scale test. 3 default scales (\[2, 1, 0.5\]) are used
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HigherHRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_512x512.py) | 512x512 | 0.706 | 0.881 | 0.771 | 0.747 | 0.901 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_512x512-8ae85183_20200713.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_512x512_20200713.log.json) |
| [HigherHRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_640x640.py) | 640x640 | 0.706 | 0.880 | 0.770 | 0.749 | 0.902 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_640x640-a22fe938_20200712.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_640x640_20200712.log.json) |
| [HigherHRNet-w48](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w48_coco_512x512.py) | 512x512 | 0.716 | 0.884 | 0.775 | 0.755 | 0.901 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet48_coco_512x512-60fedcbc_20200712.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet48_coco_512x512_20200712.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_coco.yml
Vendored
+106
View File
@@ -0,0 +1,106 @@
Collections:
- Name: HigherHRNet
Paper:
Title: 'HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose
Estimation'
URL: http://openaccess.thecvf.com/content_CVPR_2020/html/Cheng_HigherHRNet_Scale-Aware_Representation_Learning_for_Bottom-Up_Human_Pose_Estimation_CVPR_2020_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/backbones/higherhrnet.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_512x512.py
In Collection: HigherHRNet
Metadata:
Architecture: &id001
- Associative Embedding
- HigherHRNet
Training Data: COCO
Name: associative_embedding_higherhrnet_w32_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.677
AP@0.5: 0.87
AP@0.75: 0.738
AR: 0.723
AR@0.5: 0.89
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_512x512-8ae85183_20200713.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_640x640.py
In Collection: HigherHRNet
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_higherhrnet_w32_coco_640x640
Results:
- Dataset: COCO
Metrics:
AP: 0.686
AP@0.5: 0.871
AP@0.75: 0.747
AR: 0.733
AR@0.5: 0.898
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_640x640-a22fe938_20200712.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w48_coco_512x512.py
In Collection: HigherHRNet
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_higherhrnet_w48_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.686
AP@0.5: 0.873
AP@0.75: 0.741
AR: 0.731
AR@0.5: 0.892
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet48_coco_512x512-60fedcbc_20200712.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_512x512.py
In Collection: HigherHRNet
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_higherhrnet_w32_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.706
AP@0.5: 0.881
AP@0.75: 0.771
AR: 0.747
AR@0.5: 0.901
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_512x512-8ae85183_20200713.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_640x640.py
In Collection: HigherHRNet
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_higherhrnet_w32_coco_640x640
Results:
- Dataset: COCO
Metrics:
AP: 0.706
AP@0.5: 0.88
AP@0.75: 0.77
AR: 0.749
AR@0.5: 0.902
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_640x640-a22fe938_20200712.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w48_coco_512x512.py
In Collection: HigherHRNet
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_higherhrnet_w48_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.716
AP@0.5: 0.884
AP@0.75: 0.775
AR: 0.755
AR@0.5: 0.901
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet48_coco_512x512-60fedcbc_20200712.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_udp_coco.md
Vendored
+75
View File
@@ -0,0 +1,75 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2020/html/Cheng_HigherHRNet_Scale-Aware_Representation_Learning_for_Bottom-Up_Human_Pose_Estimation_CVPR_2020_paper.html">HigherHRNet (CVPR'2020)</a></summary>
```bibtex
@inproceedings{cheng2020higherhrnet,
title={HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation},
author={Cheng, Bowen and Xiao, Bin and Wang, Jingdong and Shi, Honghui and Huang, Thomas S and Zhang, Lei},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={5386--5395},
year={2020}
}
```
</details>
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2020/html/Huang_The_Devil_Is_in_the_Details_Delving_Into_Unbiased_Data_CVPR_2020_paper.html">UDP (CVPR'2020)</a></summary>
```bibtex
@InProceedings{Huang_2020_CVPR,
author = {Huang, Junjie and Zhu, Zheng and Guo, Feng and Huang, Guan},
title = {The Devil Is in the Details: Delving Into Unbiased Data Processing for Human Pose Estimation},
booktitle = {The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://link.springer.com/chapter/10.1007/978-3-319-10602-1_48">COCO (ECCV'2014)</a></summary>
```bibtex
@inproceedings{lin2014microsoft,
title={Microsoft coco: Common objects in context},
author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence},
booktitle={European conference on computer vision},
pages={740--755},
year={2014},
organization={Springer}
}
```
</details>
Results on COCO val2017 without multi-scale test
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HigherHRNet-w32_udp](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_512x512_udp.py) | 512x512 | 0.678 | 0.862 | 0.736 | 0.724 | 0.890 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_512x512_udp-8cc64794_20210222.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_512x512_udp_20210222.log.json) |
| [HigherHRNet-w48_udp](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w48_coco_512x512_udp.py) | 512x512 | 0.690 | 0.872 | 0.750 | 0.734 | 0.891 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet48_coco_512x512_udp-7cad61ef_20210222.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet48_coco_512x512_udp_20210222.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_udp_coco.yml
Vendored
+43
View File
@@ -0,0 +1,43 @@
Collections:
- Name: HigherHRNet
Paper:
Title: 'HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose
Estimation'
URL: http://openaccess.thecvf.com/content_CVPR_2020/html/Cheng_HigherHRNet_Scale-Aware_Representation_Learning_for_Bottom-Up_Human_Pose_Estimation_CVPR_2020_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/backbones/higherhrnet.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_512x512_udp.py
In Collection: HigherHRNet
Metadata:
Architecture: &id001
- Associative Embedding
- HigherHRNet
- UDP
Training Data: COCO
Name: associative_embedding_higherhrnet_w32_coco_512x512_udp
Results:
- Dataset: COCO
Metrics:
AP: 0.678
AP@0.5: 0.862
AP@0.75: 0.736
AR: 0.724
AR@0.5: 0.89
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_coco_512x512_udp-8cc64794_20210222.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w48_coco_512x512_udp.py
In Collection: HigherHRNet
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_higherhrnet_w48_coco_512x512_udp
Results:
- Dataset: COCO
Metrics:
AP: 0.69
AP@0.5: 0.872
AP@0.75: 0.75
AR: 0.734
AR@0.5: 0.891
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet48_coco_512x512_udp-7cad61ef_20210222.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_512x512.py
Vendored
+193
View File
@@ -0,0 +1,193 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128, 256],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=32,
num_joints=17,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[32],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_512x512_udp.py
Vendored
+197
View File
@@ -0,0 +1,197 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128, 256],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=32,
num_joints=17,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[32],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=False,
align_corners=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True,
use_udp=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40,
use_udp=True),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
use_udp=True,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1], use_udp=True),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
],
use_udp=True),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_640x640.py
Vendored
+193
View File
@@ -0,0 +1,193 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=640,
base_size=320,
base_sigma=2,
heatmap_size=[160, 320],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=32,
num_joints=17,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[32],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w32_coco_640x640_udp.py
Vendored
+197
View File
@@ -0,0 +1,197 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=640,
base_size=320,
base_sigma=2,
heatmap_size=[160, 320],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=32,
num_joints=17,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[32],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=False,
align_corners=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True,
use_udp=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40,
use_udp=True),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
use_udp=True,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1], use_udp=True),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
],
use_udp=True),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w48_coco_512x512.py
Vendored
+193
View File
@@ -0,0 +1,193 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128, 256],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w48-8ef0771d.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(48, 96)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(48, 96, 192)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(48, 96, 192, 384))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=48,
num_joints=17,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[48],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w48_coco_512x512_udp.py
Vendored
+197
View File
@@ -0,0 +1,197 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128, 256],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w48-8ef0771d.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(48, 96)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(48, 96, 192)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(48, 96, 192, 384))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=48,
num_joints=17,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[48],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=False,
align_corners=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True,
use_udp=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40,
use_udp=True),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
use_udp=True,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1], use_udp=True),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
],
use_udp=True),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hourglass_ae_coco.md
Vendored
+63
View File
@@ -0,0 +1,63 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [BACKBONE] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">HourglassAENet (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://link.springer.com/chapter/10.1007/978-3-319-10602-1_48">COCO (ECCV'2014)</a></summary>
```bibtex
@inproceedings{lin2014microsoft,
title={Microsoft coco: Common objects in context},
author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence},
booktitle={European conference on computer vision},
pages={740--755},
year={2014},
organization={Springer}
}
```
</details>
Results on COCO val2017 without multi-scale test
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [pose_hourglass_ae](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hourglass_ae_coco_512x512.py) | 512x512 | 0.613 | 0.833 | 0.667 | 0.659 | 0.850 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hourglass_ae/hourglass_ae_coco_512x512-90af499f_20210920.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hourglass_ae/hourglass_ae_coco_512x512_20210920.log.json) |
Results on COCO val2017 with multi-scale test. 3 default scales (\[2, 1, 0.5\]) are used
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [pose_hourglass_ae](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hourglass_ae_coco_512x512.py) | 512x512 | 0.667 | 0.855 | 0.723 | 0.707 | 0.877 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hourglass_ae/hourglass_ae_coco_512x512-90af499f_20210920.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hourglass_ae/hourglass_ae_coco_512x512_20210920.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hourglass_ae_coco.yml
Vendored
+41
View File
@@ -0,0 +1,41 @@
Collections:
- Name: Associative Embedding
Paper:
Title: 'Associative embedding: End-to-end learning for joint detection and grouping'
URL: https://arxiv.org/abs/1611.05424
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/algorithms/associative_embedding.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hourglass_ae_coco_512x512.py
In Collection: Associative Embedding
Metadata:
Architecture: &id001
- Associative Embedding
- HourglassAENet
Training Data: COCO
Name: associative_embedding_hourglass_ae_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.613
AP@0.5: 0.833
AP@0.75: 0.667
AR: 0.659
AR@0.5: 0.85
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hourglass_ae/hourglass_ae_coco_512x512-90af499f_20210920.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hourglass_ae_coco_512x512.py
In Collection: Associative Embedding
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_hourglass_ae_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.667
AP@0.5: 0.855
AP@0.75: 0.723
AR: 0.707
AR@0.5: 0.877
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hourglass_ae/hourglass_ae_coco_512x512-90af499f_20210920.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hourglass_ae_coco_512x512.py
Vendored
+167
View File
@@ -0,0 +1,167 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained=None,
backbone=dict(
type='HourglassAENet',
num_stacks=4,
out_channels=34,
),
keypoint_head=dict(
type='AEMultiStageHead',
in_channels=34,
out_channels=34,
num_stages=4,
num_deconv_layers=0,
extra=dict(final_conv_kernel=0),
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=4,
ae_loss_type='exp',
with_heatmaps_loss=[True, True, True, True],
with_ae_loss=[True, True, True, True],
push_loss_factor=[0.001, 0.001, 0.001, 0.001],
pull_loss_factor=[0.001, 0.001, 0.001, 0.001],
heatmaps_loss_factor=[1.0, 1.0, 1.0, 1.0])),
train_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
img_size=data_cfg['image_size']),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True, True, True],
with_ae=[True, True, True, True],
select_output_index=[3],
project2image=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='MultitaskGatherTarget',
pipeline_list=[
[dict(type='BottomUpGenerateTarget', sigma=2, max_num_people=30)],
],
pipeline_indices=[0] * 4,
keys=['targets', 'masks', 'joints']),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=6),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_coco.md
Vendored
+65
View File
@@ -0,0 +1,65 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2019/html/Sun_Deep_High-Resolution_Representation_Learning_for_Human_Pose_Estimation_CVPR_2019_paper.html">HRNet (CVPR'2019)</a></summary>
```bibtex
@inproceedings{sun2019deep,
title={Deep high-resolution representation learning for human pose estimation},
author={Sun, Ke and Xiao, Bin and Liu, Dong and Wang, Jingdong},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={5693--5703},
year={2019}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://link.springer.com/chapter/10.1007/978-3-319-10602-1_48">COCO (ECCV'2014)</a></summary>
```bibtex
@inproceedings{lin2014microsoft,
title={Microsoft coco: Common objects in context},
author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence},
booktitle={European conference on computer vision},
pages={740--755},
year={2014},
organization={Springer}
}
```
</details>
Results on COCO val2017 without multi-scale test
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w32_coco_512x512.py) | 512x512 | 0.654 | 0.863 | 0.720 | 0.710 | 0.892 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_coco_512x512-bcb8c247_20200816.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_coco_512x512_20200816.log.json) |
| [HRNet-w48](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_coco_512x512.py) | 512x512 | 0.665 | 0.860 | 0.727 | 0.716 | 0.889 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_coco_512x512-cf72fcdf_20200816.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_coco_512x512_20200816.log.json) |
Results on COCO val2017 with multi-scale test. 3 default scales (\[2, 1, 0.5\]) are used
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w32_coco_512x512.py) | 512x512 | 0.698 | 0.877 | 0.760 | 0.748 | 0.907 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_coco_512x512-bcb8c247_20200816.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_coco_512x512_20200816.log.json) |
| [HRNet-w48](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_coco_512x512.py) | 512x512 | 0.712 | 0.880 | 0.771 | 0.757 | 0.909 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_coco_512x512-cf72fcdf_20200816.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_coco_512x512_20200816.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_coco.yml
Vendored
+73
View File
@@ -0,0 +1,73 @@
Collections:
- Name: HRNet
Paper:
Title: Deep high-resolution representation learning for human pose estimation
URL: http://openaccess.thecvf.com/content_CVPR_2019/html/Sun_Deep_High-Resolution_Representation_Learning_for_Human_Pose_Estimation_CVPR_2019_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/backbones/hrnet.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w32_coco_512x512.py
In Collection: HRNet
Metadata:
Architecture: &id001
- Associative Embedding
- HRNet
Training Data: COCO
Name: associative_embedding_hrnet_w32_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.654
AP@0.5: 0.863
AP@0.75: 0.72
AR: 0.71
AR@0.5: 0.892
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_coco_512x512-bcb8c247_20200816.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_coco_512x512.py
In Collection: HRNet
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_hrnet_w48_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.665
AP@0.5: 0.86
AP@0.75: 0.727
AR: 0.716
AR@0.5: 0.889
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_coco_512x512-cf72fcdf_20200816.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w32_coco_512x512.py
In Collection: HRNet
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_hrnet_w32_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.698
AP@0.5: 0.877
AP@0.75: 0.76
AR: 0.748
AR@0.5: 0.907
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_coco_512x512-bcb8c247_20200816.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_coco_512x512.py
In Collection: HRNet
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_hrnet_w48_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.712
AP@0.5: 0.88
AP@0.75: 0.771
AR: 0.757
AR@0.5: 0.909
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_coco_512x512-cf72fcdf_20200816.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_udp_coco.md
Vendored
+75
View File
@@ -0,0 +1,75 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [BACKBONE] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2019/html/Sun_Deep_High-Resolution_Representation_Learning_for_Human_Pose_Estimation_CVPR_2019_paper.html">HRNet (CVPR'2019)</a></summary>
```bibtex
@inproceedings{sun2019deep,
title={Deep high-resolution representation learning for human pose estimation},
author={Sun, Ke and Xiao, Bin and Liu, Dong and Wang, Jingdong},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={5693--5703},
year={2019}
}
```
</details>
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2020/html/Huang_The_Devil_Is_in_the_Details_Delving_Into_Unbiased_Data_CVPR_2020_paper.html">UDP (CVPR'2020)</a></summary>
```bibtex
@InProceedings{Huang_2020_CVPR,
author = {Huang, Junjie and Zhu, Zheng and Guo, Feng and Huang, Guan},
title = {The Devil Is in the Details: Delving Into Unbiased Data Processing for Human Pose Estimation},
booktitle = {The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://link.springer.com/chapter/10.1007/978-3-319-10602-1_48">COCO (ECCV'2014)</a></summary>
```bibtex
@inproceedings{lin2014microsoft,
title={Microsoft coco: Common objects in context},
author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence},
booktitle={European conference on computer vision},
pages={740--755},
year={2014},
organization={Springer}
}
```
</details>
Results on COCO val2017 without multi-scale test
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HRNet-w32_udp](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w32_coco_512x512_udp.py) | 512x512 | 0.671 | 0.863 | 0.729 | 0.717 | 0.889 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_coco_512x512_udp-91663bf9_20210220.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_coco_512x512_udp_20210220.log.json) |
| [HRNet-w48_udp](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_coco_512x512_udp.py) | 512x512 | 0.681 | 0.872 | 0.741 | 0.725 | 0.892 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_coco_512x512_udp-de08fd8c_20210222.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_coco_512x512_udp_20210222.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_udp_coco.yml
Vendored
+43
View File
@@ -0,0 +1,43 @@
Collections:
- Name: UDP
Paper:
Title: 'The Devil Is in the Details: Delving Into Unbiased Data Processing for
Human Pose Estimation'
URL: http://openaccess.thecvf.com/content_CVPR_2020/html/Huang_The_Devil_Is_in_the_Details_Delving_Into_Unbiased_Data_CVPR_2020_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/techniques/udp.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w32_coco_512x512_udp.py
In Collection: UDP
Metadata:
Architecture: &id001
- Associative Embedding
- HRNet
- UDP
Training Data: COCO
Name: associative_embedding_hrnet_w32_coco_512x512_udp
Results:
- Dataset: COCO
Metrics:
AP: 0.671
AP@0.5: 0.863
AP@0.75: 0.729
AR: 0.717
AR@0.5: 0.889
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hrnet_w32_coco_512x512_udp-91663bf9_20210220.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_coco_512x512_udp.py
In Collection: UDP
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_hrnet_w48_coco_512x512_udp
Results:
- Dataset: COCO
Metrics:
AP: 0.681
AP@0.5: 0.872
AP@0.75: 0.741
AR: 0.725
AR@0.5: 0.892
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_coco_512x512_udp-de08fd8c_20210222.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w32_coco_512x512.py
Vendored
+189
View File
@@ -0,0 +1,189 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AESimpleHead',
in_channels=32,
num_joints=17,
num_deconv_layers=0,
tag_per_joint=True,
with_ae_loss=[True],
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w32_coco_512x512_udp.py
Vendored
+193
View File
@@ -0,0 +1,193 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AESimpleHead',
in_channels=32,
num_joints=17,
num_deconv_layers=0,
tag_per_joint=True,
with_ae_loss=[True],
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=False,
align_corners=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True,
use_udp=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40,
use_udp=True),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
use_udp=True,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1], use_udp=True),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
],
use_udp=True),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w32_coco_640x640.py
Vendored
+189
View File
@@ -0,0 +1,189 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=640,
base_size=320,
base_sigma=2,
heatmap_size=[160],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AESimpleHead',
in_channels=32,
num_joints=17,
num_deconv_layers=0,
tag_per_joint=True,
with_ae_loss=[True],
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w32_coco_640x640_udp.py
Vendored
+193
View File
@@ -0,0 +1,193 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=640,
base_size=320,
base_sigma=2,
heatmap_size=[160],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AESimpleHead',
in_channels=32,
num_joints=17,
num_deconv_layers=0,
tag_per_joint=True,
with_ae_loss=[True],
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=False,
align_corners=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True,
use_udp=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40,
use_udp=True),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
use_udp=True,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1], use_udp=True),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
],
use_udp=True),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_coco_512x512.py
Vendored
+189
View File
@@ -0,0 +1,189 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w48-8ef0771d.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(48, 96)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(48, 96, 192)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(48, 96, 192, 384))),
),
keypoint_head=dict(
type='AESimpleHead',
in_channels=48,
num_joints=17,
num_deconv_layers=0,
tag_per_joint=True,
with_ae_loss=[True],
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_coco_512x512_udp.py
Vendored
+193
View File
@@ -0,0 +1,193 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w48-8ef0771d.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(48, 96)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(48, 96, 192)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(48, 96, 192, 384))),
),
keypoint_head=dict(
type='AESimpleHead',
in_channels=48,
num_joints=17,
num_deconv_layers=0,
tag_per_joint=True,
with_ae_loss=[True],
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=False,
align_corners=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True,
use_udp=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40,
use_udp=True),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
use_udp=True,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1], use_udp=True),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
],
use_udp=True),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_coco_640x640.py
Vendored
+189
View File
@@ -0,0 +1,189 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=640,
base_size=320,
base_sigma=2,
heatmap_size=[160],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w48-8ef0771d.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(48, 96)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(48, 96, 192)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(48, 96, 192, 384))),
),
keypoint_head=dict(
type='AESimpleHead',
in_channels=48,
num_joints=17,
num_deconv_layers=0,
tag_per_joint=True,
with_ae_loss=[True],
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=8),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_coco_640x640_udp.py
Vendored
+193
View File
@@ -0,0 +1,193 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=640,
base_size=320,
base_sigma=2,
heatmap_size=[160],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w48-8ef0771d.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(48, 96)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(48, 96, 192)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(48, 96, 192, 384))),
),
keypoint_head=dict(
type='AESimpleHead',
in_channels=48,
num_joints=17,
num_deconv_layers=0,
tag_per_joint=True,
with_ae_loss=[True],
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=False,
align_corners=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True,
use_udp=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40,
use_udp=True),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
use_udp=True,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1], use_udp=True),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
],
use_udp=True),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=8),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/mobilenetv2_coco.md
Vendored
+63
View File
@@ -0,0 +1,63 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [BACKBONE] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_cvpr_2018/html/Sandler_MobileNetV2_Inverted_Residuals_CVPR_2018_paper.html">MobilenetV2 (CVPR'2018)</a></summary>
```bibtex
@inproceedings{sandler2018mobilenetv2,
title={Mobilenetv2: Inverted residuals and linear bottlenecks},
author={Sandler, Mark and Howard, Andrew and Zhu, Menglong and Zhmoginov, Andrey and Chen, Liang-Chieh},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={4510--4520},
year={2018}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://link.springer.com/chapter/10.1007/978-3-319-10602-1_48">COCO (ECCV'2014)</a></summary>
```bibtex
@inproceedings{lin2014microsoft,
title={Microsoft coco: Common objects in context},
author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence},
booktitle={European conference on computer vision},
pages={740--755},
year={2014},
organization={Springer}
}
```
</details>
Results on COCO val2017 without multi-scale test
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [pose_mobilenetv2](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/mobilenetv2_coco_512x512.py) | 512x512 | 0.380 | 0.671 | 0.368 | 0.473 | 0.741 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/mobilenetv2_coco_512x512-4d96e309_20200816.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/mobilenetv2_coco_512x512_20200816.log.json) |
Results on COCO val2017 with multi-scale test. 3 default scales (\[2, 1, 0.5\]) are used
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [pose_mobilenetv2](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/mobilenetv2_coco_512x512.py) | 512x512 | 0.442 | 0.696 | 0.422 | 0.517 | 0.766 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/mobilenetv2_coco_512x512-4d96e309_20200816.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/mobilenetv2_coco_512x512_20200816.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/mobilenetv2_coco.yml
Vendored
+41
View File
@@ -0,0 +1,41 @@
Collections:
- Name: MobilenetV2
Paper:
Title: 'Mobilenetv2: Inverted residuals and linear bottlenecks'
URL: http://openaccess.thecvf.com/content_cvpr_2018/html/Sandler_MobileNetV2_Inverted_Residuals_CVPR_2018_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/backbones/mobilenetv2.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/mobilenetv2_coco_512x512.py
In Collection: MobilenetV2
Metadata:
Architecture: &id001
- Associative Embedding
- MobilenetV2
Training Data: COCO
Name: associative_embedding_mobilenetv2_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.38
AP@0.5: 0.671
AP@0.75: 0.368
AR: 0.473
AR@0.5: 0.741
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/mobilenetv2_coco_512x512-4d96e309_20200816.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/mobilenetv2_coco_512x512.py
In Collection: MobilenetV2
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_mobilenetv2_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.442
AP@0.5: 0.696
AP@0.75: 0.422
AR: 0.517
AR@0.5: 0.766
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/mobilenetv2_coco_512x512-4d96e309_20200816.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/mobilenetv2_coco_512x512.py
Vendored
+158
View File
@@ -0,0 +1,158 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='mmcls://mobilenet_v2',
backbone=dict(type='MobileNetV2', widen_factor=1., out_indices=(7, )),
keypoint_head=dict(
type='AESimpleHead',
in_channels=1280,
num_joints=17,
tag_per_joint=True,
with_ae_loss=[True],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=1,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res101_coco_512x512.py
Vendored
+158
View File
@@ -0,0 +1,158 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='torchvision://resnet101',
backbone=dict(type='ResNet', depth=101),
keypoint_head=dict(
type='AESimpleHead',
in_channels=2048,
num_joints=17,
tag_per_joint=True,
with_ae_loss=[True],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res101_coco_640x640.py
Vendored
+158
View File
@@ -0,0 +1,158 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=640,
base_size=320,
base_sigma=2,
heatmap_size=[160],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='torchvision://resnet101',
backbone=dict(type='ResNet', depth=101),
keypoint_head=dict(
type='AESimpleHead',
in_channels=2048,
num_joints=17,
tag_per_joint=True,
with_ae_loss=[True],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res152_coco_512x512.py
Vendored
+158
View File
@@ -0,0 +1,158 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='torchvision://resnet152',
backbone=dict(type='ResNet', depth=152),
keypoint_head=dict(
type='AESimpleHead',
in_channels=2048,
num_joints=17,
tag_per_joint=True,
with_ae_loss=[True],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res152_coco_640x640.py
Vendored
+158
View File
@@ -0,0 +1,158 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=640,
base_size=320,
base_sigma=2,
heatmap_size=[160],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='torchvision://resnet152',
backbone=dict(type='ResNet', depth=152),
keypoint_head=dict(
type='AESimpleHead',
in_channels=2048,
num_joints=17,
tag_per_joint=True,
with_ae_loss=[True],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res50_coco_512x512.py
Vendored
+159
View File
@@ -0,0 +1,159 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='torchvision://resnet50',
backbone=dict(type='ResNet', depth=50),
keypoint_head=dict(
type='AESimpleHead',
in_channels=2048,
num_joints=17,
tag_per_joint=True,
with_ae_loss=[True],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0],
)),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=1,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res50_coco_640x640.py
Vendored
+158
View File
@@ -0,0 +1,158 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
data_cfg = dict(
image_size=640,
base_size=320,
base_sigma=2,
heatmap_size=[160],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='torchvision://resnet50',
backbone=dict(type='ResNet', depth=50),
keypoint_head=dict(
type='AESimpleHead',
in_channels=2048,
num_joints=17,
tag_per_joint=True,
with_ae_loss=[True],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=17,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
workers_per_gpu=1,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/resnet_coco.md
Vendored
+69
View File
@@ -0,0 +1,69 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [BACKBONE] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html">ResNet (CVPR'2016)</a></summary>
```bibtex
@inproceedings{he2016deep,
title={Deep residual learning for image recognition},
author={He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={770--778},
year={2016}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://link.springer.com/chapter/10.1007/978-3-319-10602-1_48">COCO (ECCV'2014)</a></summary>
```bibtex
@inproceedings{lin2014microsoft,
title={Microsoft coco: Common objects in context},
author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence},
booktitle={European conference on computer vision},
pages={740--755},
year={2014},
organization={Springer}
}
```
</details>
Results on COCO val2017 without multi-scale test
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [pose_resnet_50](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res50_coco_512x512.py) | 512x512 | 0.466 | 0.742 | 0.479 | 0.552 | 0.797 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/res50_coco_512x512-5521bead_20200816.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/res50_coco_512x512_20200816.log.json) |
| [pose_resnet_50](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res50_coco_640x640.py) | 640x640 | 0.479 | 0.757 | 0.487 | 0.566 | 0.810 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/res50_coco_640x640-2046f9cb_20200822.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/res50_coco_640x640_20200822.log.json) |
| [pose_resnet_101](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res101_coco_512x512.py) | 512x512 | 0.554 | 0.807 | 0.599 | 0.622 | 0.841 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/res101_coco_512x512-e0c95157_20200816.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/res101_coco_512x512_20200816.log.json) |
| [pose_resnet_152](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res152_coco_512x512.py) | 512x512 | 0.595 | 0.829 | 0.648 | 0.651 | 0.856 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/res152_coco_512x512-364eb38d_20200822.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/res152_coco_512x512_20200822.log.json) |
Results on COCO val2017 with multi-scale test. 3 default scales (\[2, 1, 0.5\]) are used
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [pose_resnet_50](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res50_coco_512x512.py) | 512x512 | 0.503 | 0.765 | 0.521 | 0.591 | 0.821 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/res50_coco_512x512-5521bead_20200816.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/res50_coco_512x512_20200816.log.json) |
| [pose_resnet_50](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res50_coco_640x640.py) | 640x640 | 0.525 | 0.784 | 0.542 | 0.610 | 0.832 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/res50_coco_640x640-2046f9cb_20200822.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/res50_coco_640x640_20200822.log.json) |
| [pose_resnet_101](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res101_coco_512x512.py) | 512x512 | 0.603 | 0.831 | 0.641 | 0.668 | 0.870 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/res101_coco_512x512-e0c95157_20200816.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/res101_coco_512x512_20200816.log.json) |
| [pose_resnet_152](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res152_coco_512x512.py) | 512x512 | 0.660 | 0.860 | 0.713 | 0.709 | 0.889 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/res152_coco_512x512-364eb38d_20200822.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/res152_coco_512x512_20200822.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/resnet_coco.yml
Vendored
+137
View File
@@ -0,0 +1,137 @@
Collections:
- Name: Associative Embedding
Paper:
Title: 'Associative embedding: End-to-end learning for joint detection and grouping'
URL: https://arxiv.org/abs/1611.05424
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/algorithms/associative_embedding.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res50_coco_512x512.py
In Collection: Associative Embedding
Metadata:
Architecture: &id001
- Associative Embedding
- ResNet
Training Data: COCO
Name: associative_embedding_res50_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.466
AP@0.5: 0.742
AP@0.75: 0.479
AR: 0.552
AR@0.5: 0.797
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/res50_coco_512x512-5521bead_20200816.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res50_coco_640x640.py
In Collection: Associative Embedding
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_res50_coco_640x640
Results:
- Dataset: COCO
Metrics:
AP: 0.479
AP@0.5: 0.757
AP@0.75: 0.487
AR: 0.566
AR@0.5: 0.81
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/res50_coco_640x640-2046f9cb_20200822.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res101_coco_512x512.py
In Collection: Associative Embedding
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_res101_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.554
AP@0.5: 0.807
AP@0.75: 0.599
AR: 0.622
AR@0.5: 0.841
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/res101_coco_512x512-e0c95157_20200816.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res152_coco_512x512.py
In Collection: Associative Embedding
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_res152_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.595
AP@0.5: 0.829
AP@0.75: 0.648
AR: 0.651
AR@0.5: 0.856
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/res152_coco_512x512-364eb38d_20200822.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res50_coco_512x512.py
In Collection: Associative Embedding
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_res50_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.503
AP@0.5: 0.765
AP@0.75: 0.521
AR: 0.591
AR@0.5: 0.821
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/res50_coco_512x512-5521bead_20200816.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res50_coco_640x640.py
In Collection: Associative Embedding
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_res50_coco_640x640
Results:
- Dataset: COCO
Metrics:
AP: 0.525
AP@0.5: 0.784
AP@0.75: 0.542
AR: 0.61
AR@0.5: 0.832
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/res50_coco_640x640-2046f9cb_20200822.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res101_coco_512x512.py
In Collection: Associative Embedding
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_res101_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.603
AP@0.5: 0.831
AP@0.75: 0.641
AR: 0.668
AR@0.5: 0.87
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/res101_coco_512x512-e0c95157_20200816.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/res152_coco_512x512.py
In Collection: Associative Embedding
Metadata:
Architecture: *id001
Training Data: COCO
Name: associative_embedding_res152_coco_512x512
Results:
- Dataset: COCO
Metrics:
AP: 0.66
AP@0.5: 0.86
AP@0.75: 0.713
AR: 0.709
AR@0.5: 0.889
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/res152_coco_512x512-364eb38d_20200822.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_crowdpose.md
Vendored
+61
View File
@@ -0,0 +1,61 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2020/html/Cheng_HigherHRNet_Scale-Aware_Representation_Learning_for_Bottom-Up_Human_Pose_Estimation_CVPR_2020_paper.html">HigherHRNet (CVPR'2020)</a></summary>
```bibtex
@inproceedings{cheng2020higherhrnet,
title={HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation},
author={Cheng, Bowen and Xiao, Bin and Wang, Jingdong and Shi, Honghui and Huang, Thomas S and Zhang, Lei},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={5386--5395},
year={2020}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2019/html/Li_CrowdPose_Efficient_Crowded_Scenes_Pose_Estimation_and_a_New_Benchmark_CVPR_2019_paper.html">CrowdPose (CVPR'2019)</a></summary>
```bibtex
@article{li2018crowdpose,
title={CrowdPose: Efficient Crowded Scenes Pose Estimation and A New Benchmark},
author={Li, Jiefeng and Wang, Can and Zhu, Hao and Mao, Yihuan and Fang, Hao-Shu and Lu, Cewu},
journal={arXiv preprint arXiv:1812.00324},
year={2018}
}
```
</details>
Results on CrowdPose test without multi-scale test
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AP (E) | AP (M) | AP (H) | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: | :------: |
| [HigherHRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_w32_crowdpose_512x512.py) | 512x512 | 0.655 | 0.859 | 0.705 | 0.728 | 0.660 | 0.577 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_crowdpose_512x512-1aa4a132_20201017.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_crowdpose_512x512_20201017.log.json) |
Results on CrowdPose test with multi-scale test. 2 scales (\[2, 1\]) are used
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AP (E) | AP (M) | AP (H) | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: | :------: |
| [HigherHRNet-w32](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_w32_crowdpose_512x512.py) | 512x512 | 0.661 | 0.864 | 0.710 | 0.742 | 0.670 | 0.566 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_crowdpose_512x512-1aa4a132_20201017.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_crowdpose_512x512_20201017.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_crowdpose.yml
Vendored
+44
View File
@@ -0,0 +1,44 @@
Collections:
- Name: HigherHRNet
Paper:
Title: 'HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose
Estimation'
URL: http://openaccess.thecvf.com/content_CVPR_2020/html/Cheng_HigherHRNet_Scale-Aware_Representation_Learning_for_Bottom-Up_Human_Pose_Estimation_CVPR_2020_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/backbones/higherhrnet.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_w32_crowdpose_512x512.py
In Collection: HigherHRNet
Metadata:
Architecture: &id001
- Associative Embedding
- HigherHRNet
Training Data: CrowdPose
Name: associative_embedding_higherhrnet_w32_crowdpose_512x512
Results:
- Dataset: CrowdPose
Metrics:
AP: 0.655
AP (E): 0.728
AP (H): 0.577
AP (M): 0.66
AP@0.5: 0.859
AP@0.75: 0.705
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_crowdpose_512x512-1aa4a132_20201017.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_w32_crowdpose_512x512.py
In Collection: HigherHRNet
Metadata:
Architecture: *id001
Training Data: CrowdPose
Name: associative_embedding_higherhrnet_w32_crowdpose_512x512
Results:
- Dataset: CrowdPose
Metrics:
AP: 0.661
AP (E): 0.742
AP (H): 0.566
AP (M): 0.67
AP@0.5: 0.864
AP@0.75: 0.71
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/higher_hrnet32_crowdpose_512x512-1aa4a132_20201017.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_w32_crowdpose_512x512.py
Vendored
+192
View File
@@ -0,0 +1,192 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/crowdpose.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128, 256],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=32,
num_joints=14,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[32],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/crowdpose'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_trainval.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_w32_crowdpose_512x512_udp.py
Vendored
+196
View File
@@ -0,0 +1,196 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/crowdpose.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128, 256],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=32,
num_joints=14,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[32],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=False,
align_corners=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True,
use_udp=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40,
use_udp=True),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
use_udp=True,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1], use_udp=True),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
],
use_udp=True),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/crowdpose'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_trainval.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_w32_crowdpose_640x640.py
Vendored
+192
View File
@@ -0,0 +1,192 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/crowdpose.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=640,
base_size=320,
base_sigma=2,
heatmap_size=[160, 320],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=32,
num_joints=14,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[32],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/crowdpose'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_trainval.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_w32_crowdpose_640x640_udp.py
Vendored
+196
View File
@@ -0,0 +1,196 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/crowdpose.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=640,
base_size=320,
base_sigma=2,
heatmap_size=[160, 320],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=32,
num_joints=14,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[32],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=False,
align_corners=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True,
use_udp=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40,
use_udp=True),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
use_udp=True,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1], use_udp=True),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
],
use_udp=True),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/crowdpose'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_trainval.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_w48_crowdpose_512x512.py
Vendored
+192
View File
@@ -0,0 +1,192 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/crowdpose.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128, 256],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w48-8ef0771d.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(48, 96)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(48, 96, 192)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(48, 96, 192, 384))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=48,
num_joints=14,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[48],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/crowdpose'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_trainval.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/higherhrnet_w48_crowdpose_512x512_udp.py
Vendored
+196
View File
@@ -0,0 +1,196 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/crowdpose.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128, 256],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=2,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w48-8ef0771d.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(48, 96)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(48, 96, 192)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(48, 96, 192, 384))),
),
keypoint_head=dict(
type='AEHigherResolutionHead',
in_channels=48,
num_joints=14,
tag_per_joint=True,
extra=dict(final_conv_kernel=1, ),
num_deconv_layers=1,
num_deconv_filters=[48],
num_deconv_kernels=[4],
num_basic_blocks=4,
cat_output=[True],
with_ae_loss=[True, False],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=2,
ae_loss_type='exp',
with_ae_loss=[True, False],
push_loss_factor=[0.001, 0.001],
pull_loss_factor=[0.001, 0.001],
with_heatmaps_loss=[True, True],
heatmaps_loss_factor=[1.0, 1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True, True],
with_ae=[True, False],
project2image=False,
align_corners=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True,
use_udp=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40,
use_udp=True),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
use_udp=True,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1], use_udp=True),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
],
use_udp=True),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/crowdpose'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_trainval.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/mobilenetv2_crowdpose_512x512.py
Vendored
+157
View File
@@ -0,0 +1,157 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/crowdpose.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='mmcls://mobilenet_v2',
backbone=dict(type='MobileNetV2', widen_factor=1., out_indices=(7, )),
keypoint_head=dict(
type='AESimpleHead',
in_channels=1280,
num_joints=14,
tag_per_joint=True,
with_ae_loss=[True],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/crowdpose'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_trainval.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/res101_crowdpose_512x512.py
Vendored
+157
View File
@@ -0,0 +1,157 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/crowdpose.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='torchvision://resnet101',
backbone=dict(type='ResNet', depth=101),
keypoint_head=dict(
type='AESimpleHead',
in_channels=2048,
num_joints=14,
tag_per_joint=True,
with_ae_loss=[True],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/crowdpose'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_trainval.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/res152_crowdpose_512x512.py
Vendored
+157
View File
@@ -0,0 +1,157 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/crowdpose.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='torchvision://resnet152',
backbone=dict(type='ResNet', depth=152),
keypoint_head=dict(
type='AESimpleHead',
in_channels=2048,
num_joints=14,
tag_per_joint=True,
with_ae_loss=[True],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/crowdpose'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_trainval.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/crowdpose/res50_crowdpose_512x512.py
Vendored
+158
View File
@@ -0,0 +1,158 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/crowdpose.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.0015,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[200, 260])
total_epochs = 300
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='torchvision://resnet50',
backbone=dict(type='ResNet', depth=50),
keypoint_head=dict(
type='AESimpleHead',
in_channels=2048,
num_joints=14,
tag_per_joint=True,
with_ae_loss=[True],
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=14,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.001],
pull_loss_factor=[0.001],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0],
)),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/crowdpose'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=24),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_trainval.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpCrowdPoseDataset',
ann_file=f'{data_root}/annotations/mmpose_crowdpose_test.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/mhp/hrnet_mhp.md
Vendored
+62
View File
@@ -0,0 +1,62 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1611.05424">Associative Embedding (NIPS'2017)</a></summary>
```bibtex
@inproceedings{newell2017associative,
title={Associative embedding: End-to-end learning for joint detection and grouping},
author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
booktitle={Advances in neural information processing systems},
pages={2277--2287},
year={2017}
}
```
</details>
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2019/html/Sun_Deep_High-Resolution_Representation_Learning_for_Human_Pose_Estimation_CVPR_2019_paper.html">HRNet (CVPR'2019)</a></summary>
```bibtex
@inproceedings{sun2019deep,
title={Deep high-resolution representation learning for human pose estimation},
author={Sun, Ke and Xiao, Bin and Liu, Dong and Wang, Jingdong},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={5693--5703},
year={2019}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://dl.acm.org/doi/abs/10.1145/3240508.3240509">MHP (ACM MM'2018)</a></summary>
```bibtex
@inproceedings{zhao2018understanding,
title={Understanding humans in crowded scenes: Deep nested adversarial learning and a new benchmark for multi-human parsing},
author={Zhao, Jian and Li, Jianshu and Cheng, Yu and Sim, Terence and Yan, Shuicheng and Feng, Jiashi},
booktitle={Proceedings of the 26th ACM international conference on Multimedia},
pages={792--800},
year={2018}
}
```
</details>
Results on MHP v2.0 validation set without multi-scale test
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HRNet-w48](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_mhp_512x512.py) | 512x512 | 0.583 | 0.895 | 0.666 | 0.656 | 0.931 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_mhp_512x512-85a6ab6f_20201229.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_mhp_512x512_20201229.log.json) |
Results on MHP v2.0 validation set with multi-scale test. 3 default scales (\[2, 1, 0.5\]) are used
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :----------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [HRNet-w48](/configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_mhp_512x512.py) | 512x512 | 0.592 | 0.898 | 0.673 | 0.664 | 0.932 | [ckpt](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_mhp_512x512-85a6ab6f_20201229.pth) | [log](https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_mhp_512x512_20201229.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/mhp/hrnet_mhp.yml
Vendored
+41
View File
@@ -0,0 +1,41 @@
Collections:
- Name: HRNet
Paper:
Title: Deep high-resolution representation learning for human pose estimation
URL: http://openaccess.thecvf.com/content_CVPR_2019/html/Sun_Deep_High-Resolution_Representation_Learning_for_Human_Pose_Estimation_CVPR_2019_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/backbones/hrnet.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_mhp_512x512.py
In Collection: HRNet
Metadata:
Architecture: &id001
- Associative Embedding
- HRNet
Training Data: MHP
Name: associative_embedding_hrnet_w48_mhp_512x512
Results:
- Dataset: MHP
Metrics:
AP: 0.583
AP@0.5: 0.895
AP@0.75: 0.666
AR: 0.656
AR@0.5: 0.931
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_mhp_512x512-85a6ab6f_20201229.pth
- Config: configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/hrnet_w48_mhp_512x512.py
In Collection: HRNet
Metadata:
Architecture: *id001
Training Data: MHP
Name: associative_embedding_hrnet_w48_mhp_512x512
Results:
- Dataset: MHP
Metrics:
AP: 0.592
AP@0.5: 0.898
AP@0.75: 0.673
AR: 0.664
AR@0.5: 0.932
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/bottom_up/hrnet_w48_mhp_512x512-85a6ab6f_20201229.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/associative_embedding/mhp/hrnet_w48_mhp_512x512.py
Vendored
+187
View File
@@ -0,0 +1,187 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/mhp.py'
]
checkpoint_config = dict(interval=50)
evaluation = dict(interval=50, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=0.005,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[400, 550])
total_epochs = 600
channel_cfg = dict(
dataset_joints=16,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15])
data_cfg = dict(
image_size=512,
base_size=256,
base_sigma=2,
heatmap_size=[128],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
num_scales=1,
scale_aware_sigma=False,
)
# model settings
model = dict(
type='AssociativeEmbedding',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w48-8ef0771d.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(48, 96)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(48, 96, 192)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(48, 96, 192, 384))),
),
keypoint_head=dict(
type='AESimpleHead',
in_channels=48,
num_joints=16,
num_deconv_layers=0,
tag_per_joint=True,
with_ae_loss=[True],
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(
type='MultiLossFactory',
num_joints=16,
num_stages=1,
ae_loss_type='exp',
with_ae_loss=[True],
push_loss_factor=[0.01],
pull_loss_factor=[0.01],
with_heatmaps_loss=[True],
heatmaps_loss_factor=[1.0])),
train_cfg=dict(),
test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
align_corners=False,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=True))
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='BottomUpRandomAffine',
rot_factor=30,
scale_factor=[0.75, 1.5],
scale_type='short',
trans_factor=40),
dict(type='BottomUpRandomFlip', flip_prob=0.5),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='BottomUpGenerateTarget',
sigma=2,
max_num_people=30,
),
dict(
type='Collect',
keys=['img', 'joints', 'targets', 'masks'],
meta_keys=[]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='BottomUpGetImgSize', test_scale_factor=[1]),
dict(
type='BottomUpResizeAlign',
transforms=[
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'aug_data', 'test_scale_factor', 'base_size',
'center', 'scale', 'flip_index'
]),
]
test_pipeline = val_pipeline
data_root = 'data/mhp'
data = dict(
workers_per_gpu=2,
train_dataloader=dict(samples_per_gpu=16),
val_dataloader=dict(samples_per_gpu=1),
test_dataloader=dict(samples_per_gpu=1),
train=dict(
type='BottomUpMhpDataset',
ann_file=f'{data_root}/annotations/mhp_train.json',
img_prefix=f'{data_root}/train/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='BottomUpMhpDataset',
ann_file=f'{data_root}/annotations/mhp_val.json',
img_prefix=f'{data_root}/val/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='BottomUpMhpDataset',
ann_file=f'{data_root}/annotations/mhp_val.json',
img_prefix=f'{data_root}/val/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/deeppose/README.md
Vendored
+24
View File
@@ -0,0 +1,24 @@
# DeepPose: Human pose estimation via deep neural networks
## Introduction
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_cvpr_2014/html/Toshev_DeepPose_Human_Pose_2014_CVPR_paper.html">DeepPose (CVPR'2014)</a></summary>
```bibtex
@inproceedings{toshev2014deeppose,
title={Deeppose: Human pose estimation via deep neural networks},
author={Toshev, Alexander and Szegedy, Christian},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={1653--1660},
year={2014}
}
```
</details>
DeepPose first proposes using deep neural networks (DNNs) to tackle the problem of human pose estimation.
It follows the top-down paradigm, that first detects human bounding boxes and then estimates poses.
It learns to directly regress the human body keypoint coordinates.
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/deeppose/coco/res101_coco_256x192.py
Vendored
+132
View File
@@ -0,0 +1,132 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet101',
backbone=dict(type='ResNet', depth=101, num_stages=4, out_indices=(3, )),
neck=dict(type='GlobalAveragePooling'),
keypoint_head=dict(
type='DeepposeRegressionHead',
in_channels=2048,
num_joints=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='SmoothL1Loss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(flip_test=True))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTargetRegression'),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/deeppose/coco/res152_coco_256x192.py
Vendored
+132
View File
@@ -0,0 +1,132 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet152',
backbone=dict(type='ResNet', depth=152, num_stages=4, out_indices=(3, )),
neck=dict(type='GlobalAveragePooling'),
keypoint_head=dict(
type='DeepposeRegressionHead',
in_channels=2048,
num_joints=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='SmoothL1Loss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(flip_test=True))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTargetRegression'),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/deeppose/coco/res50_coco_256x192.py
Vendored
+132
View File
@@ -0,0 +1,132 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet50',
backbone=dict(type='ResNet', depth=50, num_stages=4, out_indices=(3, )),
neck=dict(type='GlobalAveragePooling'),
keypoint_head=dict(
type='DeepposeRegressionHead',
in_channels=2048,
num_joints=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='SmoothL1Loss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(flip_test=True))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTargetRegression'),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/deeppose/coco/resnet_coco.md
Vendored
+59
View File
@@ -0,0 +1,59 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_cvpr_2014/html/Toshev_DeepPose_Human_Pose_2014_CVPR_paper.html">DeepPose (CVPR'2014)</a></summary>
```bibtex
@inproceedings{toshev2014deeppose,
title={Deeppose: Human pose estimation via deep neural networks},
author={Toshev, Alexander and Szegedy, Christian},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={1653--1660},
year={2014}
}
```
</details>
<!-- [BACKBONE] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html">ResNet (CVPR'2016)</a></summary>
```bibtex
@inproceedings{he2016deep,
title={Deep residual learning for image recognition},
author={He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={770--778},
year={2016}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://link.springer.com/chapter/10.1007/978-3-319-10602-1_48">COCO (ECCV'2014)</a></summary>
```bibtex
@inproceedings{lin2014microsoft,
title={Microsoft coco: Common objects in context},
author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence},
booktitle={European conference on computer vision},
pages={740--755},
year={2014},
organization={Springer}
}
```
</details>
Results on COCO val2017 with detector having human AP of 56.4 on COCO val2017 dataset
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [deeppose_resnet_50](/configs/body/2d_kpt_sview_rgb_img/deeppose/coco/res50_coco_256x192.py) | 256x192 | 0.526 | 0.816 | 0.586 | 0.638 | 0.887 | [ckpt](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res50_coco_256x192-f6de6c0e_20210205.pth) | [log](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res50_coco_256x192_20210205.log.json) |
| [deeppose_resnet_101](/configs/body/2d_kpt_sview_rgb_img/deeppose/coco/res101_coco_256x192.py) | 256x192 | 0.560 | 0.832 | 0.628 | 0.668 | 0.900 | [ckpt](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res101_coco_256x192-2f247111_20210205.pth) | [log](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res101_coco_256x192_20210205.log.json) |
| [deeppose_resnet_152](/configs/body/2d_kpt_sview_rgb_img/deeppose/coco/res152_coco_256x192.py) | 256x192 | 0.583 | 0.843 | 0.659 | 0.686 | 0.907 | [ckpt](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res152_coco_256x192-7df89a88_20210205.pth) | [log](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res152_coco_256x192_20210205.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/deeppose/coco/resnet_coco.yml
Vendored
+57
View File
@@ -0,0 +1,57 @@
Collections:
- Name: ResNet
Paper:
Title: Deep residual learning for image recognition
URL: http://openaccess.thecvf.com/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/backbones/resnet.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/deeppose/coco/res50_coco_256x192.py
In Collection: ResNet
Metadata:
Architecture: &id001
- DeepPose
- ResNet
Training Data: COCO
Name: deeppose_res50_coco_256x192
Results:
- Dataset: COCO
Metrics:
AP: 0.526
AP@0.5: 0.816
AP@0.75: 0.586
AR: 0.638
AR@0.5: 0.887
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res50_coco_256x192-f6de6c0e_20210205.pth
- Config: configs/body/2d_kpt_sview_rgb_img/deeppose/coco/res101_coco_256x192.py
In Collection: ResNet
Metadata:
Architecture: *id001
Training Data: COCO
Name: deeppose_res101_coco_256x192
Results:
- Dataset: COCO
Metrics:
AP: 0.56
AP@0.5: 0.832
AP@0.75: 0.628
AR: 0.668
AR@0.5: 0.9
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res101_coco_256x192-2f247111_20210205.pth
- Config: configs/body/2d_kpt_sview_rgb_img/deeppose/coco/res152_coco_256x192.py
In Collection: ResNet
Metadata:
Architecture: *id001
Training Data: COCO
Name: deeppose_res152_coco_256x192
Results:
- Dataset: COCO
Metrics:
AP: 0.583
AP@0.5: 0.843
AP@0.75: 0.659
AR: 0.686
AR@0.5: 0.907
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res152_coco_256x192-7df89a88_20210205.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/deeppose/mpii/res101_mpii_256x256.py
Vendored
+120
View File
@@ -0,0 +1,120 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/mpii.py'
]
evaluation = dict(interval=10, metric='PCKh', save_best='PCKh')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
log_config = dict(
interval=50, hooks=[
dict(type='TextLoggerHook'),
])
channel_cfg = dict(
num_output_channels=16,
dataset_joints=16,
dataset_channel=list(range(16)),
inference_channel=list(range(16)))
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet101',
backbone=dict(type='ResNet', depth=101, num_stages=4, out_indices=(3, )),
neck=dict(type='GlobalAveragePooling'),
keypoint_head=dict(
type='DeepposeRegressionHead',
in_channels=2048,
num_joints=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='SmoothL1Loss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(flip_test=True))
data_cfg = dict(
image_size=[256, 256],
heatmap_size=[64, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
use_gt_bbox=True,
bbox_file=None,
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTargetRegression'),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=['image_file', 'center', 'scale', 'rotation', 'flip_pairs']),
]
test_pipeline = val_pipeline
data_root = 'data/mpii'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownMpiiDataset',
ann_file=f'{data_root}/annotations/mpii_train.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownMpiiDataset',
ann_file=f'{data_root}/annotations/mpii_val.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownMpiiDataset',
ann_file=f'{data_root}/annotations/mpii_val.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/deeppose/mpii/res152_mpii_256x256.py
Vendored
+120
View File
@@ -0,0 +1,120 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/mpii.py'
]
evaluation = dict(interval=10, metric='PCKh', save_best='PCKh')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
log_config = dict(
interval=50, hooks=[
dict(type='TextLoggerHook'),
])
channel_cfg = dict(
num_output_channels=16,
dataset_joints=16,
dataset_channel=list(range(16)),
inference_channel=list(range(16)))
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet152',
backbone=dict(type='ResNet', depth=152, num_stages=4, out_indices=(3, )),
neck=dict(type='GlobalAveragePooling'),
keypoint_head=dict(
type='DeepposeRegressionHead',
in_channels=2048,
num_joints=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='SmoothL1Loss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(flip_test=True))
data_cfg = dict(
image_size=[256, 256],
heatmap_size=[64, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
use_gt_bbox=True,
bbox_file=None,
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTargetRegression'),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=['image_file', 'center', 'scale', 'rotation', 'flip_pairs']),
]
test_pipeline = val_pipeline
data_root = 'data/mpii'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownMpiiDataset',
ann_file=f'{data_root}/annotations/mpii_train.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownMpiiDataset',
ann_file=f'{data_root}/annotations/mpii_val.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownMpiiDataset',
ann_file=f'{data_root}/annotations/mpii_val.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/deeppose/mpii/res50_mpii_256x256.py
Vendored
+120
View File
@@ -0,0 +1,120 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/mpii.py'
]
evaluation = dict(interval=10, metric='PCKh', save_best='PCKh')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
log_config = dict(
interval=50, hooks=[
dict(type='TextLoggerHook'),
])
channel_cfg = dict(
num_output_channels=16,
dataset_joints=16,
dataset_channel=list(range(16)),
inference_channel=list(range(16)))
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet50',
backbone=dict(type='ResNet', depth=50, num_stages=4, out_indices=(3, )),
neck=dict(type='GlobalAveragePooling'),
keypoint_head=dict(
type='DeepposeRegressionHead',
in_channels=2048,
num_joints=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='SmoothL1Loss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(flip_test=True))
data_cfg = dict(
image_size=[256, 256],
heatmap_size=[64, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
use_gt_bbox=True,
bbox_file=None,
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTargetRegression'),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=['image_file', 'center', 'scale', 'rotation', 'flip_pairs']),
]
test_pipeline = val_pipeline
data_root = 'data/mpii'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownMpiiDataset',
ann_file=f'{data_root}/annotations/mpii_train.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownMpiiDataset',
ann_file=f'{data_root}/annotations/mpii_val.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownMpiiDataset',
ann_file=f'{data_root}/annotations/mpii_val.json',
img_prefix=f'{data_root}/images/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/deeppose/mpii/resnet_mpii.md
Vendored
+58
View File
@@ -0,0 +1,58 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_cvpr_2014/html/Toshev_DeepPose_Human_Pose_2014_CVPR_paper.html">DeepPose (CVPR'2014)</a></summary>
```bibtex
@inproceedings{toshev2014deeppose,
title={Deeppose: Human pose estimation via deep neural networks},
author={Toshev, Alexander and Szegedy, Christian},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={1653--1660},
year={2014}
}
```
</details>
<!-- [BACKBONE] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html">ResNet (CVPR'2016)</a></summary>
```bibtex
@inproceedings{he2016deep,
title={Deep residual learning for image recognition},
author={He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={770--778},
year={2016}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_cvpr_2014/html/Andriluka_2D_Human_Pose_2014_CVPR_paper.html">MPII (CVPR'2014)</a></summary>
```bibtex
@inproceedings{andriluka14cvpr,
author = {Mykhaylo Andriluka and Leonid Pishchulin and Peter Gehler and Schiele, Bernt},
title = {2D Human Pose Estimation: New Benchmark and State of the Art Analysis},
booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year = {2014},
month = {June}
}
```
</details>
Results on MPII val set
| Arch | Input Size | Mean | Mean@0.1 | ckpt | log |
| :--- | :--------: | :------: | :------: |:------: |:------: |
| [deeppose_resnet_50](/configs/body/2d_kpt_sview_rgb_img/deeppose/mpii/res50_mpii_256x256.py) | 256x256 | 0.825 | 0.174 | [ckpt](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res50_mpii_256x256-c63cd0b6_20210203.pth) | [log](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res50_mpii_256x256_20210203.log.json) |
| [deeppose_resnet_101](/configs/body/2d_kpt_sview_rgb_img/deeppose/mpii/res101_mpii_256x256.py) | 256x256 | 0.841 | 0.193 | [ckpt](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res101_mpii_256x256-87516a90_20210205.pth) | [log](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res101_mpii_256x256_20210205.log.json) |
| [deeppose_resnet_152](/configs/body/2d_kpt_sview_rgb_img/deeppose/mpii/res152_mpii_256x256.py) | 256x256 | 0.850 | 0.198 | [ckpt](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res152_mpii_256x256-15f5e6f9_20210205.pth) | [log](https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res152_mpii_256x256_20210205.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/deeppose/mpii/resnet_mpii.yml
Vendored
+48
View File
@@ -0,0 +1,48 @@
Collections:
- Name: ResNet
Paper:
Title: Deep residual learning for image recognition
URL: http://openaccess.thecvf.com/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/backbones/resnet.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/deeppose/mpii/res50_mpii_256x256.py
In Collection: ResNet
Metadata:
Architecture: &id001
- DeepPose
- ResNet
Training Data: MPII
Name: deeppose_res50_mpii_256x256
Results:
- Dataset: MPII
Metrics:
Mean: 0.825
Mean@0.1: 0.174
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res50_mpii_256x256-c63cd0b6_20210203.pth
- Config: configs/body/2d_kpt_sview_rgb_img/deeppose/mpii/res101_mpii_256x256.py
In Collection: ResNet
Metadata:
Architecture: *id001
Training Data: MPII
Name: deeppose_res101_mpii_256x256
Results:
- Dataset: MPII
Metrics:
Mean: 0.841
Mean@0.1: 0.193
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res101_mpii_256x256-87516a90_20210205.pth
- Config: configs/body/2d_kpt_sview_rgb_img/deeppose/mpii/res152_mpii_256x256.py
In Collection: ResNet
Metadata:
Architecture: *id001
Training Data: MPII
Name: deeppose_res152_mpii_256x256
Results:
- Dataset: MPII
Metrics:
Mean: 0.85
Mean@0.1: 0.198
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/top_down/deeppose/deeppose_res152_mpii_256x256-15f5e6f9_20210205.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/README.md
Vendored
+10
View File
@@ -0,0 +1,10 @@
# Top-down heatmap-based pose estimation
Top-down methods divide the task into two stages: human detection and pose estimation.
They perform human detection first, followed by single-person pose estimation given human bounding boxes.
Instead of estimating keypoint coordinates directly, the pose estimator will produce heatmaps which represent the
likelihood of being a keypoint.
Various neural network models have been proposed for better performance.
The popular ones include stacked hourglass networks, and HRNet.
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/ViTPose_base_aic_256x192.py
Vendored
+151
View File
@@ -0,0 +1,151 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='ViT',
img_size=(256, 192),
patch_size=16,
embed_dim=768,
depth=12,
num_heads=12,
ratio=1,
use_checkpoint=False,
mlp_ratio=4,
qkv_bias=True,
drop_path_rate=0.3,
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=768,
num_deconv_layers=2,
num_deconv_filters=(256, 256),
num_deconv_kernels=(4, 4),
extra=dict(final_conv_kernel=1, ),
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=2),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/ViTPose_huge_aic_256x192.py
Vendored
+151
View File
@@ -0,0 +1,151 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='ViT',
img_size=(256, 192),
patch_size=16,
embed_dim=1280,
depth=32,
num_heads=16,
ratio=1,
use_checkpoint=False,
mlp_ratio=4,
qkv_bias=True,
drop_path_rate=0.55,
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=1280,
num_deconv_layers=2,
num_deconv_filters=(256, 256),
num_deconv_kernels=(4, 4),
extra=dict(final_conv_kernel=1, ),
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=2),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/ViTPose_large_aic_256x192.py
Vendored
+151
View File
@@ -0,0 +1,151 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='ViT',
img_size=(256, 192),
patch_size=16,
embed_dim=1024,
depth=24,
num_heads=16,
ratio=1,
use_checkpoint=False,
mlp_ratio=4,
qkv_bias=True,
drop_path_rate=0.55,
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=1024,
num_deconv_layers=2,
num_deconv_filters=(256, 256),
num_deconv_kernels=(4, 4),
extra=dict(final_conv_kernel=1, ),
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=2),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/ViTPose_small_aic_256x192.py
Vendored
+151
View File
@@ -0,0 +1,151 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='ViT',
img_size=(256, 192),
patch_size=16,
embed_dim=384,
depth=12,
num_heads=12,
ratio=1,
use_checkpoint=False,
mlp_ratio=4,
qkv_bias=True,
drop_path_rate=0.3,
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=384,
num_deconv_layers=2,
num_deconv_filters=(256, 256),
num_deconv_kernels=(4, 4),
extra=dict(final_conv_kernel=1, ),
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=2),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/hrnet_aic.md
Vendored
+39
View File
@@ -0,0 +1,39 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2019/html/Sun_Deep_High-Resolution_Representation_Learning_for_Human_Pose_Estimation_CVPR_2019_paper.html">HRNet (CVPR'2019)</a></summary>
```bibtex
@inproceedings{sun2019deep,
title={Deep high-resolution representation learning for human pose estimation},
author={Sun, Ke and Xiao, Bin and Liu, Dong and Wang, Jingdong},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={5693--5703},
year={2019}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1711.06475">AI Challenger (ArXiv'2017)</a></summary>
```bibtex
@article{wu2017ai,
title={Ai challenger: A large-scale dataset for going deeper in image understanding},
author={Wu, Jiahong and Zheng, He and Zhao, Bo and Li, Yixin and Yan, Baoming and Liang, Rui and Wang, Wenjia and Zhou, Shipei and Lin, Guosen and Fu, Yanwei and others},
journal={arXiv preprint arXiv:1711.06475},
year={2017}
}
```
</details>
Results on AIC val set with ground-truth bounding boxes
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [pose_hrnet_w32](/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/hrnet_w32_aic_256x192.py) | 256x192 | 0.323 | 0.762 | 0.219 | 0.366 | 0.789 | [ckpt](https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w32_aic_256x192-30a4e465_20200826.pth) | [log](https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w32_aic_256x192_20200826.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/hrnet_aic.yml
Vendored
+24
View File
@@ -0,0 +1,24 @@
Collections:
- Name: HRNet
Paper:
Title: Deep high-resolution representation learning for human pose estimation
URL: http://openaccess.thecvf.com/content_CVPR_2019/html/Sun_Deep_High-Resolution_Representation_Learning_for_Human_Pose_Estimation_CVPR_2019_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/backbones/hrnet.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/hrnet_w32_aic_256x192.py
In Collection: HRNet
Metadata:
Architecture:
- HRNet
Training Data: AI Challenger
Name: topdown_heatmap_hrnet_w32_aic_256x192
Results:
- Dataset: AI Challenger
Metrics:
AP: 0.323
AP@0.5: 0.762
AP@0.75: 0.219
AR: 0.366
AR@0.5: 0.789
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w32_aic_256x192-30a4e465_20200826.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/hrnet_w32_aic_256x192.py
Vendored
+166
View File
@@ -0,0 +1,166 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=32,
out_channels=channel_cfg['num_output_channels'],
num_deconv_layers=0,
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=2),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/hrnet_w32_aic_384x288.py
Vendored
+166
View File
@@ -0,0 +1,166 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w32-36af842e.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))),
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=32,
out_channels=channel_cfg['num_output_channels'],
num_deconv_layers=0,
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[288, 384],
heatmap_size=[72, 96],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=3),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/hrnet_w48_aic_256x192.py
Vendored
+166
View File
@@ -0,0 +1,166 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w48-8ef0771d.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(48, 96)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(48, 96, 192)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(48, 96, 192, 384))),
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=48,
out_channels=channel_cfg['num_output_channels'],
num_deconv_layers=0,
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=2),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=32,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/hrnet_w48_aic_384x288.py
Vendored
+167
View File
@@ -0,0 +1,167 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup=None,
# warmup='linear',
# warmup_iters=500,
# warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained='https://download.openmmlab.com/mmpose/'
'pretrain_models/hrnet_w48-8ef0771d.pth',
backbone=dict(
type='HRNet',
in_channels=3,
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(48, 96)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(48, 96, 192)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(48, 96, 192, 384))),
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=48,
out_channels=channel_cfg['num_output_channels'],
num_deconv_layers=0,
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[288, 384],
heatmap_size=[72, 96],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=3),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=32,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/res101_aic_256x192.py
Vendored
+135
View File
@@ -0,0 +1,135 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet101',
backbone=dict(type='ResNet', depth=101),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=2048,
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=2),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/res101_aic_384x288.py
Vendored
+135
View File
@@ -0,0 +1,135 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet101',
backbone=dict(type='ResNet', depth=101),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=2048,
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[288, 384],
heatmap_size=[72, 96],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=3),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=32,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/res152_aic_256x192.py
Vendored
+135
View File
@@ -0,0 +1,135 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet152',
backbone=dict(type='ResNet', depth=152),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=2048,
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=2),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=32,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/res152_aic_384x288.py
Vendored
+135
View File
@@ -0,0 +1,135 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet152',
backbone=dict(type='ResNet', depth=152),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=2048,
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[288, 384],
heatmap_size=[72, 96],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=3),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=32,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/res50_aic_256x192.py
Vendored
+134
View File
@@ -0,0 +1,134 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet50',
backbone=dict(type='ResNet', depth=50),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=2048,
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=2),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/res50_aic_384x288.py
Vendored
+135
View File
@@ -0,0 +1,135 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/aic.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
channel_cfg = dict(
num_output_channels=14,
dataset_joints=14,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
],
inference_channel=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet50',
backbone=dict(type='ResNet', depth=50),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=2048,
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
data_cfg = dict(
image_size=[288, 384],
heatmap_size=[72, 96],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=True,
det_bbox_thr=0.0,
bbox_file='data/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(type='TopDownGenerateTarget', sigma=3),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/aic'
data = dict(
samples_per_gpu=64,
workers_per_gpu=2,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_train.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_train_20170902/'
'keypoint_train_images_20170902/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownAicDataset',
ann_file=f'{data_root}/annotations/aic_val.json',
img_prefix=f'{data_root}/ai_challenger_keypoint_validation_20170911/'
'keypoint_validation_images_20170911/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}))
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/resnet_aic.md
Vendored
+55
View File
@@ -0,0 +1,55 @@
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_ECCV_2018/html/Bin_Xiao_Simple_Baselines_for_ECCV_2018_paper.html">SimpleBaseline2D (ECCV'2018)</a></summary>
```bibtex
@inproceedings{xiao2018simple,
title={Simple baselines for human pose estimation and tracking},
author={Xiao, Bin and Wu, Haiping and Wei, Yichen},
booktitle={Proceedings of the European conference on computer vision (ECCV)},
pages={466--481},
year={2018}
}
```
</details>
<!-- [BACKBONE] -->
<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html">ResNet (CVPR'2016)</a></summary>
```bibtex
@inproceedings{he2016deep,
title={Deep residual learning for image recognition},
author={He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={770--778},
year={2016}
}
```
</details>
<!-- [DATASET] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/1711.06475">AI Challenger (ArXiv'2017)</a></summary>
```bibtex
@article{wu2017ai,
title={Ai challenger: A large-scale dataset for going deeper in image understanding},
author={Wu, Jiahong and Zheng, He and Zhao, Bo and Li, Yixin and Yan, Baoming and Liang, Rui and Wang, Wenjia and Zhou, Shipei and Lin, Guosen and Fu, Yanwei and others},
journal={arXiv preprint arXiv:1711.06475},
year={2017}
}
```
</details>
Results on AIC val set with ground-truth bounding boxes
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :-------------- | :-----------: | :------: | :------: | :------: | :------: | :------: |:------: |:------: |
| [pose_resnet_101](/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/res101_aic_256x192.py) | 256x192 | 0.294 | 0.736 | 0.174 | 0.337 | 0.763 | [ckpt](https://download.openmmlab.com/mmpose/top_down/resnet/res101_aic_256x192-79b35445_20200826.pth) | [log](https://download.openmmlab.com/mmpose/top_down/resnet/res101_aic_256x192_20200826.log.json) |
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/resnet_aic.yml
Vendored
+25
View File
@@ -0,0 +1,25 @@
Collections:
- Name: SimpleBaseline2D
Paper:
Title: Simple baselines for human pose estimation and tracking
URL: http://openaccess.thecvf.com/content_ECCV_2018/html/Bin_Xiao_Simple_Baselines_for_ECCV_2018_paper.html
README: https://github.com/open-mmlab/mmpose/blob/master/docs/en/papers/algorithms/simplebaseline2d.md
Models:
- Config: configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/aic/res101_aic_256x192.py
In Collection: SimpleBaseline2D
Metadata:
Architecture:
- SimpleBaseline2D
- ResNet
Training Data: AI Challenger
Name: topdown_heatmap_res101_aic_256x192
Results:
- Dataset: AI Challenger
Metrics:
AP: 0.294
AP@0.5: 0.736
AP@0.75: 0.174
AR: 0.337
AR@0.5: 0.763
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/top_down/resnet/res101_aic_256x192-79b35445_20200826.pth
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/2xmspn50_coco_256x192.py
Vendored
+165
View File
@@ -0,0 +1,165 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-3,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
log_config = dict(
interval=50, hooks=[
dict(type='TextLoggerHook'),
])
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet50',
backbone=dict(
type='MSPN',
unit_channels=256,
num_stages=2,
num_units=4,
num_blocks=[3, 4, 6, 3],
norm_cfg=dict(type='BN')),
keypoint_head=dict(
type='TopdownHeatmapMSMUHead',
out_shape=(64, 48),
unit_channels=256,
out_channels=channel_cfg['num_output_channels'],
num_stages=2,
num_units=4,
use_prm=False,
norm_cfg=dict(type='BN'),
loss_keypoint=([
dict(
type='JointsMSELoss', use_target_weight=True, loss_weight=0.25)
] * 3 + [
dict(
type='JointsOHKMMSELoss',
use_target_weight=True,
loss_weight=1.)
]) * 2),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='megvii',
shift_heatmap=False,
modulate_kernel=5))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
use_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='TopDownGenerateTarget',
kernel=[(15, 15), (11, 11), (9, 9), (7, 7)] + [(11, 11), (9, 9),
(7, 7), (5, 5)],
encoding='Megvii'),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=[
'img',
],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=32,
workers_per_gpu=4,
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/2xrsn50_coco_256x192.py
Vendored
+165
View File
@@ -0,0 +1,165 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-3,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
log_config = dict(
interval=50, hooks=[
dict(type='TextLoggerHook'),
])
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='RSN',
unit_channels=256,
num_stages=2,
num_units=4,
num_blocks=[3, 4, 6, 3],
num_steps=4,
norm_cfg=dict(type='BN')),
keypoint_head=dict(
type='TopdownHeatmapMSMUHead',
out_shape=(64, 48),
unit_channels=256,
out_channels=channel_cfg['num_output_channels'],
num_stages=2,
num_units=4,
use_prm=False,
norm_cfg=dict(type='BN'),
loss_keypoint=([
dict(
type='JointsMSELoss', use_target_weight=True, loss_weight=0.25)
] * 3 + [
dict(
type='JointsOHKMMSELoss',
use_target_weight=True,
loss_weight=1.)
]) * 2),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='megvii',
shift_heatmap=False,
modulate_kernel=5))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
use_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='TopDownGenerateTarget',
kernel=[(15, 15), (11, 11), (9, 9), (7, 7)] + [(11, 11), (9, 9),
(7, 7), (5, 5)],
encoding='Megvii'),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=[
'img',
],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=32,
workers_per_gpu=4,
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/3xmspn50_coco_256x192.py
Vendored
+165
View File
@@ -0,0 +1,165 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-3,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
log_config = dict(
interval=50, hooks=[
dict(type='TextLoggerHook'),
])
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet50',
backbone=dict(
type='MSPN',
unit_channels=256,
num_stages=3,
num_units=4,
num_blocks=[3, 4, 6, 3],
norm_cfg=dict(type='BN')),
keypoint_head=dict(
type='TopdownHeatmapMSMUHead',
out_shape=(64, 48),
unit_channels=256,
out_channels=channel_cfg['num_output_channels'],
num_stages=3,
num_units=4,
use_prm=False,
norm_cfg=dict(type='BN'),
loss_keypoint=([
dict(
type='JointsMSELoss', use_target_weight=True, loss_weight=0.25)
] * 3 + [
dict(
type='JointsOHKMMSELoss',
use_target_weight=True,
loss_weight=1.)
]) * 3),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='megvii',
shift_heatmap=False,
modulate_kernel=5))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
use_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='TopDownGenerateTarget',
kernel=[(15, 15), (11, 11), (9, 9), (7, 7)] * 2 + [(11, 11), (9, 9),
(7, 7), (5, 5)],
encoding='Megvii'),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=[
'img',
],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=32,
workers_per_gpu=4,
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/3xrsn50_coco_256x192.py
Vendored
+165
View File
@@ -0,0 +1,165 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-3,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
log_config = dict(
interval=50, hooks=[
dict(type='TextLoggerHook'),
])
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='RSN',
unit_channels=256,
num_stages=3,
num_units=4,
num_blocks=[3, 4, 6, 3],
num_steps=4,
norm_cfg=dict(type='BN')),
keypoint_head=dict(
type='TopdownHeatmapMSMUHead',
out_shape=(64, 48),
unit_channels=256,
out_channels=channel_cfg['num_output_channels'],
num_stages=3,
num_units=4,
use_prm=False,
norm_cfg=dict(type='BN'),
loss_keypoint=([
dict(
type='JointsMSELoss', use_target_weight=True, loss_weight=0.25)
] * 3 + [
dict(
type='JointsOHKMMSELoss',
use_target_weight=True,
loss_weight=1.)
]) * 3),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='megvii',
shift_heatmap=False,
modulate_kernel=5))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
use_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='TopDownGenerateTarget',
kernel=[(15, 15), (11, 11), (9, 9), (7, 7)] * 2 + [(11, 11), (9, 9),
(7, 7), (5, 5)],
encoding='Megvii'),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=[
'img',
],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=32,
workers_per_gpu=4,
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/4xmspn50_coco_256x192.py
Vendored
+165
View File
@@ -0,0 +1,165 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(
type='Adam',
lr=5e-3,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
log_config = dict(
interval=50, hooks=[
dict(type='TextLoggerHook'),
])
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained='torchvision://resnet50',
backbone=dict(
type='MSPN',
unit_channels=256,
num_stages=4,
num_units=4,
num_blocks=[3, 4, 6, 3],
norm_cfg=dict(type='BN')),
keypoint_head=dict(
type='TopdownHeatmapMSMUHead',
out_shape=(64, 48),
unit_channels=256,
out_channels=channel_cfg['num_output_channels'],
num_stages=4,
num_units=4,
use_prm=False,
norm_cfg=dict(type='BN'),
loss_keypoint=([
dict(
type='JointsMSELoss', use_target_weight=True, loss_weight=0.25)
] * 3 + [
dict(
type='JointsOHKMMSELoss',
use_target_weight=True,
loss_weight=1.)
]) * 4),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='megvii',
shift_heatmap=False,
modulate_kernel=5))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
use_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='TopDownGenerateTarget',
kernel=[(15, 15), (11, 11), (9, 9), (7, 7)] * 3 + [(11, 11), (9, 9),
(7, 7), (5, 5)],
encoding='Megvii'),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine'),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=[
'img',
],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=32,
workers_per_gpu=4,
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_base_coco_256x192.py
Vendored
+170
View File
@@ -0,0 +1,170 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(type='AdamW', lr=5e-4, betas=(0.9, 0.999), weight_decay=0.1,
constructor='LayerDecayOptimizerConstructor',
paramwise_cfg=dict(
num_layers=12,
layer_decay_rate=0.75,
custom_keys={
'bias': dict(decay_multi=0.),
'pos_embed': dict(decay_mult=0.),
'relative_position_bias_table': dict(decay_mult=0.),
'norm': dict(decay_mult=0.)
}
)
)
optimizer_config = dict(grad_clip=dict(max_norm=1., norm_type=2))
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
target_type = 'GaussianHeatmap'
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='ViT',
img_size=(256, 192),
patch_size=16,
embed_dim=768,
depth=12,
num_heads=12,
ratio=1,
use_checkpoint=False,
mlp_ratio=4,
qkv_bias=True,
drop_path_rate=0.3,
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=768,
num_deconv_layers=2,
num_deconv_filters=(256, 256),
num_deconv_kernels=(4, 4),
extra=dict(final_conv_kernel=1, ),
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=False,
target_type=target_type,
modulate_kernel=11,
use_udp=True))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine', use_udp=True),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='TopDownGenerateTarget',
sigma=2,
encoding='UDP',
target_type=target_type),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine', use_udp=True),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_base_simple_coco_256x192.py
Vendored
+171
View File
@@ -0,0 +1,171 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(type='AdamW', lr=5e-4, betas=(0.9, 0.999), weight_decay=0.1,
constructor='LayerDecayOptimizerConstructor',
paramwise_cfg=dict(
num_layers=12,
layer_decay_rate=0.75,
custom_keys={
'bias': dict(decay_multi=0.),
'pos_embed': dict(decay_mult=0.),
'relative_position_bias_table': dict(decay_mult=0.),
'norm': dict(decay_mult=0.)
}
)
)
optimizer_config = dict(grad_clip=dict(max_norm=1., norm_type=2))
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
target_type = 'GaussianHeatmap'
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='ViT',
img_size=(256, 192),
patch_size=16,
embed_dim=768,
depth=12,
num_heads=12,
ratio=1,
use_checkpoint=False,
mlp_ratio=4,
qkv_bias=True,
drop_path_rate=0.3,
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=768,
num_deconv_layers=0,
num_deconv_filters=[],
num_deconv_kernels=[],
upsample=4,
extra=dict(final_conv_kernel=3, ),
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=False,
target_type=target_type,
modulate_kernel=11,
use_udp=True))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine', use_udp=True),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='TopDownGenerateTarget',
sigma=2,
encoding='UDP',
target_type=target_type),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine', use_udp=True),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_huge_coco_256x192.py
Vendored
+170
View File
@@ -0,0 +1,170 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(type='AdamW', lr=5e-4, betas=(0.9, 0.999), weight_decay=0.1,
constructor='LayerDecayOptimizerConstructor',
paramwise_cfg=dict(
num_layers=32,
layer_decay_rate=0.85,
custom_keys={
'bias': dict(decay_multi=0.),
'pos_embed': dict(decay_mult=0.),
'relative_position_bias_table': dict(decay_mult=0.),
'norm': dict(decay_mult=0.)
}
)
)
optimizer_config = dict(grad_clip=dict(max_norm=1., norm_type=2))
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
target_type = 'GaussianHeatmap'
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='ViT',
img_size=(256, 192),
patch_size=16,
embed_dim=1280,
depth=32,
num_heads=16,
ratio=1,
use_checkpoint=False,
mlp_ratio=4,
qkv_bias=True,
drop_path_rate=0.55,
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=1280,
num_deconv_layers=2,
num_deconv_filters=(256, 256),
num_deconv_kernels=(4, 4),
extra=dict(final_conv_kernel=1, ),
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=False,
target_type=target_type,
modulate_kernel=11,
use_udp=True))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine', use_udp=True),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='TopDownGenerateTarget',
sigma=2,
encoding='UDP',
target_type=target_type),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine', use_udp=True),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_huge_simple_coco_256x192.py
Vendored
+171
View File
@@ -0,0 +1,171 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(type='AdamW', lr=5e-4, betas=(0.9, 0.999), weight_decay=0.1,
constructor='LayerDecayOptimizerConstructor',
paramwise_cfg=dict(
num_layers=32,
layer_decay_rate=0.85,
custom_keys={
'bias': dict(decay_multi=0.),
'pos_embed': dict(decay_mult=0.),
'relative_position_bias_table': dict(decay_mult=0.),
'norm': dict(decay_mult=0.)
}
)
)
optimizer_config = dict(grad_clip=dict(max_norm=1., norm_type=2))
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
target_type = 'GaussianHeatmap'
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='ViT',
img_size=(256, 192),
patch_size=16,
embed_dim=1280,
depth=32,
num_heads=16,
ratio=1,
use_checkpoint=False,
mlp_ratio=4,
qkv_bias=True,
drop_path_rate=0.55,
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=1280,
num_deconv_layers=0,
num_deconv_filters=[],
num_deconv_kernels=[],
upsample=4,
extra=dict(final_conv_kernel=3, ),
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=False,
target_type=target_type,
modulate_kernel=11,
use_udp=True))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine', use_udp=True),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='TopDownGenerateTarget',
sigma=2,
encoding='UDP',
target_type=target_type),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine', use_udp=True),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)
third_party/hamer/third-party/ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_large_coco_256x192.py
Vendored
+170
View File
@@ -0,0 +1,170 @@
_base_ = [
'../../../../_base_/default_runtime.py',
'../../../../_base_/datasets/coco.py'
]
evaluation = dict(interval=10, metric='mAP', save_best='AP')
optimizer = dict(type='AdamW', lr=5e-4, betas=(0.9, 0.999), weight_decay=0.1,
constructor='LayerDecayOptimizerConstructor',
paramwise_cfg=dict(
num_layers=24,
layer_decay_rate=0.8,
custom_keys={
'bias': dict(decay_multi=0.),
'pos_embed': dict(decay_mult=0.),
'relative_position_bias_table': dict(decay_mult=0.),
'norm': dict(decay_mult=0.)
}
)
)
optimizer_config = dict(grad_clip=dict(max_norm=1., norm_type=2))
# learning policy
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[170, 200])
total_epochs = 210
target_type = 'GaussianHeatmap'
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
# model settings
model = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='ViT',
img_size=(256, 192),
patch_size=16,
embed_dim=1024,
depth=24,
num_heads=16,
ratio=1,
use_checkpoint=False,
mlp_ratio=4,
qkv_bias=True,
drop_path_rate=0.5,
),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=1024,
num_deconv_layers=2,
num_deconv_filters=(256, 256),
num_deconv_kernels=(4, 4),
extra=dict(final_conv_kernel=1, ),
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=False,
target_type=target_type,
modulate_kernel=11,
use_udp=True))
data_cfg = dict(
image_size=[192, 256],
heatmap_size=[48, 64],
num_output_channels=channel_cfg['num_output_channels'],
num_joints=channel_cfg['dataset_joints'],
dataset_channel=channel_cfg['dataset_channel'],
inference_channel=channel_cfg['inference_channel'],
soft_nms=False,
nms_thr=1.0,
oks_thr=0.9,
vis_thr=0.2,
use_gt_bbox=False,
det_bbox_thr=0.0,
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownRandomFlip', flip_prob=0.5),
dict(
type='TopDownHalfBodyTransform',
num_joints_half_body=8,
prob_half_body=0.3),
dict(
type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
dict(type='TopDownAffine', use_udp=True),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='TopDownGenerateTarget',
sigma=2,
encoding='UDP',
target_type=target_type),
dict(
type='Collect',
keys=['img', 'target', 'target_weight'],
meta_keys=[
'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
'rotation', 'bbox_score', 'flip_pairs'
]),
]
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='TopDownAffine', use_udp=True),
dict(type='ToTensor'),
dict(
type='NormalizeTensor',
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
dict(
type='Collect',
keys=['img'],
meta_keys=[
'image_file', 'center', 'scale', 'rotation', 'bbox_score',
'flip_pairs'
]),
]
test_pipeline = val_pipeline
data_root = 'data/coco'
data = dict(
samples_per_gpu=64,
workers_per_gpu=4,
val_dataloader=dict(samples_per_gpu=32),
test_dataloader=dict(samples_per_gpu=32),
train=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_train2017.json',
img_prefix=f'{data_root}/train2017/',
data_cfg=data_cfg,
pipeline=train_pipeline,
dataset_info={{_base_.dataset_info}}),
val=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=val_pipeline,
dataset_info={{_base_.dataset_info}}),
test=dict(
type='TopDownCocoDataset',
ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
img_prefix=f'{data_root}/val2017/',
data_cfg=data_cfg,
pipeline=test_pipeline,
dataset_info={{_base_.dataset_info}}),
)

Some files were not shown because too many files have changed in this diff Show More