Skip to content

GitLab

"docs/vscode:/vscode.git/clone" did not exist on "b32ab0705edb8401d93b47e28dc5fc4d37b2e39b"
Commit ca8a762a authored by chenzk's avatar chenzk
Browse files

v1.0

parents
Pipeline #716 failed with stages
in 0 seconds
Hide whitespace changes
Inline Side-by-side
Showing with 2000 additions and 0 deletions
configs/body_2d_keypoint/rtmpose/humanart/rtmpose-s_8xb256-420e_humanart-256x192.py 0 → 100644
View file @ ca8a762a
_base_ = ['../../../_base_/default_runtime.py']
# runtime
max_epochs = 420
stage2_num_epochs = 30
base_lr = 4e-3
train_cfg = dict(max_epochs=max_epochs, val_interval=10)
randomness = dict(seed=21)
# optimizer
optim_wrapper = dict(
type='OptimWrapper',
optimizer=dict(type='AdamW', lr=base_lr, weight_decay=0.),
paramwise_cfg=dict(
norm_decay_mult=0, bias_decay_mult=0, bypass_duplicate=True))
# learning rate
param_scheduler = [
dict(
type='LinearLR',
start_factor=1.0e-5,
by_epoch=False,
begin=0,
end=1000),
dict(
# use cosine lr from 210 to 420 epoch
type='CosineAnnealingLR',
eta_min=base_lr * 0.05,
begin=max_epochs // 2,
end=max_epochs,
T_max=max_epochs // 2,
by_epoch=True,
convert_to_iter_based=True),
]
# automatically scaling LR based on the actual training batch size
auto_scale_lr = dict(base_batch_size=1024)
# codec settings
codec = dict(
type='SimCCLabel',
input_size=(192, 256),
sigma=(4.9, 5.66),
simcc_split_ratio=2.0,
normalize=False,
use_dark=False)
# model settings
model = dict(
type='TopdownPoseEstimator',
data_preprocessor=dict(
type='PoseDataPreprocessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
bgr_to_rgb=True),
backbone=dict(
_scope_='mmdet',
type='CSPNeXt',
arch='P5',
expand_ratio=0.5,
deepen_factor=0.33,
widen_factor=0.5,
out_indices=(4, ),
channel_attention=True,
norm_cfg=dict(type='SyncBN'),
act_cfg=dict(type='SiLU'),
init_cfg=dict(
type='Pretrained',
prefix='backbone.',
checkpoint='https://download.openmmlab.com/mmpose/v1/projects/'
'rtmpose/cspnext-s_udp-aic-coco_210e-256x192-92f5a029_20230130.pth' # noqa
)),
head=dict(
type='RTMCCHead',
in_channels=512,
out_channels=17,
input_size=codec['input_size'],
in_featuremap_size=(6, 8),
simcc_split_ratio=codec['simcc_split_ratio'],
final_layer_kernel_size=7,
gau_cfg=dict(
hidden_dims=256,
s=128,
expansion_factor=2,
dropout_rate=0.,
drop_path=0.,
act_fn='SiLU',
use_rel_bias=False,
pos_enc=False),
loss=dict(
type='KLDiscretLoss',
use_target_weight=True,
beta=10.,
label_softmax=True),
decoder=codec),
test_cfg=dict(flip_test=True))
# base dataset settings
dataset_type = 'HumanArtDataset'
data_mode = 'topdown'
data_root = 'data/'
backend_args = dict(backend='local')
# backend_args = dict(
# backend='petrel',
# path_mapping=dict({
# f'{data_root}': 's3://openmmlab/datasets/detection/coco/',
# f'{data_root}': 's3://openmmlab/datasets/detection/coco/'
# }))
# pipelines
train_pipeline = [
dict(type='LoadImage', backend_args=backend_args),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(
type='RandomBBoxTransform', scale_factor=[0.6, 1.4], rotate_factor=80),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='mmdet.YOLOXHSVRandomAug'),
dict(
type='Albumentation',
transforms=[
dict(type='Blur', p=0.1),
dict(type='MedianBlur', p=0.1),
dict(
type='CoarseDropout',
max_holes=1,
max_height=0.4,
max_width=0.4,
min_holes=1,
min_height=0.2,
min_width=0.2,
p=1.),
]),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
val_pipeline = [
dict(type='LoadImage', backend_args=backend_args),
dict(type='GetBBoxCenterScale'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='PackPoseInputs')
]
train_pipeline_stage2 = [
dict(type='LoadImage', backend_args=backend_args),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(
type='RandomBBoxTransform',
shift_factor=0.,
scale_factor=[0.75, 1.25],
rotate_factor=60),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='mmdet.YOLOXHSVRandomAug'),
dict(
type='Albumentation',
transforms=[
dict(type='Blur', p=0.1),
dict(type='MedianBlur', p=0.1),
dict(
type='CoarseDropout',
max_holes=1,
max_height=0.4,
max_width=0.4,
min_holes=1,
min_height=0.2,
min_width=0.2,
p=0.5),
]),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
# data loaders
train_dataloader = dict(
batch_size=256,
num_workers=10,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='HumanArt/annotations/training_humanart_coco.json',
data_prefix=dict(img=''),
pipeline=train_pipeline,
))
val_dataloader = dict(
batch_size=64,
num_workers=10,
persistent_workers=True,
drop_last=False,
sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='HumanArt/annotations/validation_humanart.json',
# bbox_file=f'{data_root}HumanArt/person_detection_results/'
# 'HumanArt_validation_detections_AP_H_56_person.json',
data_prefix=dict(img=''),
test_mode=True,
pipeline=val_pipeline,
))
test_dataloader = val_dataloader
# hooks
default_hooks = dict(
checkpoint=dict(save_best='coco/AP', rule='greater', max_keep_ckpts=1))
custom_hooks = [
dict(
type='EMAHook',
ema_type='ExpMomentumEMA',
momentum=0.0002,
update_buffers=True,
priority=49),
dict(
type='mmdet.PipelineSwitchHook',
switch_epoch=max_epochs - stage2_num_epochs,
switch_pipeline=train_pipeline_stage2)
]
# evaluators
val_evaluator = dict(
type='CocoMetric',
ann_file=data_root + 'HumanArt/annotations/validation_humanart.json')
test_evaluator = val_evaluator
configs/body_2d_keypoint/rtmpose/humanart/rtmpose-t_8xb256-420e_humanart-256x192.py 0 → 100644
View file @ ca8a762a
_base_ = ['../../../_base_/default_runtime.py']
# runtime
max_epochs = 420
stage2_num_epochs = 30
base_lr = 4e-3
train_cfg = dict(max_epochs=max_epochs, val_interval=10)
randomness = dict(seed=21)
# optimizer
optim_wrapper = dict(
type='OptimWrapper',
optimizer=dict(type='AdamW', lr=base_lr, weight_decay=0.),
paramwise_cfg=dict(
norm_decay_mult=0, bias_decay_mult=0, bypass_duplicate=True))
# learning rate
param_scheduler = [
dict(
type='LinearLR',
start_factor=1.0e-5,
by_epoch=False,
begin=0,
end=1000),
dict(
# use cosine lr from 210 to 420 epoch
type='CosineAnnealingLR',
eta_min=base_lr * 0.05,
begin=max_epochs // 2,
end=max_epochs,
T_max=max_epochs // 2,
by_epoch=True,
convert_to_iter_based=True),
]
# automatically scaling LR based on the actual training batch size
auto_scale_lr = dict(base_batch_size=1024)
# codec settings
codec = dict(
type='SimCCLabel',
input_size=(192, 256),
sigma=(4.9, 5.66),
simcc_split_ratio=2.0,
normalize=False,
use_dark=False)
# model settings
model = dict(
type='TopdownPoseEstimator',
data_preprocessor=dict(
type='PoseDataPreprocessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
bgr_to_rgb=True),
backbone=dict(
_scope_='mmdet',
type='CSPNeXt',
arch='P5',
expand_ratio=0.5,
deepen_factor=0.167,
widen_factor=0.375,
out_indices=(4, ),
channel_attention=True,
norm_cfg=dict(type='SyncBN'),
act_cfg=dict(type='SiLU'),
init_cfg=dict(
type='Pretrained',
prefix='backbone.',
checkpoint='https://download.openmmlab.com/mmpose/v1/projects/'
'rtmpose/cspnext-tiny_udp-aic-coco_210e-256x192-cbed682d_20230130.pth' # noqa
)),
head=dict(
type='RTMCCHead',
in_channels=384,
out_channels=17,
input_size=codec['input_size'],
in_featuremap_size=(6, 8),
simcc_split_ratio=codec['simcc_split_ratio'],
final_layer_kernel_size=7,
gau_cfg=dict(
hidden_dims=256,
s=128,
expansion_factor=2,
dropout_rate=0.,
drop_path=0.,
act_fn='SiLU',
use_rel_bias=False,
pos_enc=False),
loss=dict(
type='KLDiscretLoss',
use_target_weight=True,
beta=10.,
label_softmax=True),
decoder=codec),
test_cfg=dict(flip_test=True))
# base dataset settings
dataset_type = 'HumanArtDataset'
data_mode = 'topdown'
data_root = 'data/'
backend_args = dict(backend='local')
# backend_args = dict(
# backend='petrel',
# path_mapping=dict({
# f'{data_root}': 's3://openmmlab/datasets/detection/coco/',
# f'{data_root}': 's3://openmmlab/datasets/detection/coco/'
# }))
# pipelines
train_pipeline = [
dict(type='LoadImage', backend_args=backend_args),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(
type='RandomBBoxTransform', scale_factor=[0.6, 1.4], rotate_factor=80),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='mmdet.YOLOXHSVRandomAug'),
dict(
type='Albumentation',
transforms=[
dict(type='Blur', p=0.1),
dict(type='MedianBlur', p=0.1),
dict(
type='CoarseDropout',
max_holes=1,
max_height=0.4,
max_width=0.4,
min_holes=1,
min_height=0.2,
min_width=0.2,
p=1.),
]),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
val_pipeline = [
dict(type='LoadImage', backend_args=backend_args),
dict(type='GetBBoxCenterScale'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='PackPoseInputs')
]
train_pipeline_stage2 = [
dict(type='LoadImage', backend_args=backend_args),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(
type='RandomBBoxTransform',
shift_factor=0.,
scale_factor=[0.75, 1.25],
rotate_factor=60),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='mmdet.YOLOXHSVRandomAug'),
dict(
type='Albumentation',
transforms=[
dict(type='Blur', p=0.1),
dict(type='MedianBlur', p=0.1),
dict(
type='CoarseDropout',
max_holes=1,
max_height=0.4,
max_width=0.4,
min_holes=1,
min_height=0.2,
min_width=0.2,
p=0.5),
]),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
# data loaders
train_dataloader = dict(
batch_size=256,
num_workers=10,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='HumanArt/annotations/training_humanart_coco.json',
data_prefix=dict(img=''),
pipeline=train_pipeline,
))
val_dataloader = dict(
batch_size=64,
num_workers=10,
persistent_workers=True,
drop_last=False,
sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='HumanArt/annotations/validation_humanart.json',
# bbox_file=f'{data_root}HumanArt/person_detection_results/'
# 'HumanArt_validation_detections_AP_H_56_person.json',
data_prefix=dict(img=''),
test_mode=True,
pipeline=val_pipeline,
))
test_dataloader = val_dataloader
# hooks
default_hooks = dict(
checkpoint=dict(save_best='coco/AP', rule='greater', max_keep_ckpts=1))
custom_hooks = [
# Turn off EMA while training the tiny model
# dict(
# type='EMAHook',
# ema_type='ExpMomentumEMA',
# momentum=0.0002,
# update_buffers=True,
# priority=49),
dict(
type='mmdet.PipelineSwitchHook',
switch_epoch=max_epochs - stage2_num_epochs,
switch_pipeline=train_pipeline_stage2)
]
# evaluators
val_evaluator = dict(
type='CocoMetric',
ann_file=data_root + 'HumanArt/annotations/validation_humanart.json')
test_evaluator = val_evaluator
configs/body_2d_keypoint/rtmpose/humanart/rtmpose_humanart.md 0 → 100644
View file @ ca8a762a
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://link.springer.com/chapter/10.1007/978-3-030-58580-8_27">RTMPose (arXiv'2023)</a></summary>
```bibtex
@misc{https://doi.org/10.48550/arxiv.2303.07399,
doi = {10.48550/ARXIV.2303.07399},
url = {https://arxiv.org/abs/2303.07399},
author = {Jiang, Tao and Lu, Peng and Zhang, Li and Ma, Ningsheng and Han, Rui and Lyu, Chengqi and Li, Yining and Chen, Kai},
keywords = {Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {RTMPose: Real-Time Multi-Person Pose Estimation based on MMPose},
publisher = {arXiv},
year = {2023},
copyright = {Creative Commons Attribution 4.0 International}
}
```
</details>
<!-- [BACKBONE] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/2212.07784">RTMDet (arXiv'2022)</a></summary>
```bibtex
@misc{lyu2022rtmdet,
title={RTMDet: An Empirical Study of Designing Real-Time Object Detectors},
author={Chengqi Lyu and Wenwei Zhang and Haian Huang and Yue Zhou and Yudong Wang and Yanyi Liu and Shilong Zhang and Kai Chen},
year={2022},
eprint={2212.07784},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
</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>
<details>
<summary align="right"><a href="https://idea-research.github.io/HumanArt/">Human-Art (CVPR'2023)</a></summary>
```bibtex
@inproceedings{ju2023humanart,
title={Human-Art: A Versatile Human-Centric Dataset Bridging Natural and Artificial Scenes},
author={Ju, Xuan and Zeng, Ailing and Jianan, Wang and Qiang, Xu and Lei, Zhang},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),
year={2023}}
```
</details>
Results on Human-Art validation dataset with detector having human AP of 56.2 on Human-Art validation dataset
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :-------------------------------------------- | :--------: | :---: | :-------------: | :-------------: | :---: | :-------------: | :-------------------------------------------: | :-------------------------------------------: |
| [rtmpose-t-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-t_8xb256-420e_coco-256x192.py) | 256x192 | 0.161 | 0.283 | 0.154 | 0.221 | 0.373 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-tiny_simcc-coco_pt-aic-coco_420e-256x192-e613ba3f_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-tiny_simcc-coco_pt-aic-coco_420e-256x192-e613ba3f_20230127.json) |
| [rtmpose-t-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-t_8xb256-420e_humanart-256x192.py) | 256x192 | 0.249 | 0.395 | 0.256 | 0.323 | 0.485 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.json) |
| [rtmpose-s-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-s_8xb256-420e_coco-256x192.py) | 256x192 | 0.199 | 0.328 | 0.198 | 0.261 | 0.418 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_simcc-coco_pt-aic-coco_420e-256x192-8edcf0d7_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_simcc-coco_pt-aic-coco_420e-256x192-8edcf0d7_20230127.json) |
| [rtmpose-s-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-s_8xb256-420e_humanart-256x192.py) | 256x192 | 0.311 | 0.462 | 0.323 | 0.381 | 0.540 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.json) |
| [rtmpose-m-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-m_8xb256-420e_coco-256x192.py) | 256x192 | 0.239 | 0.372 | 0.243 | 0.302 | 0.455 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_simcc-coco_pt-aic-coco_420e-256x192-d8dd5ca4_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_simcc-coco_pt-aic-coco_420e-256x192-d8dd5ca4_20230127.json) |
| [rtmpose-m-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-m_8xb256-420e_humanart-256x192.py) | 256x192 | 0.355 | 0.503 | 0.377 | 0.417 | 0.568 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.json) |
| [rtmpose-l-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-l_8xb256-420e_coco-256x192.py) | 256x192 | 0.260 | 0.393 | 0.267 | 0.323 | 0.472 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_simcc-coco_pt-aic-coco_420e-256x192-1352a4d2_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_simcc-coco_pt-aic-coco_420e-256x192-1352a4d2_20230127.json) |
| [rtmpose-l-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-l_8xb256-420e_humanart-256x192.py) | 256x192 | 0.378 | 0.521 | 0.399 | 0.442 | 0.584 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.json) |
Results on Human-Art validation dataset with ground-truth bounding-box
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :-------------------------------------------- | :--------: | :---: | :-------------: | :-------------: | :---: | :-------------: | :-------------------------------------------: | :-------------------------------------------: |
| [rtmpose-t-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-t_8xb256-420e_coco-256x192.py) | 256x192 | 0.444 | 0.725 | 0.453 | 0.488 | 0.750 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-tiny_simcc-coco_pt-aic-coco_420e-256x192-e613ba3f_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-tiny_simcc-coco_pt-aic-coco_420e-256x192-e613ba3f_20230127.json) |
| [rtmpose-t-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-t_8xb256-420e_humanart-256x192.py) | 256x192 | 0.655 | 0.872 | 0.720 | 0.693 | 0.890 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.json) |
| [rtmpose-s-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-s_8xb256-420e_coco-256x192.py) | 256x192 | 0.480 | 0.739 | 0.498 | 0.521 | 0.763 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_simcc-coco_pt-aic-coco_420e-256x192-8edcf0d7_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_simcc-coco_pt-aic-coco_420e-256x192-8edcf0d7_20230127.json) |
| [rtmpose-s-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-s_8xb256-420e_humanart-256x192.py) | 256x192 | 0.698 | 0.893 | 0.768 | 0.732 | 0.903 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.json) |
| [rtmpose-m-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-m_8xb256-420e_coco-256x192.py) | 256x192 | 0.532 | 0.765 | 0.563 | 0.571 | 0.789 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_simcc-coco_pt-aic-coco_420e-256x192-d8dd5ca4_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_simcc-coco_pt-aic-coco_420e-256x192-d8dd5ca4_20230127.json) |
| [rtmpose-m-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-m_8xb256-420e_humanart-256x192.py) | 256x192 | 0.728 | 0.895 | 0.791 | 0.759 | 0.906 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.json) |
| [rtmpose-l-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-l_8xb256-420e_coco-256x192.py) | 256x192 | 0.564 | 0.789 | 0.602 | 0.599 | 0.808 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_simcc-coco_pt-aic-coco_420e-256x192-1352a4d2_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_simcc-coco_pt-aic-coco_420e-256x192-1352a4d2_20230127.json) |
| [rtmpose-l-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-l_8xb256-420e_humanart-256x192.py) | 256x192 | 0.753 | 0.905 | 0.812 | 0.783 | 0.915 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.json) |
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 |
| :-------------------------------------------- | :--------: | :---: | :-------------: | :-------------: | :---: | :-------------: | :-------------------------------------------: | :-------------------------------------------: |
| [rtmpose-t-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-t_8xb256-420e_coco-256x192.py) | 256x192 | 0.682 | 0.883 | 0.759 | 0.736 | 0.920 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-tiny_simcc-coco_pt-aic-coco_420e-256x192-e613ba3f_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-tiny_simcc-coco_pt-aic-coco_420e-256x192-e613ba3f_20230127.json) |
| [rtmpose-t-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-t_8xb256-420e_humanart-256x192.py) | 256x192 | 0.665 | 0.875 | 0.739 | 0.721 | 0.916 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.json) |
| [rtmpose-s-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-s_8xb256-420e_coco-256x192.py) | 256x192 | 0.716 | 0.892 | 0.789 | 0.768 | 0.929 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_simcc-coco_pt-aic-coco_420e-256x192-8edcf0d7_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_simcc-coco_pt-aic-coco_420e-256x192-8edcf0d7_20230127.json) |
| [rtmpose-s-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-s_8xb256-420e_humanart-256x192.py) | 256x192 | 0.706 | 0.888 | 0.780 | 0.759 | 0.928 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.json) |
| [rtmpose-m-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-m_8xb256-420e_coco-256x192.py) | 256x192 | 0.746 | 0.899 | 0.817 | 0.795 | 0.935 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_simcc-coco_pt-aic-coco_420e-256x192-d8dd5ca4_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_simcc-coco_pt-aic-coco_420e-256x192-d8dd5ca4_20230127.json) |
| [rtmpose-m-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-m_8xb256-420e_humanart-256x192.py) | 256x192 | 0.725 | 0.892 | 0.795 | 0.775 | 0.929 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.json) |
| [rtmpose-l-coco](/configs/body_2d_keypoint/rtmpose/coco/rtmpose-l_8xb256-420e_coco-256x192.py) | 256x192 | 0.758 | 0.906 | 0.826 | 0.806 | 0.942 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_simcc-coco_pt-aic-coco_420e-256x192-1352a4d2_20230127.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_simcc-coco_pt-aic-coco_420e-256x192-1352a4d2_20230127.json) |
| [rtmpose-l-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-l_8xb256-420e_humanart-256x192.py) | 256x192 | 0.748 | 0.901 | 0.816 | 0.796 | 0.938 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.json) |
Results on COCO val2017 with ground-truth bounding box
| Arch | Input Size | AP | AP<sup>50</sup> | AP<sup>75</sup> | AR | AR<sup>50</sup> | ckpt | log |
| :-------------------------------------------- | :--------: | :---: | :-------------: | :-------------: | :---: | :-------------: | :-------------------------------------------: | :-------------------------------------------: |
| [rtmpose-t-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-t_8xb256-420e_humanart-256x192.py) | 256x192 | 0.679 | 0.895 | 0.755 | 0.710 | 0.907 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.json) |
| [rtmpose-s-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-s_8xb256-420e_humanart-256x192.py) | 256x192 | 0.725 | 0.916 | 0.798 | 0.753 | 0.925 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.json) |
| [rtmpose-m-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-m_8xb256-420e_humanart-256x192.py) | 256x192 | 0.744 | 0.916 | 0.818 | 0.770 | 0.930 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.json) |
| [rtmpose-l-humanart-coco](/configs/body_2d_keypoint/rtmpose/humanart/rtmpose-l_8xb256-420e_humanart-256x192.py) | 256x192 | 0.770 | 0.927 | 0.840 | 0.794 | 0.939 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.json) |
configs/body_2d_keypoint/rtmpose/humanart/rtmpose_humanart.yml 0 → 100644
View file @ ca8a762a
Collections:
- Name: RTMPose
Paper:
Title: "RTMPose: Real-Time Multi-Person Pose Estimation based on MMPose"
URL: https://arxiv.org/abs/2303.07399
README: https://github.com/open-mmlab/mmpose/blob/main/projects/rtmpose/README.md
Models:
- Config: configs/body_2d_keypoint/rtmpose/humanart/rtmpose-l_8xb256-420e_humanart-256x192.py
In Collection: RTMPose
Metadata:
Architecture: &id001
- RTMPose
Training Data: &id002
- COCO
- Human-Art
Name: rtmpose-l_8xb256-420e_humanart-256x192
Results:
- Dataset: COCO
Metrics:
AP: 0.748
AP@0.5: 0.901
AP@0.75: 0.816
AR: 0.796
AR@0.5: 0.938
Task: Body 2D Keypoint
- Dataset: Human-Art
Metrics:
AP: 0.378
AP@0.5: 0.521
AP@0.75: 0.399
AR: 0.442
AR@0.5: 0.584
Task: Body 2D Keypoint
- Dataset: Human-Art(GT)
Metrics:
AP: 0.753
AP@0.5: 0.905
AP@0.75: 0.812
AR: 0.783
AR@0.5: 0.915
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-l_8xb256-420e_humanart-256x192-389f2cb0_20230611.pth
- Config: configs/body_2d_keypoint/rtmpose/humanart/rtmpose-m_8xb256-420e_humanart-256x192.py
In Collection: RTMPose
Metadata:
Architecture: *id001
Training Data: *id002
Name: rtmpose-m_8xb256-420e_humanart-256x192
Results:
- Dataset: COCO
Metrics:
AP: 0.725
AP@0.5: 0.892
AP@0.75: 0.795
AR: 0.775
AR@0.5: 0.929
Task: Body 2D Keypoint
- Dataset: Human-Art
Metrics:
AP: 0.355
AP@0.5: 0.503
AP@0.75: 0.377
AR: 0.417
AR@0.5: 0.568
Task: Body 2D Keypoint
- Dataset: Human-Art(GT)
Metrics:
AP: 0.728
AP@0.5: 0.895
AP@0.75: 0.791
AR: 0.759
AR@0.5: 0.906
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_8xb256-420e_humanart-256x192-8430627b_20230611.pth
- Config: configs/body_2d_keypoint/rtmpose/humanart/rtmpose-s_8xb256-420e_humanart-256x192.py
In Collection: RTMPose
Metadata:
Architecture: *id001
Training Data: *id002
Name: rtmpose-s_8xb256-420e_humanart-256x192
Results:
- Dataset: COCO
Metrics:
AP: 0.706
AP@0.5: 0.888
AP@0.75: 0.780
AR: 0.759
AR@0.5: 0.928
Task: Body 2D Keypoint
- Dataset: Human-Art
Metrics:
AP: 0.311
AP@0.5: 0.462
AP@0.75: 0.323
AR: 0.381
AR@0.5: 0.540
Task: Body 2D Keypoint
- Dataset: Human-Art(GT)
Metrics:
AP: 0.698
AP@0.5: 0.893
AP@0.75: 0.768
AR: 0.732
AR@0.5: 0.903
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-s_8xb256-420e_humanart-256x192-5a3ac943_20230611.pth
- Config: configs/body_2d_keypoint/rtmpose/humanart/rtmpose-t_8xb256-420e_humanart-256x192.py
In Collection: RTMPose
Metadata:
Architecture: *id001
Training Data: *id002
Name: rtmpose-t_8xb256-420e_humanart-256x192
Results:
- Dataset: COCO
Metrics:
AP: 0.665
AP@0.5: 0.875
AP@0.75: 0.739
AR: 0.721
AR@0.5: 0.916
Task: Body 2D Keypoint
- Dataset: Human-Art
Metrics:
AP: 0.249
AP@0.5: 0.395
AP@0.75: 0.256
AR: 0.323
AR@0.5: 0.485
Task: Body 2D Keypoint
- Dataset: Human-Art(GT)
Metrics:
AP: 0.655
AP@0.5: 0.872
AP@0.75: 0.720
AR: 0.693
AR@0.5: 0.890
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-t_8xb256-420e_humanart-256x192-60b68c98_20230612.pth
configs/body_2d_keypoint/rtmpose/mpii/rtmpose-m_8xb64-210e_mpii-256x256.py 0 → 100644
View file @ ca8a762a
_base_ = ['../../../_base_/default_runtime.py']
# runtime
max_epochs = 210
stage2_num_epochs = 30
base_lr = 4e-3
train_cfg = dict(max_epochs=max_epochs, val_interval=10)
# optimizer
optim_wrapper = dict(
type='OptimWrapper',
optimizer=dict(type='AdamW', lr=base_lr, weight_decay=0.05),
paramwise_cfg=dict(
norm_decay_mult=0, bias_decay_mult=0, bypass_duplicate=True))
# learning policy
param_scheduler = [
dict(
type='LinearLR',
start_factor=1.0e-5,
by_epoch=False,
begin=0,
end=1000),
dict(
# use cosine lr from 210 to 420 epoch
type='CosineAnnealingLR',
eta_min=base_lr * 0.05,
begin=max_epochs // 2,
end=max_epochs,
T_max=max_epochs // 2,
by_epoch=True,
convert_to_iter_based=True),
]
# automatically scaling LR based on the actual training batch size
auto_scale_lr = dict(base_batch_size=1024)
# codec settings
codec = dict(
type='SimCCLabel',
input_size=(256, 256),
sigma=(5.66, 5.66),
simcc_split_ratio=2.0,
normalize=False,
use_dark=False)
# model settings
model = dict(
type='TopdownPoseEstimator',
data_preprocessor=dict(
type='PoseDataPreprocessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
bgr_to_rgb=True),
backbone=dict(
_scope_='mmdet',
type='CSPNeXt',
arch='P5',
expand_ratio=0.5,
deepen_factor=0.67,
widen_factor=0.75,
out_indices=(4, ),
channel_attention=True,
norm_cfg=dict(type='SyncBN'),
act_cfg=dict(type='SiLU'),
init_cfg=dict(
type='Pretrained',
prefix='backbone.',
checkpoint='https://download.openmmlab.com/mmpose/v1/projects/'
'rtmposev1/cspnext-m_udp-aic-coco_210e-256x192-f2f7d6f6_20230130.pth' # noqa
)),
head=dict(
type='RTMCCHead',
in_channels=768,
out_channels=16,
input_size=codec['input_size'],
in_featuremap_size=tuple([s // 32 for s in codec['input_size']]),
simcc_split_ratio=codec['simcc_split_ratio'],
final_layer_kernel_size=7,
gau_cfg=dict(
hidden_dims=256,
s=128,
expansion_factor=2,
dropout_rate=0.,
drop_path=0.,
act_fn='SiLU',
use_rel_bias=False,
pos_enc=False),
loss=dict(
type='KLDiscretLoss',
use_target_weight=True,
beta=10.,
label_softmax=True),
decoder=codec),
test_cfg=dict(flip_test=True))
# base dataset settings
dataset_type = 'MpiiDataset'
data_mode = 'topdown'
data_root = 'data/mpii/'
backend_args = dict(backend='local')
# backend_args = dict(
# backend='petrel',
# path_mapping=dict({
# f'{data_root}': 's3://openmmlab/datasets/pose/MPI/',
# f'{data_root}': 's3://openmmlab/datasets/pose/MPI/'
# }))
# pipelines
train_pipeline = [
dict(type='LoadImage', backend_args=backend_args),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(
type='RandomBBoxTransform', scale_factor=[0.6, 1.4], rotate_factor=80),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='mmdet.YOLOXHSVRandomAug'),
dict(
type='Albumentation',
transforms=[
dict(type='Blur', p=0.1),
dict(type='MedianBlur', p=0.1),
dict(
type='CoarseDropout',
max_holes=1,
max_height=0.4,
max_width=0.4,
min_holes=1,
min_height=0.2,
min_width=0.2,
p=1.),
]),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
val_pipeline = [
dict(type='LoadImage', backend_args=backend_args),
dict(type='GetBBoxCenterScale'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='PackPoseInputs')
]
train_pipeline_stage2 = [
dict(type='LoadImage', backend_args=backend_args),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(
type='RandomBBoxTransform',
shift_factor=0.,
scale_factor=[0.75, 1.25],
rotate_factor=60),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='mmdet.YOLOXHSVRandomAug'),
dict(
type='Albumentation',
transforms=[
dict(type='Blur', p=0.1),
dict(type='MedianBlur', p=0.1),
dict(
type='CoarseDropout',
max_holes=1,
max_height=0.4,
max_width=0.4,
min_holes=1,
min_height=0.2,
min_width=0.2,
p=0.5),
]),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
# data loaders
train_dataloader = dict(
batch_size=64,
num_workers=10,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/mpii_train.json',
data_prefix=dict(img='images/'),
pipeline=train_pipeline,
))
val_dataloader = dict(
batch_size=32,
num_workers=10,
persistent_workers=True,
drop_last=False,
sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/mpii_val.json',
headbox_file=f'{data_root}/annotations/mpii_gt_val.mat',
data_prefix=dict(img='images/'),
test_mode=True,
pipeline=val_pipeline,
))
test_dataloader = val_dataloader
# hooks
default_hooks = dict(
checkpoint=dict(save_best='PCK', rule='greater', max_keep_ckpts=1))
custom_hooks = [
dict(
type='EMAHook',
ema_type='ExpMomentumEMA',
momentum=0.0002,
update_buffers=True,
priority=49),
dict(
type='mmdet.PipelineSwitchHook',
switch_epoch=max_epochs - stage2_num_epochs,
switch_pipeline=train_pipeline_stage2)
]
# evaluators
val_evaluator = dict(type='MpiiPCKAccuracy')
test_evaluator = val_evaluator
configs/body_2d_keypoint/rtmpose/mpii/rtmpose_mpii.md 0 → 100644
View file @ ca8a762a
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://link.springer.com/chapter/10.1007/978-3-030-58580-8_27">RTMPose (arXiv'2023)</a></summary>
```bibtex
@misc{https://doi.org/10.48550/arxiv.2303.07399,
doi = {10.48550/ARXIV.2303.07399},
url = {https://arxiv.org/abs/2303.07399},
author = {Jiang, Tao and Lu, Peng and Zhang, Li and Ma, Ningsheng and Han, Rui and Lyu, Chengqi and Li, Yining and Chen, Kai},
keywords = {Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {RTMPose: Real-Time Multi-Person Pose Estimation based on MMPose},
publisher = {arXiv},
year = {2023},
copyright = {Creative Commons Attribution 4.0 International}
}
```
</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 / w. flip | Mean@0.1 | ckpt | log |
| :------------------------------------------------------- | :--------: | :------------: | :------: | :------------------------------------------------------: | :------------------------------------------------------: |
| [rtmpose-m](/configs/body_2d_keypoint/rtmpose/mpii/rtmpose-m_8xb64-210e_mpii-256x256.py) | 256x256 | 0.907 | 0.348 | [ckpt](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_simcc-mpii_pt-aic-coco_210e-256x256-ec4dbec8_20230206.pth) | [log](https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_simcc-mpii_pt-aic-coco_210e-256x256-ec4dbec8_20230206.json) |
configs/body_2d_keypoint/rtmpose/mpii/rtmpose_mpii.yml 0 → 100644
View file @ ca8a762a
Models:
- Config: configs/body_2d_keypoint/rtmpose/mpii/rtmpose-m_8xb64-210e_mpii-256x256.py
In Collection: RTMPose
Metadata:
Architecture:
- RTMPose
Training Data: MPII
Name: rtmpose-m_8xb64-210e_mpii-256x256
Results:
- Dataset: MPII
Metrics:
Mean: 0.907
Mean@0.1: 0.348
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/rtmpose-m_simcc-mpii_pt-aic-coco_210e-256x256-ec4dbec8_20230206.pth
configs/body_2d_keypoint/simcc/README.md 0 → 100644
View file @ ca8a762a
# Top-down SimCC-based pose estimation
Top-down methods divide the task into two stages: object detection, followed by single-object pose estimation given object bounding boxes. At the 2nd stage, SimCC based methods reformulate human pose estimation as two classification tasks for horizontal and vertical coordinates, and uniformly divide each pixel into several bins, thus obtain the keypoint coordinates given the features extracted from the bounding box area, following the paradigm introduced in [SimCC: a Simple Coordinate Classification Perspective for Human Pose Estimation](https://arxiv.org/abs/2107.03332).
<div align=center>
<img src="https://user-images.githubusercontent.com/13503330/189811385-6395d118-055b-4bad-89e8-f84ffa2c2aa6.png">
</div>
## Results and Models
### COCO Dataset
Results on COCO val2017 with detector having human AP of 56.4 on COCO val2017 dataset
| Model | Input Size | AP | AR | Details and Download |
| :---------------------------: | :--------: | :---: | :---: | :-----------------------------------------------: |
| ResNet-50+SimCC | 384x288 | 0.735 | 0.790 | [resnet_coco.md](./coco/resnet_coco.md) |
| ResNet-50+SimCC | 256x192 | 0.721 | 0.781 | [resnet_coco.md](./coco/resnet_coco.md) |
| S-ViPNAS-MobileNet-V3+SimCC | 256x192 | 0.695 | 0.755 | [vipnas_coco.md](./coco/vipnas_coco.md) |
| MobileNet-V2+SimCC(wo/deconv) | 256x192 | 0.620 | 0.678 | [mobilenetv2_coco.md](./coco/mobilenetv2_coco.md) |
configs/body_2d_keypoint/simcc/coco/mobilenetv2_coco.md 0 → 100644
View file @ ca8a762a
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/2107.03332">SimCC (ECCV'2022)</a></summary>
```bibtex
@misc{https://doi.org/10.48550/arxiv.2107.03332,
title={SimCC: a Simple Coordinate Classification Perspective for Human Pose Estimation},
author={Li, Yanjie and Yang, Sen and Liu, Peidong and Zhang, Shoukui and Wang, Yunxiao and Wang, Zhicheng and Yang, Wankou and Xia, Shu-Tao},
year={2021}
}
```
</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 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 |
| :-------------------------------------------- | :--------: | :---: | :-------------: | :-------------: | :---: | :-------------: | :-------------------------------------------: | :-------------------------------------------: |
| [simcc_mobilenetv2_wo_deconv](/configs/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192.py) | 256x192 | 0.620 | 0.855 | 0.697 | 0.678 | 0.902 | [ckpt](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192-4b0703bb_20221010.pth) | [log](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192-4b0703bb_20221010.log.json) |
configs/body_2d_keypoint/simcc/coco/mobilenetv2_coco.yml 0 → 100644
View file @ ca8a762a
Models:
- Config: configs/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192.py
In Collection: SimCC
Metadata:
Architecture: &id001
- SimCC
- MobilenetV2
Training Data: COCO
Name: simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192
Results:
- Dataset: COCO
Metrics:
AP: 0.62
AP@0.5: 0.855
AP@0.75: 0.697
AR: 0.678
AR@0.5: 0.902
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192-4b0703bb_20221010.pth
configs/body_2d_keypoint/simcc/coco/resnet_coco.md 0 → 100644
View file @ ca8a762a
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/2107.03332">SimCC (ECCV'2022)</a></summary>
```bibtex
@misc{https://doi.org/10.48550/arxiv.2107.03332,
title={SimCC: a Simple Coordinate Classification Perspective for Human Pose Estimation},
author={Li, Yanjie and Yang, Sen and Liu, Peidong and Zhang, Shoukui and Wang, Yunxiao and Wang, Zhicheng and Yang, Wankou and Xia, Shu-Tao},
year={2021}
}
```
</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 |
| :-------------------------------------------- | :--------: | :---: | :-------------: | :-------------: | :---: | :-------------: | :-------------------------------------------: | :-------------------------------------------: |
| [simcc_resnet_50](/configs/body_2d_keypoint/simcc/coco/simcc_res50_8xb64-210e_coco-256x192.py) | 256x192 | 0.721 | 0.897 | 0.798 | 0.781 | 0.937 | [ckpt](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_res50_8xb64-210e_coco-256x192-8e0f5b59_20220919.pth) | [log](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_res50_8xb64-210e_coco-256x192-8e0f5b59_20220919.log.json) |
| [simcc_resnet_50](/configs/body_2d_keypoint/simcc/coco/simcc_res50_8xb32-140e_coco-384x288.py) | 384x288 | 0.735 | 0.899 | 0.800 | 0.790 | 0.939 | [ckpt](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_res50_8xb32-140e_coco-384x288-45c3ba34_20220913.pth) | [log](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_res50_8xb32-140e_coco-384x288-45c3ba34_20220913.log.json) |
configs/body_2d_keypoint/simcc/coco/resnet_coco.yml 0 → 100644
View file @ ca8a762a
Collections:
- Name: SimCC
Paper:
Title: A Simple Coordinate Classification Perspective for Human Pose Estimation
URL: https://arxiv.org/abs/2107.03332
README: https://github.com/open-mmlab/mmpose/blob/main/docs/src/papers/algorithms/simcc.md
Models:
- Config: configs/body_2d_keypoint/simcc/coco/simcc_res50_8xb64-210e_coco-256x192.py
In Collection: SimCC
Metadata:
Architecture: &id001
- SimCC
- ResNet
Training Data: COCO
Name: simcc_res50_8xb64-210e_coco-256x192
Results:
- Dataset: COCO
Metrics:
AP: 0.721
AP@0.5: 0.900
AP@0.75: 0.798
AR: 0.781
AR@0.5: 0.937
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_res50_8xb64-210e_coco-256x192-8e0f5b59_20220919.pth
- Config: configs/body_2d_keypoint/simcc/coco/simcc_res50_8xb32-140e_coco-384x288.py
In Collection: SimCC
Metadata:
Architecture: *id001
Training Data: COCO
Name: simcc_res50_8xb32-140e_coco-384x288
Results:
- Dataset: COCO
Metrics:
AP: 0.735
AP@0.5: 0.899
AP@0.75: 0.800
AR: 0.790
AR@0.5: 0.939
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_res50_8xb32-140e_coco-384x288-45c3ba34_20220913.pth
configs/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192.py 0 → 100644
View file @ ca8a762a
_base_ = ['../../../_base_/default_runtime.py']
# runtime
train_cfg = dict(max_epochs=210, val_interval=10)
# optimizer
optim_wrapper = dict(optimizer=dict(
type='Adam',
lr=5e-4,
))
# learning policy
param_scheduler = [
dict(
type='LinearLR', begin=0, end=500, start_factor=0.001,
by_epoch=False), # warm-up
dict(
type='MultiStepLR',
begin=0,
end=train_cfg['max_epochs'],
milestones=[170, 200],
gamma=0.1,
by_epoch=True)
]
# automatically scaling LR based on the actual training batch size
auto_scale_lr = dict(base_batch_size=512)
# codec settings
codec = dict(
type='SimCCLabel', input_size=(192, 256), sigma=6.0, simcc_split_ratio=2.0)
# model settings
model = dict(
type='TopdownPoseEstimator',
data_preprocessor=dict(
type='PoseDataPreprocessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
bgr_to_rgb=True),
backbone=dict(
type='MobileNetV2',
widen_factor=1.,
out_indices=(7, ),
init_cfg=dict(
type='Pretrained',
checkpoint='mmcls://mobilenet_v2',
)),
head=dict(
type='SimCCHead',
in_channels=1280,
out_channels=17,
input_size=codec['input_size'],
in_featuremap_size=tuple([s // 32 for s in codec['input_size']]),
simcc_split_ratio=codec['simcc_split_ratio'],
deconv_out_channels=None,
loss=dict(type='KLDiscretLoss', use_target_weight=True),
decoder=codec),
test_cfg=dict(flip_test=True, ))
# base dataset settings
dataset_type = 'CocoDataset'
data_mode = 'topdown'
data_root = 'data/coco/'
# pipelines
train_pipeline = [
dict(type='LoadImage'),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(type='RandomBBoxTransform'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
val_pipeline = [
dict(type='LoadImage'),
dict(type='GetBBoxCenterScale'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='PackPoseInputs')
]
# data loaders
train_dataloader = dict(
batch_size=64,
num_workers=2,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_train2017.json',
data_prefix=dict(img='train2017/'),
pipeline=train_pipeline,
))
val_dataloader = dict(
batch_size=32,
num_workers=2,
persistent_workers=True,
drop_last=False,
sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_val2017.json',
bbox_file=f'{data_root}person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
data_prefix=dict(img='val2017/'),
test_mode=True,
pipeline=val_pipeline,
))
test_dataloader = val_dataloader
# hooks
default_hooks = dict(checkpoint=dict(save_best='coco/AP', rule='greater'))
# evaluators
val_evaluator = dict(
type='CocoMetric',
ann_file=data_root + 'annotations/person_keypoints_val2017.json')
test_evaluator = val_evaluator
configs/body_2d_keypoint/simcc/coco/simcc_res50_8xb32-140e_coco-384x288.py 0 → 100644
View file @ ca8a762a
_base_ = ['../../../_base_/default_runtime.py']
# runtime
train_cfg = dict(max_epochs=140, val_interval=10)
# optimizer
optim_wrapper = dict(optimizer=dict(
type='Adam',
lr=1e-3,
))
# learning policy
param_scheduler = [
dict(
type='LinearLR', begin=0, end=500, start_factor=0.001,
by_epoch=False), # warm-up
dict(
type='MultiStepLR',
begin=0,
end=train_cfg['max_epochs'],
milestones=[90, 120],
gamma=0.1,
by_epoch=True)
]
# automatically scaling LR based on the actual training batch size
auto_scale_lr = dict(base_batch_size=512)
# codec settings
codec = dict(
type='SimCCLabel', input_size=(288, 384), sigma=6.0, simcc_split_ratio=2.0)
# model settings
model = dict(
type='TopdownPoseEstimator',
data_preprocessor=dict(
type='PoseDataPreprocessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
bgr_to_rgb=True),
backbone=dict(
type='ResNet',
depth=50,
init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50'),
),
head=dict(
type='SimCCHead',
in_channels=2048,
out_channels=17,
input_size=codec['input_size'],
in_featuremap_size=tuple([s // 32 for s in codec['input_size']]),
simcc_split_ratio=codec['simcc_split_ratio'],
loss=dict(type='KLDiscretLoss', use_target_weight=True),
decoder=codec),
test_cfg=dict(flip_test=True))
# base dataset settings
dataset_type = 'CocoDataset'
data_mode = 'topdown'
data_root = 'data/coco/'
# pipelines
train_pipeline = [
dict(type='LoadImage'),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(type='RandomBBoxTransform'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
test_pipeline = [
dict(type='LoadImage'),
dict(type='GetBBoxCenterScale'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='PackPoseInputs')
]
# data loaders
train_dataloader = dict(
batch_size=32,
num_workers=2,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_train2017.json',
data_prefix=dict(img='train2017/'),
pipeline=train_pipeline,
))
val_dataloader = dict(
batch_size=32,
num_workers=2,
persistent_workers=True,
drop_last=False,
sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_val2017.json',
bbox_file=f'{data_root}person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
data_prefix=dict(img='val2017/'),
test_mode=True,
pipeline=test_pipeline,
))
test_dataloader = val_dataloader
# hooks
default_hooks = dict(checkpoint=dict(save_best='coco/AP', rule='greater'))
# evaluators
val_evaluator = dict(
type='CocoMetric',
ann_file=data_root + 'annotations/person_keypoints_val2017.json')
test_evaluator = val_evaluator
configs/body_2d_keypoint/simcc/coco/simcc_res50_8xb64-210e_coco-256x192.py 0 → 100644
View file @ ca8a762a
_base_ = ['../../../_base_/default_runtime.py']
# runtime
train_cfg = dict(max_epochs=210, val_interval=10)
# optimizer
optim_wrapper = dict(optimizer=dict(
type='Adam',
lr=1e-3,
))
# learning policy
param_scheduler = [
dict(
type='LinearLR', begin=0, end=500, start_factor=0.001,
by_epoch=False), # warm-up
dict(type='MultiStepLR', milestones=[170, 200], gamma=0.1, by_epoch=True)
]
# automatically scaling LR based on the actual training batch size
auto_scale_lr = dict(base_batch_size=512)
# codec settings
codec = dict(
type='SimCCLabel', input_size=(192, 256), sigma=6.0, simcc_split_ratio=2.0)
# model settings
model = dict(
type='TopdownPoseEstimator',
data_preprocessor=dict(
type='PoseDataPreprocessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
bgr_to_rgb=True),
backbone=dict(
type='ResNet',
depth=50,
init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50'),
),
head=dict(
type='SimCCHead',
in_channels=2048,
out_channels=17,
input_size=codec['input_size'],
in_featuremap_size=tuple([s // 32 for s in codec['input_size']]),
simcc_split_ratio=codec['simcc_split_ratio'],
loss=dict(type='KLDiscretLoss', use_target_weight=True),
decoder=codec),
test_cfg=dict(flip_test=True))
# base dataset settings
dataset_type = 'CocoDataset'
data_mode = 'topdown'
data_root = 'data/coco/'
# pipelines
train_pipeline = [
dict(type='LoadImage'),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(type='RandomBBoxTransform'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
test_pipeline = [
dict(type='LoadImage'),
dict(type='GetBBoxCenterScale'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='PackPoseInputs')
]
# data loaders
train_dataloader = dict(
batch_size=64,
num_workers=2,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_train2017.json',
data_prefix=dict(img='train2017/'),
pipeline=train_pipeline,
))
val_dataloader = dict(
batch_size=32,
num_workers=2,
persistent_workers=True,
drop_last=False,
sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_val2017.json',
bbox_file=f'{data_root}person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
data_prefix=dict(img='val2017/'),
test_mode=True,
pipeline=test_pipeline,
))
test_dataloader = val_dataloader
# hooks
default_hooks = dict(checkpoint=dict(save_best='coco/AP', rule='greater'))
# evaluators
val_evaluator = dict(
type='CocoMetric',
ann_file=data_root + 'annotations/person_keypoints_val2017.json')
test_evaluator = val_evaluator
configs/body_2d_keypoint/simcc/coco/simcc_vipnas-mbv3_8xb64-210e_coco-256x192.py 0 → 100644
View file @ ca8a762a
_base_ = ['../../../_base_/default_runtime.py']
# runtime
train_cfg = dict(max_epochs=210, val_interval=10)
# optimizer
optim_wrapper = dict(optimizer=dict(
type='Adam',
lr=5e-4,
))
# learning policy
param_scheduler = [
dict(
type='LinearLR', begin=0, end=500, start_factor=0.001,
by_epoch=False), # warm-up
dict(
type='MultiStepLR',
begin=0,
end=train_cfg['max_epochs'],
milestones=[170, 200],
gamma=0.1,
by_epoch=True)
]
# automatically scaling LR based on the actual training batch size
auto_scale_lr = dict(base_batch_size=512)
# codec settings
codec = dict(
type='SimCCLabel', input_size=(192, 256), sigma=6.0, simcc_split_ratio=2.0)
# model settings
model = dict(
type='TopdownPoseEstimator',
data_preprocessor=dict(
type='PoseDataPreprocessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
bgr_to_rgb=True),
backbone=dict(type='ViPNAS_MobileNetV3'),
head=dict(
type='SimCCHead',
in_channels=160,
out_channels=17,
input_size=codec['input_size'],
in_featuremap_size=tuple([s // 32 for s in codec['input_size']]),
simcc_split_ratio=codec['simcc_split_ratio'],
deconv_type='vipnas',
deconv_out_channels=(160, 160, 160),
deconv_num_groups=(160, 160, 160),
loss=dict(type='KLDiscretLoss', use_target_weight=True),
decoder=codec),
test_cfg=dict(flip_test=True, ))
# base dataset settings
dataset_type = 'CocoDataset'
data_mode = 'topdown'
data_root = 'data/coco/'
# pipelines
train_pipeline = [
dict(type='LoadImage'),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(type='RandomBBoxTransform'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
val_pipeline = [
dict(type='LoadImage'),
dict(type='GetBBoxCenterScale'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='PackPoseInputs')
]
# data loaders
train_dataloader = dict(
batch_size=64,
num_workers=2,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_train2017.json',
data_prefix=dict(img='train2017/'),
pipeline=train_pipeline,
))
val_dataloader = dict(
batch_size=32,
num_workers=2,
persistent_workers=True,
drop_last=False,
sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_val2017.json',
bbox_file=data_root + 'person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
data_prefix=dict(img='val2017/'),
test_mode=True,
pipeline=val_pipeline,
))
test_dataloader = val_dataloader
# hooks
default_hooks = dict(checkpoint=dict(save_best='coco/AP', rule='greater'))
# evaluators
val_evaluator = dict(
type='CocoMetric',
ann_file=data_root + 'annotations/person_keypoints_val2017.json')
test_evaluator = val_evaluator
configs/body_2d_keypoint/simcc/coco/vipnas_coco.md 0 → 100644
View file @ ca8a762a
<!-- [ALGORITHM] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/2107.03332">SimCC (ECCV'2022)</a></summary>
```bibtex
@misc{https://doi.org/10.48550/arxiv.2107.03332,
title={SimCC: a Simple Coordinate Classification Perspective for Human Pose Estimation},
author={Li, Yanjie and Yang, Sen and Liu, Peidong and Zhang, Shoukui and Wang, Yunxiao and Wang, Zhicheng and Yang, Wankou and Xia, Shu-Tao},
year={2021}
}
```
</details>
<!-- [BACKBONE] -->
<details>
<summary align="right"><a href="https://arxiv.org/abs/2105.10154">ViPNAS (CVPR'2021)</a></summary>
```bibtex
@article{xu2021vipnas,
title={ViPNAS: Efficient Video Pose Estimation via Neural Architecture Search},
author={Xu, Lumin and Guan, Yingda and Jin, Sheng and Liu, Wentao and Qian, Chen and Luo, Ping and Ouyang, Wanli and Wang, Xiaogang},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
year={2021}
}
```
</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 |
| :-------------------------------------------- | :--------: | :---: | :-------------: | :-------------: | :---: | :-------------: | :-------------------------------------------: | :-------------------------------------------: |
| [simcc_S-ViPNAS-MobileNetV3](/configs/body_2d_keypoint/simcc/coco/simcc_vipnas-mbv3_8xb64-210e_coco-256x192.py) | 256x192 | 0.695 | 0.883 | 0.772 | 0.755 | 0.927 | [ckpt](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_vipnas-mbv3_8xb64-210e_coco-256x192-719f3489_20220922.pth) | [log](https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_vipnas-mbv3_8xb64-210e_coco-256x192-719f3489_20220922.log.json) |
configs/body_2d_keypoint/simcc/coco/vipnas_coco.yml 0 → 100644
View file @ ca8a762a
Models:
- Config: configs/body_2d_keypoint/simcc/coco/simcc_vipnas-mbv3_8xb64-210e_coco-256x192.py
In Collection: SimCC
Metadata:
Architecture: &id001
- SimCC
- ViPNAS
Training Data: COCO
Name: simcc_vipnas-mbv3_8xb64-210e_coco-256x192
Results:
- Dataset: COCO
Metrics:
AP: 0.695
AP@0.5: 0.883
AP@0.75: 0.772
AR: 0.755
AR@0.5: 0.927
Task: Body 2D Keypoint
Weights: https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/simcc/coco/simcc_vipnas-mbv3_8xb64-210e_coco-256x192-719f3489_20220922.pth
configs/body_2d_keypoint/simcc/mpii/simcc_res50_wo-deconv-8xb64-210e_mpii-256x256.py 0 → 100644
View file @ ca8a762a
_base_ = ['../../../_base_/default_runtime.py']
# runtime
train_cfg = dict(max_epochs=210, val_interval=10)
# optimizer
optim_wrapper = dict(optimizer=dict(
type='Adam',
lr=5e-4,
))
# learning policy
param_scheduler = [
dict(
type='LinearLR', begin=0, end=500, start_factor=0.001,
by_epoch=False), # warm-up
dict(
type='MultiStepLR',
begin=0,
end=train_cfg['max_epochs'],
milestones=[170, 200],
gamma=0.1,
by_epoch=True)
]
# automatically scaling LR based on the actual training batch size
auto_scale_lr = dict(base_batch_size=512)
# codec settings
codec = dict(
type='SimCCLabel', input_size=(256, 256), sigma=6.0, simcc_split_ratio=2.0)
# model settings
model = dict(
type='TopdownPoseEstimator',
data_preprocessor=dict(
type='PoseDataPreprocessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
bgr_to_rgb=True),
backbone=dict(
type='ResNet',
depth=50,
init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50'),
),
head=dict(
type='SimCCHead',
in_channels=2048,
out_channels=16,
input_size=codec['input_size'],
in_featuremap_size=tuple([s // 32 for s in codec['input_size']]),
simcc_split_ratio=codec['simcc_split_ratio'],
deconv_out_channels=None,
loss=dict(type='KLDiscretLoss', use_target_weight=True),
decoder=codec),
test_cfg=dict(
flip_test=True,
shift_coords=True,
))
# base dataset settings
dataset_type = 'MpiiDataset'
data_mode = 'topdown'
data_root = 'data/mpii/'
# pipelines
train_pipeline = [
dict(type='LoadImage'),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomBBoxTransform', shift_prob=0),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
val_pipeline = [
dict(type='LoadImage'),
dict(type='GetBBoxCenterScale'),
dict(type='TopdownAffine', input_size=codec['input_size']),
dict(type='PackPoseInputs')
]
# data loaders
train_dataloader = dict(
batch_size=64,
num_workers=2,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/mpii_train.json',
data_prefix=dict(img='images/'),
pipeline=train_pipeline,
))
val_dataloader = dict(
batch_size=32,
num_workers=2,
persistent_workers=True,
drop_last=False,
sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/mpii_val.json',
headbox_file=f'{data_root}/annotations/mpii_gt_val.mat',
data_prefix=dict(img='images/'),
test_mode=True,
pipeline=val_pipeline,
))
test_dataloader = val_dataloader
# hooks
default_hooks = dict(checkpoint=dict(save_best='PCK', rule='greater'))
# evaluators
val_evaluator = dict(type='MpiiPCKAccuracy')
test_evaluator = val_evaluator
configs/body_2d_keypoint/topdown_heatmap/README.md 0 → 100644
View file @ ca8a762a
# Top-down heatmap-based pose estimation
Top-down methods divide the task into two stages: object detection, followed by single-object pose estimation given object bounding boxes. Instead of estimating keypoint coordinates directly, the pose estimator will produce heatmaps which represent the likelihood of being a keypoint, following the paradigm introduced in [Simple Baselines for Human Pose Estimation and Tracking](http://openaccess.thecvf.com/content_ECCV_2018/html/Bin_Xiao_Simple_Baselines_for_ECCV_2018_paper.html).
<div align=center>
<img src="https://user-images.githubusercontent.com/15977946/146522977-5f355832-e9c1-442f-a34f-9d24fb0aefa8.png" height=400>
</div>
## Results and Models
### COCO Dataset
Results on COCO val2017 with detector having human AP of 56.4 on COCO val2017 dataset
| Model | Input Size | AP | AR | Details and Download |
| :-------------: | :--------: | :---: | :---: | :-------------------------------------------------: |
| ViTPose-h | 256x192 | 0.790 | 0.840 | [vitpose_coco.md](./coco/vitpose_coco.md) |
| HRNet-w48+UDP | 256x192 | 0.768 | 0.817 | [hrnet_udp_coco.md](./coco/hrnet_udp_coco.md) |
| MSPN 4-stg | 256x192 | 0.765 | 0.826 | [mspn_coco.md](./coco/mspn_coco.md) |
| HRNet-w48+Dark | 256x192 | 0.764 | 0.814 | [hrnet_dark_coco.md](./coco/hrnet_dark_coco.md) |
| HRNet-w48 | 256x192 | 0.756 | 0.809 | [hrnet_coco.md](./coco/hrnet_coco.md) |
| HRFormer-B | 256x192 | 0.754 | 0.807 | [hrformer_coco.md](./coco/hrformer_coco.md) |
| RSN-50-3x | 256x192 | 0.750 | 0.814 | [rsn_coco.md](./coco/rsn_coco.md) |
| CSPNeXt-l | 256x192 | 0.750 | 0.800 | [cspnext_udp_coco.md](./coco/cspnext_udp_coco.md) |
| HRNet-w32 | 256x192 | 0.749 | 0.804 | [hrnet_coco.md](./coco/hrnet_coco.md) |
| Swin-L | 256x192 | 0.743 | 0.798 | [swin_coco.md](./coco/swin_coco.md) |
| ViTPose-s | 256x192 | 0.739 | 0.792 | [vitpose_coco.md](./coco/vitpose_coco.md) |
| HRFormer-S | 256x192 | 0.738 | 0.793 | [hrformer_coco.md](./coco/hrformer_coco.md) |
| Swin-B | 256x192 | 0.737 | 0.794 | [swin_coco.md](./coco/swin_coco.md) |
| SEResNet-101 | 256x192 | 0.734 | 0.790 | [seresnet_coco.md](./coco/seresnet_coco.md) |
| SCNet-101 | 256x192 | 0.733 | 0.789 | [scnet_coco.md](./coco/scnet_coco.md) |
| ResNet-101+Dark | 256x192 | 0.733 | 0.786 | [resnet_dark_coco.md](./coco/resnet_dark_coco.md) |
| CSPNeXt-m | 256x192 | 0.732 | 0.785 | [cspnext_udp_coco.md](./coco/cspnext_udp_coco.md) |
| ResNetV1d-101 | 256x192 | 0.732 | 0.785 | [resnetv1d_coco.md](./coco/resnetv1d_coco.md) |
| SEResNet-50 | 256x192 | 0.729 | 0.784 | [seresnet_coco.md](./coco/seresnet_coco.md) |
| SCNet-50 | 256x192 | 0.728 | 0.784 | [scnet_coco.md](./coco/scnet_coco.md) |
| ResNet-101 | 256x192 | 0.726 | 0.783 | [resnet_coco.md](./coco/resnet_coco.md) |
| ResNeXt-101 | 256x192 | 0.726 | 0.781 | [resnext_coco.md](./coco/resnext_coco.md) |
| HourglassNet | 256x256 | 0.726 | 0.780 | [hourglass_coco.md](./coco/hourglass_coco.md) |
| ResNeSt-101 | 256x192 | 0.725 | 0.781 | [resnest_coco.md](./coco/resnest_coco.md) |
| RSN-50 | 256x192 | 0.724 | 0.790 | [rsn_coco.md](./coco/rsn_coco.md) |
| Swin-T | 256x192 | 0.724 | 0.782 | [swin_coco.md](./coco/swin_coco.md) |
| MSPN 1-stg | 256x192 | 0.723 | 0.788 | [mspn_coco.md](./coco/mspn_coco.md) |
| ResNetV1d-50 | 256x192 | 0.722 | 0.777 | [resnetv1d_coco.md](./coco/resnetv1d_coco.md) |
| ResNeSt-50 | 256x192 | 0.720 | 0.775 | [resnest_coco.md](./coco/resnest_coco.md) |
| ResNet-50 | 256x192 | 0.718 | 0.774 | [resnet_coco.md](./coco/resnet_coco.md) |
| ResNeXt-50 | 256x192 | 0.715 | 0.771 | [resnext_coco.md](./coco/resnext_coco.md) |
| PVT-S | 256x192 | 0.714 | 0.773 | [pvt_coco.md](./coco/pvt_coco.md) |
| CSPNeXt-s | 256x192 | 0.697 | 0.753 | [cspnext_udp_coco.md](./coco/cspnext_udp_coco.md) |
| LiteHRNet-30 | 256x192 | 0.676 | 0.736 | [litehrnet_coco.md](./coco/litehrnet_coco.md) |
| CSPNeXt-tiny | 256x192 | 0.665 | 0.723 | [cspnext_udp_coco.md](./coco/cspnext_udp_coco.md) |
| MobileNet-v2 | 256x192 | 0.648 | 0.709 | [mobilenetv2_coco.md](./coco/mobilenetv2_coco.md) |
| LiteHRNet-18 | 256x192 | 0.642 | 0.705 | [litehrnet_coco.md](./coco/litehrnet_coco.md) |
| CPM | 256x192 | 0.627 | 0.689 | [cpm_coco.md](./coco/cpm_coco.md) |
| ShuffleNet-v2 | 256x192 | 0.602 | 0.668 | [shufflenetv2_coco.md](./coco/shufflenetv2_coco.md) |
| ShuffleNet-v1 | 256x192 | 0.587 | 0.654 | [shufflenetv1_coco.md](./coco/shufflenetv1_coco.md) |
| AlexNet | 256x192 | 0.448 | 0.521 | [alexnet_coco.md](./coco/alexnet_coco.md) |
### MPII Dataset
| Model | Input Size | PCKh@0.5 | PCKh@0.1 | Details and Download |
| :------------: | :--------: | :------: | :------: | :-------------------------------------------------: |
| HRNet-w48+Dark | 256x256 | 0.905 | 0.360 | [hrnet_dark_mpii.md](./mpii/hrnet_dark_mpii.md) |
| HRNet-w48 | 256x256 | 0.902 | 0.303 | [hrnet_mpii.md](./mpii/cspnext_udp_mpii.md) |
| HRNet-w48 | 256x256 | 0.901 | 0.337 | [hrnet_mpii.md](./mpii/hrnet_mpii.md) |
| HRNet-w32 | 256x256 | 0.900 | 0.334 | [hrnet_mpii.md](./mpii/hrnet_mpii.md) |
| HourglassNet | 256x256 | 0.889 | 0.317 | [hourglass_mpii.md](./mpii/hourglass_mpii.md) |
| ResNet-152 | 256x256 | 0.889 | 0.303 | [resnet_mpii.md](./mpii/resnet_mpii.md) |
| ResNetV1d-152 | 256x256 | 0.888 | 0.300 | [resnetv1d_mpii.md](./mpii/resnetv1d_mpii.md) |
| SCNet-50 | 256x256 | 0.888 | 0.290 | [scnet_mpii.md](./mpii/scnet_mpii.md) |
| ResNeXt-152 | 256x256 | 0.887 | 0.294 | [resnext_mpii.md](./mpii/resnext_mpii.md) |
| SEResNet-50 | 256x256 | 0.884 | 0.292 | [seresnet_mpii.md](./mpii/seresnet_mpii.md) |
| ResNet-50 | 256x256 | 0.882 | 0.286 | [resnet_mpii.md](./mpii/resnet_mpii.md) |
| ResNetV1d-50 | 256x256 | 0.881 | 0.290 | [resnetv1d_mpii.md](./mpii/resnetv1d_mpii.md) |
| CPM | 368x368\* | 0.876 | 0.285 | [cpm_mpii.md](./mpii/cpm_mpii.md) |
| LiteHRNet-30 | 256x256 | 0.869 | 0.271 | [litehrnet_mpii.md](./mpii/litehrnet_mpii.md) |
| LiteHRNet-18 | 256x256 | 0.859 | 0.260 | [litehrnet_mpii.md](./mpii/litehrnet_mpii.md) |
| MobileNet-v2 | 256x256 | 0.854 | 0.234 | [mobilenetv2_mpii.md](./mpii/mobilenetv2_mpii.md) |
| ShuffleNet-v2 | 256x256 | 0.828 | 0.205 | [shufflenetv2_mpii.md](./mpii/shufflenetv2_mpii.md) |
| ShuffleNet-v1 | 256x256 | 0.824 | 0.195 | [shufflenetv1_mpii.md](./mpii/shufflenetv1_mpii.md) |
### CrowdPose Dataset
Results on CrowdPose test with [YOLOv3](https://github.com/eriklindernoren/PyTorch-YOLOv3) human detector
| Model | Input Size | AP | AR | Details and Download |
| :--------: | :--------: | :---: | :---: | :--------------------------------------------------------: |
| HRNet-w32 | 256x192 | 0.675 | 0.816 | [hrnet_crowdpose.md](./crowdpose/hrnet_crowdpose.md) |
| CSPNeXt-m | 256x192 | 0.662 | 0.755 | [hrnet_crowdpose.md](./crowdpose/cspnext_udp_crowdpose.md) |
| ResNet-101 | 256x192 | 0.647 | 0.800 | [resnet_crowdpose.md](./crowdpose/resnet_crowdpose.md) |
| HRNet-w32 | 256x192 | 0.637 | 0.785 | [resnet_crowdpose.md](./crowdpose/resnet_crowdpose.md) |
### AIC Dataset
Results on AIC val set with ground-truth bounding boxes.
| Model | Input Size | AP | AR | Details and Download |
| :--------: | :--------: | :---: | :---: | :----------------------------------: |
| HRNet-w32 | 256x192 | 0.323 | 0.366 | [hrnet_aic.md](./aic/hrnet_aic.md) |
| ResNet-101 | 256x192 | 0.294 | 0.337 | [resnet_aic.md](./aic/resnet_aic.md) |
### JHMDB Dataset
| Model | Input Size | PCK(norm. by person size) | PCK (norm. by torso size) | Details and Download |
| :-------: | :--------: | :-----------------------: | :-----------------------: | :----------------------------------------: |
| ResNet-50 | 256x256 | 96.0 | 80.1 | [resnet_jhmdb.md](./jhmdb/resnet_jhmdb.md) |
| CPM | 368x368 | 89.8 | 65.7 | [cpm_jhmdb.md](./jhmdb/cpm_jhmdb.md) |
### PoseTrack2018 Dataset
Results on PoseTrack2018 val with ground-truth bounding boxes.
| Model | Input Size | AP | Details and Download |
| :-------: | :--------: | :--: | :----------------------------------------------------------: |
| HRNet-w48 | 256x192 | 84.6 | [hrnet_posetrack18.md](./posetrack18/hrnet_posetrack18.md) |
| HRNet-w32 | 256x192 | 83.4 | [hrnet_posetrack18.md](./posetrack18/hrnet_posetrack18.md) |
| ResNet-50 | 256x192 | 81.2 | [resnet_posetrack18.md](./posetrack18/resnet_posetrack18.md) |
### Human-Art Dataset
Results on Human-Art validation dataset with detector having human AP of 56.2 on Human-Art validation dataset
| Model | Input Size | AP | AR | Details and Download |
| :-------: | :--------: | :---: | :---: | :---------------------------------------------------: |
| ViTPose-s | 256x192 | 0.381 | 0.448 | [vitpose_humanart.md](./humanart/vitpose_humanart.md) |
| ViTPose-b | 256x192 | 0.410 | 0.475 | [vitpose_humanart.md](./humanart/vitpose_humanart.md) |
Results on Human-Art validation dataset with ground-truth bounding-box
| Model | Input Size | AP | AR | Details and Download |
| :-------: | :--------: | :---: | :---: | :---------------------------------------------------: |
| ViTPose-s | 256x192 | 0.738 | 0.768 | [vitpose_humanart.md](./humanart/vitpose_humanart.md) |
| ViTPose-b | 256x192 | 0.759 | 0.790 | [vitpose_humanart.md](./humanart/vitpose_humanart.md) |
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment