Poseur: Direct Human Pose Regression with Transformers
Poseur: Direct Human Pose Regression with Transformers,
Weian Mao*, Yongtao Ge*, Chunhua Shen, Zhi Tian, Xinlong Wang, Zhibin Wang, Anton van den Hengel
In: European Conference on Computer Vision (ECCV), 2022
arXiv preprint (arXiv 2201.07412)
(* equal contribution)
News ๐ฉ
[2023/04/17] Release a naive version of Poseur based on ViT backbone. Please see poseur_vit_base_coco_256x192.
[2023/04/17] Release a naive version of Poseur trained on COCO-Wholebody dataset. Please see poseur_coco_wholebody.
Introduction
This project is bulit upon MMPose with commit ID eeebc652842a9724259ed345c00112641d8ee06d.
Installation & Quick Start
- Install following packages
pip install easydict einops
- Follow the MMPose instruction to install the project and set up the datasets (MS-COCO).
For training on COCO, run:
./tools/dist_train.sh \
configs/poseur/coco/poseur_r50_coco_256x192.py 8 \
--work-dir work_dirs/poseur_r50_coco_256x192
For evaluating on COCO, run the following command lines:
wget https://cloudstor.aarnet.edu.au/plus/s/UXr1Dn9w6ja4fM9/download -O poseur_256x192_res50_6dec_coco.pth
./tools/dist_test.sh configs/poseur/coco/poseur_res50_coco_256x192.py \
poseur_256x192_r50_6dec_coco.pth 4 \
--eval mAP \
--cfg-options model.filp_fuse_type=\'type2\'
For visualizing on COCO, run the following command lines:
python demo/top_down_img_demo.py \
configs/poseur/coco/poseur_res50_coco_256x192.py \
poseur_256x192_res50_6dec_coco.pth \
--img-root tests/data/coco/ --json-file tests/data/coco/test_coco.json \
--out-img-root vis_results_poseur
COCO Keypoint Detection
| Name | AP | AP.5 | AP.75 | download link |
|---|---|---|---|---|
| poseur_mobilenetv2_coco_256x192 | 71.9 | 88.9 | 78.6 | model |
| poseur_mobilenetv2_coco_256x192_12dec | 72.3 | 88.9 | 78.9 | model |
| poseur_res50_coco_256x192 | 75.5 | 90.7 | 82.6 | model |
| poseur_hrnet_w32_coco_256x192 | 76.8 | 91.0 | 83.5 | model |
| poseur_hrnet_w48_coco_384x288 | 78.7 | 91.6 | 85.1 | model |
| poseur_hrformer_tiny_coco_256x192_3dec | 74.2 | 90.1 | 81.4 | model |
| poseur_hrformer_small_coco_256x192_3dec | 76.6 | 91.0 | 83.4 | model |
| poseur_hrformer_big_coco_256x192 | 78.9 | 91.9 | 85.6 | model |
| poseur_hrformer_big_coco_384x288 | 79.6 | 92.1 | 85.9 | model |
| poseur_vit_base_coco_256x192 | 76.7 | 90.6 | 83.5 | model |
COCO-WholeBody Benchmark (V0.5)
Compare Whole-body pose estimation results with other methods.
| Method | body | foot | face | hand | whole | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| AP | AR | AP | AR | AP | AR | AP | AR | AP | AR | |
| OpenPose [1] | 0.563 | 0.612 | 0.532 | 0.645 | 0.482 | 0.626 | 0.198 | 0.342 | 0.338 | 0.449 |
| HRNet [2] | 0.659 | 0.709 | 0.314 | 0.424 | 0.523 | 0.582 | 0.300 | 0.363 | 0.432 | 0.520 |
| HRNet-body [2] | 0.758 | 0.809 | - | - | - | - | - | - | - | - |
| ZoomNet [3] | 0.743 | 0.802 | 0.798 | 0.869 | 0.623 | 0.701 | 0.401 | 0.498 | 0.541 | 0.658 |
| ZoomNas [4] | 0.740 | - | 0.617 | - | 0.889 | - | 0.625 | - | 0.654 | - |
| RTMPose [5] | 0.730 | - | 0.734 | - | 0.898 | - | 0.587 | - | 0.669 | - |
| Poseur_ResNet50 | 0.655 | 0.732 | 0.615 | 0.742 | 0.844 | 0.900 | 0.560 | 0.673 | 0.587 | 0.681 |
| Poseur_HRNet_W32 | 0.680 | 0.753 | 0.668 | 0.780 | 0.863 | 0.912 | 0.604 | 0.706 | 0.620 | 0.707 |
| Poseur_HRNet_W48 | 0.692 | 0.766 | 0.689 | 0.799 | 0.861 | 0.911 | 0.621 | 0.721 | 0.633 | 0.721 |
COCO-WholeBody Pretrain Models
| Name | AP | AP.5 | AP.75 | download link |
|---|---|---|---|---|
| poseur_res50_coco_wholebody_256x192 | 65.5 | 85.0 | 71.8 | model |
| poseur_hrnet_w32_coco_wholebody_256x192 | 68.0 | 85.8 | 74.4 | model |
| poseur_hrnet_w48_coco_wholebody_256x192 | 69.2 | 86.0 | 75.7 | model |
Disclaimer:
- Due to the update of MMPose, the results are slightly different from our original paper.
- We use the official HRFormer implement from here, the implementation in mmpose has not been verified by us.
Citations
Please consider citing our papers in your publications if the project helps your research. BibTeX reference is as follows.
@inproceedings{mao2022poseur,
title={Poseur: Direct human pose regression with transformers},
author={Mao, Weian and Ge, Yongtao and Shen, Chunhua and Tian, Zhi and Wang, Xinlong and Wang, Zhibin and Hengel, Anton van den},
journal = {Proceedings of the European Conference on Computer Vision {(ECCV)}},
month = {October},
year={2022}
}Reference
[1] Andriluka, M., Pishchulin, L., Gehler, P., Schiele, B.: 2d human pose estimation: New benchmark and state of the art analysis. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2014)
[2] Sun, K., Xiao, B., Liu, D., Wang, J.: Deep high-resolution representation learning for human pose estimation. arXiv preprint arXiv:1902.09212 (2019)
[3] Sheng Jin, Lumin Xu, Jin Xu, Can Wang, Wentao Liu, Chen Qian, Wanli Ouyang, Ping Luo. Whole-Body Human Pose Estimation in the Wild. (ECCV) (2020)
[4] Lumin Xu, Sheng Jin, Wentao Liu, Chen Qian, Wanli Ouyang, Ping Luo, Xiaogang Wang: ZoomNAS: Searching for Whole-body Human Pose Estimation in the Wild In: IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) (2022)
[5] Tao Jiang, Peng Lu, Li Zhang, Ningsheng Ma, Rui Han, Chengqi Lyu, Yining Li, Kai Chen. RTMPose: Real-Time Multi-Person Pose Estimation based on MMPose. arXiv preprint arXiv:2303.07399 (2023)
License
For commercial use, please contact Chunhua Shen.