Awesome

C3P: Cross-domain Pose Prior Propagation for Weakly Supervised 3D Human Pose Estimation

Introduction

This is the official implementation for the paper, "C3P: Cross-domain Pose Prior Propagation for Weakly Supervised 3D Human Pose Estimation", ECCV 2022.

C3P_code

CMUP_open is the source for CMU dataset and ITOP_open is for ITOP dataset.

Installation

Install the corresponding dependencies in the requirement.txt:

  pip install requirement.txt

mkdir model, output, data for both CMU_open and ITOP_open.
Download preprocessed data https://drive.google.com/file/d/127P1g2SaaovZ_7gyVLh9rd-XXycXnf4y/view?usp=sharing to CMUP_open/data.
Download preprocessed data https://drive.google.com/file/d/1nEoD8qs-8XpSI7PRmMvE4Hc--H531mxC/view?usp=sharing to ITOP_open/data.

Test

  Download model https://drive.google.com/file/d/1XE3M4h5Lf9OxVWSQwKoolDpd9xF4xUch/view?usp=sharing to ITOP_open/model.
  Download model https://drive.google.com/file/d/1XE3M4h5Lf9OxVWSQwKoolDpd9xF4xUch/view?usp=sharing to MUP_open/model.
  Download model https://drive.google.com/file/d/1zjuVIOWQ_FSH4_pq0Acm7Hvjmbs8pUY4/view?usp=sharing to CMUP_open/model.
  Test ITOP dataset:

cd ITOP_open
python test.py   

  Test CMU_Panoptic dataset:

cd CMUP_open
python test.py   

 (you can test different models via modifing the corresponding code in test.py)

Train

cd CMUP_open
python train.py   

  the model and log file will be saved in output folder.

If you find our work useful in your research or publication, please cite our work:

@inproceedings{C3P,
author = {Wu, Cunlin and Xiao, Yang and Zhang, Boshen and Zhang Mingyang and Cao, Zhiguo and Zhou Tianyi, Joey},
title = {C3P: Cross-domain Pose Prior Propagation for Weakly Supervised 3D Human Pose Estimation},
booktitle = {Proceedings of the European Conference on Computer Vision (ECCV)},
year = {2022}
}

Relevant paper: A2J: Anchor-to-Joint Regression Network for 3D Articulated Pose Estimation from a Single Depth Image.

@inproceedings{A2J,
author = {Xiong, Fu and Zhang, Boshen and Xiao, Yang and Cao, Zhiguo and Yu, Taidong and Zhou Tianyi, Joey and Yuan, Junsong},
title = {A2J: Anchor-to-Joint Regression Network for 3D Articulated Pose Estimation from a Single Depth Image},
booktitle = {Proceedings of the IEEE Conference on International Conference on Computer Vision (ICCV)},
year = {2019}
}