Official PyTorch Implementation of the paper: Learning Temporal 3D Human Pose Estimation with Pseudo Labels.
Arij Bouazizi, Ulrich Kressel, and Vasileios Belagiannis
[Proceedings] [Papers with Code] [Arxiv]
To setup the environment:
cd TM_HPE
conda create -n TM_HPE python=3.8.8
conda activate TM_HPE
pip install -r requirements.txt
Due to licensing it is not possible to provide any data. Please refer to VideoPose3D for the preparation of the dataset files.
To train the model on h36m or amass, you can use the following commands:
python h36m/train_h36m.py -e 80 -k cpn_ft_h36m_dbb -arc 3,3,3,3,3
python amass/train_3dhp.py -e 80 -k cpn_ft_h36m_dbb -arc 3,3,3,3,3
To test the pretrained models, you can use the following commands:
python h36m/test_h36m.py -e 80 -k cpn_ft_h36m_dbb -arc 3,3,3,3,3
python amass/test_3dhp.py -e 80 -k cpn_ft_h36m_dbb -arc 3,3,3,3,3
We release the pretrained models for academic purpose. You can download them from. Unzip the .zip file in the /checkpoints
directory.
If you find this code useful for your research, please consider citing the following paper:
@inproceedings{bouazizi2021learning,
title={Learning temporal 3d human pose estimation with pseudo-labels},
author={Bouazizi, Arij and Kressel, Ulrich and Belagiannis, Vasileios},
booktitle={2021 17th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS)},
pages={1--8},
year={2021},
organization={IEEE}
}
Some of our code was adapted from VideoPose3D. We thank the authors for making their code public.
This work is licensed under Creative Commons Attribution-NonCommercial 4.0 International License.