MPL: Lifting 3D Human Pose from Multi-View 2D Poses
Abstract
Estimating 3D human poses from 2D images is challenging due to occlusions and projective acquisition. Learning-based approaches have been largely studied to address this challenge, both in single and multi-view setups. These solutions however fail to generalize to real-world cases due to the lack of (multi-view) ‘in-the-wild’ images paired with 3D poses for training. For this reason, we propose combining 2D pose estimation, for which large and rich training datasets exist, and 2D-to-3D pose lifting, using a transformer-based network that can be trained from synthetic 2D-3D pose pairs. Our experiments demonstrate decreases up to $45\%$ 45 % in MPJPE errors compared to the 3D pose obtained by triangulating the 2D poses. The framework’s source code is available at https://github.com/aghasemzadeh/OpenMPL .
Cite
Text
Ghasemzadeh et al. "MPL: Lifting 3D Human Pose from Multi-View 2D Poses." European Conference on Computer Vision Workshops, 2024. doi:10.1007/978-3-031-91575-8_3Markdown
[Ghasemzadeh et al. "MPL: Lifting 3D Human Pose from Multi-View 2D Poses." European Conference on Computer Vision Workshops, 2024.](https://mlanthology.org/eccvw/2024/ghasemzadeh2024eccvw-mpl/) doi:10.1007/978-3-031-91575-8_3BibTeX
@inproceedings{ghasemzadeh2024eccvw-mpl,
title = {{MPL: Lifting 3D Human Pose from Multi-View 2D Poses}},
author = {Ghasemzadeh, Seyed Abolfazl and Alahi, Alexandre and De Vleeschouwer, Christophe},
booktitle = {European Conference on Computer Vision Workshops},
year = {2024},
pages = {36-52},
doi = {10.1007/978-3-031-91575-8_3},
url = {https://mlanthology.org/eccvw/2024/ghasemzadeh2024eccvw-mpl/}
}