Single Image 3D Human Pose Estimation Using Sequential Joint Group Generation
Abstract
While 3D human pose estimation from a single image has been a topic of great interest in the computer vision community for many years, the dependencies between parts of the pose have yet to be fully explored. Motivated by this gap in the literature, we present a novel sequential joint group generation model for the 2D to 3D lifting problem. We split the joints into groups, and the network estimates the locations of each group, while leveraging previously generated 3D parts to create a plausible complete pose. The model achieves an improvement over the non-sequential baseline on the Human3.6M dataset, and our experiments show that this part-based approach has great potential and room to grow with further work in the area.
Cite
Text
Lisowski et al. "Single Image 3D Human Pose Estimation Using Sequential Joint Group Generation." European Conference on Computer Vision Workshops, 2024. doi:10.1007/978-3-031-91581-9_7Markdown
[Lisowski et al. "Single Image 3D Human Pose Estimation Using Sequential Joint Group Generation." European Conference on Computer Vision Workshops, 2024.](https://mlanthology.org/eccvw/2024/lisowski2024eccvw-single/) doi:10.1007/978-3-031-91581-9_7BibTeX
@inproceedings{lisowski2024eccvw-single,
title = {{Single Image 3D Human Pose Estimation Using Sequential Joint Group Generation}},
author = {Lisowski, Szymon and Hardy, Peter and Kim, Hansung},
booktitle = {European Conference on Computer Vision Workshops},
year = {2024},
pages = {90-100},
doi = {10.1007/978-3-031-91581-9_7},
url = {https://mlanthology.org/eccvw/2024/lisowski2024eccvw-single/}
}