One-Shot Learning for Human Affordance Detection
Abstract
The diversity of action possibilities offered by an environment, a.k.a affordances, cannot be addressed in a scalable manner simply from object categories or semantics, which are limitless. To this end, we present a one-shot learning approach that trains on one or a handful of human-scene interaction samples. Then, given a previously unseen scene, we can predict human affordances and generate the associated articulated 3D bodies. Our experiments show that our approach generates physically plausible interactions that are perceived as more natural in 60–70% of the comparisons with other methods.
Cite
Text
Pacheco-Ortega and Mayol-Cuevas. "One-Shot Learning for Human Affordance Detection." European Conference on Computer Vision Workshops, 2022. doi:10.1007/978-3-031-25066-8_46Markdown
[Pacheco-Ortega and Mayol-Cuevas. "One-Shot Learning for Human Affordance Detection." European Conference on Computer Vision Workshops, 2022.](https://mlanthology.org/eccvw/2022/pachecoortega2022eccvw-oneshot/) doi:10.1007/978-3-031-25066-8_46BibTeX
@inproceedings{pachecoortega2022eccvw-oneshot,
title = {{One-Shot Learning for Human Affordance Detection}},
author = {Pacheco-Ortega, Abel and Mayol-Cuevas, Walterio W.},
booktitle = {European Conference on Computer Vision Workshops},
year = {2022},
pages = {758-766},
doi = {10.1007/978-3-031-25066-8_46},
url = {https://mlanthology.org/eccvw/2022/pachecoortega2022eccvw-oneshot/}
}