RenDetNet: Weakly-Supervised Shadow Detection with Shadow Caster Verification
Abstract
Existing shadow detection models struggle to differentiate dark image areas from shadows. In this paper, we tackle this issue by verifying that all detected shadows are real, i.e. they have paired shadow casters. We perform this step in a physically-accurate manner by differentiable re-rendering the scene and observing the changes stemming from carving out estimated shadow casters. Thanks to this approach, the RenDetNet proposed in this paper is the first learning-based shadow detection model whose supervisory signals can be computed in a self-supervised manner. The developed system compares favourably against recent models trained on our data. As part of this publication, we release our code on github .
Cite
Text
Kubiak et al. "RenDetNet: Weakly-Supervised Shadow Detection with Shadow Caster Verification." European Conference on Computer Vision Workshops, 2024. doi:10.1007/978-3-031-91838-4_11Markdown
[Kubiak et al. "RenDetNet: Weakly-Supervised Shadow Detection with Shadow Caster Verification." European Conference on Computer Vision Workshops, 2024.](https://mlanthology.org/eccvw/2024/kubiak2024eccvw-rendetnet/) doi:10.1007/978-3-031-91838-4_11BibTeX
@inproceedings{kubiak2024eccvw-rendetnet,
title = {{RenDetNet: Weakly-Supervised Shadow Detection with Shadow Caster Verification}},
author = {Kubiak, Nikolina and Wortman, Elliot and Mustafa, Armin and Phillipson, Graeme and Jolly, Stephen and Hadfield, Simon},
booktitle = {European Conference on Computer Vision Workshops},
year = {2024},
pages = {180-193},
doi = {10.1007/978-3-031-91838-4_11},
url = {https://mlanthology.org/eccvw/2024/kubiak2024eccvw-rendetnet/}
}