Learning Depth Estimation for Transparent and Mirror Surfaces
Abstract
Inferring the depth of transparent or mirror (ToM) surfaces represents a hard challenge for either sensors, algorithms, or deep networks. We propose a simple pipeline for learning to estimate depth properly for such surfaces with neural networks, without requiring any ground-truth annotation. We unveil how to obtain reliable pseudo labels by in-painting ToM objects in images and processing them with a monocular depth estimation model. These labels can be used to fine-tune existing monocular or stereo networks, to let them learn how to deal with ToM surfaces. Experimental results on the Booster dataset show the dramatic improvements enabled by our remarkably simple proposal.
Cite
Text
Costanzino et al. "Learning Depth Estimation for Transparent and Mirror Surfaces." International Conference on Computer Vision, 2023. doi:10.1109/ICCV51070.2023.00848Markdown
[Costanzino et al. "Learning Depth Estimation for Transparent and Mirror Surfaces." International Conference on Computer Vision, 2023.](https://mlanthology.org/iccv/2023/costanzino2023iccv-learning/) doi:10.1109/ICCV51070.2023.00848BibTeX
@inproceedings{costanzino2023iccv-learning,
title = {{Learning Depth Estimation for Transparent and Mirror Surfaces}},
author = {Costanzino, Alex and Ramirez, Pierluigi Zama and Poggi, Matteo and Tosi, Fabio and Mattoccia, Stefano and Di Stefano, Luigi},
booktitle = {International Conference on Computer Vision},
year = {2023},
pages = {9244-9255},
doi = {10.1109/ICCV51070.2023.00848},
url = {https://mlanthology.org/iccv/2023/costanzino2023iccv-learning/}
}