3DB: A Framework for Debugging Computer Vision Models
Abstract
We introduce 3DB: an extendable, unified framework for testing and debugging vision models using photorealistic simulation. We demonstrate, through a wide range of use cases, that 3DB allows users to discover vulnerabilities in computer vision systems and gain insights into how models make decisions. 3DB captures and generalizes many robustness analyses from prior work, and enables one to study their interplay. Finally, we find that the insights generated by the system transfer to the physical world. 3DB will be released as a library alongside a set of examples and documentation. We attach 3DB to the submission.
Cite
Text
Leclerc et al. "3DB: A Framework for Debugging Computer Vision Models." Neural Information Processing Systems, 2022.Markdown
[Leclerc et al. "3DB: A Framework for Debugging Computer Vision Models." Neural Information Processing Systems, 2022.](https://mlanthology.org/neurips/2022/leclerc2022neurips-3db/)BibTeX
@inproceedings{leclerc2022neurips-3db,
title = {{3DB: A Framework for Debugging Computer Vision Models}},
author = {Leclerc, Guillaume and Salman, Hadi and Ilyas, Andrew and Vemprala, Sai and Engstrom, Logan and Vineet, Vibhav and Xiao, Kai and Zhang, Pengchuan and Santurkar, Shibani and Yang, Greg and Kapoor, Ashish and Madry, Aleksander},
booktitle = {Neural Information Processing Systems},
year = {2022},
url = {https://mlanthology.org/neurips/2022/leclerc2022neurips-3db/}
}