Tilted Cross-Entropy (TCE): Promoting Fairness in Semantic Segmentation

Attila Szabó, Hadi Jamali Rad, Siva-Datta Mannava

CVPRW 2021 pp. 2305-2310

doi:10.1109/CVPRW53098.2021.00261 /cvprw/2021/szabo2021cvprw-tilted/

Abstract

Traditional empirical risk minimization (ERM) for semantic segmentation can disproportionately advantage or disadvantage certain target classes in favor of an (unfair but) improved overall performance. Inspired by the recently introduced tilted ERM (TERM), we propose tilted cross-entropy (TCE) loss and adapt it to the semantic segmentation set-ting to minimize performance disparity among target classes and promote fairness. Through quantitative and qualitative performance analyses, we demonstrate that the proposed Stochastic TCE for semantic segmentation can offer improved overall fairness by efficiently minimizing the performance disparity among the target classes of Cityscapes.

PDF CVPRW Semantic Scholar

Cite

Text

Szabó et al. "Tilted Cross-Entropy (TCE): Promoting Fairness in Semantic Segmentation." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2021. doi:10.1109/CVPRW53098.2021.00261

Markdown

[Szabó et al. "Tilted Cross-Entropy (TCE): Promoting Fairness in Semantic Segmentation." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2021.](https://mlanthology.org/cvprw/2021/szabo2021cvprw-tilted/) doi:10.1109/CVPRW53098.2021.00261

BibTeX

@inproceedings{szabo2021cvprw-tilted,
  title     = {{Tilted Cross-Entropy (TCE): Promoting Fairness in Semantic Segmentation}},
  author    = {Szabó, Attila and Rad, Hadi Jamali and Mannava, Siva-Datta},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
  year      = {2021},
  pages     = {2305-2310},
  doi       = {10.1109/CVPRW53098.2021.00261},
  url       = {https://mlanthology.org/cvprw/2021/szabo2021cvprw-tilted/}
}