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

Attila Szabo, Hadi Jamali-Rad, Siva-Datta Mannava; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2021, pp. 2305-2310

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 setting 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.

Related Material


[pdf] [supp]
[bibtex]
@InProceedings{Szabo_2021_CVPR, author = {Szabo, Attila and Jamali-Rad, Hadi and Mannava, Siva-Datta}, title = {Tilted Cross-Entropy (TCE): Promoting Fairness in Semantic Segmentation}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2021}, pages = {2305-2310} }