Understanding Unequal Gender Classification Accuracy from Face Images

TL;DR

This study investigates the causes of unequal gender classification accuracy across intersectional groups in face images, finding that facial features like lip, eye, and cheek structure, rather than skin type or hair length, contribute to disparities.

Contribution

The paper introduces a comprehensive analysis revealing that facial structural differences, not skin type or hair, drive gender classification bias, and highlights the role of makeup as a gender stereotype in AI models.

Findings

Skin type is not the main factor in accuracy disparity.

Facial structural differences contribute to gender classification bias.

Lip and eye makeup influence gender prediction and propagate stereotypes.

Abstract

Recent work shows unequal performance of commercial face classification services in the gender classification task across intersectional groups defined by skin type and gender. Accuracy on dark-skinned females is significantly worse than on any other group. In this paper, we conduct several analyses to try to uncover the reason for this gap. The main finding, perhaps surprisingly, is that skin type is not the driver. This conclusion is reached via stability experiments that vary an image's skin type via color-theoretic methods, namely luminance mode-shift and optimal transport. A second suspect, hair length, is also shown not to be the driver via experiments on face images cropped to exclude the hair. Finally, using contrastive post-hoc explanation techniques for neural networks, we bring forth evidence suggesting that differences in lip, eye and cheek structure across ethnicity lead to the differences. Further, lip and eye makeup are seen as strong predictors for a female face, which is a troubling propagation of a gender stereotype.