Rethinking Common Assumptions to Mitigate Racial Bias in Face Recognition Datasets

TL;DR

This paper challenges common assumptions about racial bias in face recognition datasets, showing that training on specific racial subsets can reduce bias and improve fairness, contrary to previous beliefs.

Contribution

It demonstrates that training on only African faces can reduce bias and that adding more images of existing identities can enhance accuracy across races.

Findings

Training on African faces induces less bias.

Skewed datasets with more African faces can produce more equitable models.

Adding images of existing identities boosts accuracy across racial groups.

Abstract

Many existing works have made great strides towards reducing racial bias in face recognition. However, most of these methods attempt to rectify bias that manifests in models during training instead of directly addressing a major source of the bias, the dataset itself. Exceptions to this are BUPT-Balancedface/RFW and Fairface, but these works assume that primarily training on a single race or not racially balancing the dataset are inherently disadvantageous. We demonstrate that these assumptions are not necessarily valid. In our experiments, training on only African faces induced less bias than training on a balanced distribution of faces and distributions skewed to include more African faces produced more equitable models. We additionally notice that adding more images of existing identities to a dataset in place of adding new identities can lead to accuracy boosts across racial categories. Our code is available at https://github.com/j-alex-hanson/rethinking-race-face-datasets.