Rethinking Common Assumptions to Mitigate Racial Bias in Face Recognition Datasets
This work addresses racial bias in face recognition systems by rethinking dataset composition, offering insights for more equitable AI, though it is incremental as it builds on prior dataset-focused methods.
The paper challenges assumptions about racial bias in face recognition datasets, showing that training only on African faces reduces bias more than balanced distributions and that adding more images per identity boosts accuracy across races.
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.