How Knowledge Distillation Mitigates the Synthetic Gap in Fair Face Recognition

TL;DR

This paper demonstrates that knowledge distillation can improve the accuracy and fairness of face recognition models trained on synthetic data, reducing bias and bridging the performance gap with real data.

Contribution

It introduces a knowledge distillation strategy that enhances face recognition fairness and mitigates the synthetic gap, validated across multiple models and datasets.

Findings

KD improves performance across all ethnicities

KD reduces bias in face recognition models

KD mitigates the performance gap between real and synthetic datasets

Abstract

Leveraging the capabilities of Knowledge Distillation (KD) strategies, we devise a strategy to fight the recent retraction of face recognition datasets. Given a pretrained Teacher model trained on a real dataset, we show that carefully utilising synthetic datasets, or a mix between real and synthetic datasets to distil knowledge from this teacher to smaller students can yield surprising results. In this sense, we trained 33 different models with and without KD, on different datasets, with different architectures and losses. And our findings are consistent, using KD leads to performance gains across all ethnicities and decreased bias. In addition, it helps to mitigate the performance gap between real and synthetic datasets. This approach addresses the limitations of synthetic data training, improving both the accuracy and fairness of face recognition models.