Gone with the Bits: Revealing Racial Bias in Low-Rate Neural Compression for Facial Images

Neural compression methods are gaining popularity due to their superior rate-distortion performance over traditional methods, even at extremely low bitrates below 0.1 bpp. As deep learning models, they are prone to bias during the training, potentially leading to unfair outcomes for individuals in different groups. In this paper, we present a scalable framework for evaluating bias in 9 neural image compression models. We first demonstrate that traditional distortion metrics are ineffective in capturing bias in these models. Next, we highlight that racial bias is present in all neural compression models and can be captured by examining facial phenotype degradation in image reconstructions. Finally, we show that utilizing a racially balanced training set can reduce bias but is not a sufficient bias mitigation strategy, since the bias can be attributed to both compression model bias and classification model bias. We believe that this work is a first step towards evaluating and eliminating bias in neural image compression models.