What Should Be Balanced in a "Balanced" Face Recognition Dataset?
This work addresses the issue of biased evaluations in face recognition for researchers and practitioners, but it is incremental as it builds on existing dataset critiques.
The paper tackles the problem of demographic disparities in face recognition accuracy by showing that current 'balanced' datasets, which balance identities and images, do not address key factors like head pose and image quality, and proposes a bias-aware toolkit for creating better evaluation datasets.
The issue of demographic disparities in face recognition accuracy has attracted increasing attention in recent years. Various face image datasets have been proposed as 'fair' or 'balanced' to assess the accuracy of face recognition algorithms across demographics. These datasets typically balance the number of identities and images across demographics. It is important to note that the number of identities and images in an evaluation dataset are {\em not} driving factors for 1-to-1 face matching accuracy. Moreover, balancing the number of identities and images does not ensure balance in other factors known to impact accuracy, such as head pose, brightness, and image quality. We demonstrate these issues using several recently proposed datasets. To improve the ability to perform less biased evaluations, we propose a bias-aware toolkit that facilitates creation of cross-demographic evaluation datasets balanced on factors mentioned in this paper.