The Emergence of Race and Sex from Face Recognition System

Abstract

Can computer vision recover attributes such as sex and race(or some approximation thereof) from facial images without being trained for such classification? If so, how do these categories constructed by machines related to those constructed by society? To study this problem, we use face recognition systems to extract facial features from images and we show that we can achieve very high sex and race classification using these features. We will also discuss the implications of these results.