Face Detection on Hard Datasets

Face detection algorithms are deployed in a wide variety of applications. Unfortunately, there has been no quantitative comparison of how these detectors perform under uncontrolled circumstances. We created a dataset of low light and long distance images which possess some of the problems encountered by face detectors in the real world. We hope to advance and define the state of the art by challenging the computer vision community to compete on this dataset. The dataset we created is composed of photographs and semi-synthetic heads photographed under varying conditions of low light, atmospheric blur, and a variety of distances: 3m, 80m, and 200m. This paper describes the performance of the participants’ face detectors against those of the Viola Jones detector and three leading commercial face detectors. We compared each detector’s ability to both detect and localize faces and eyes.