How Large-Scale Training Samples Effect Face Detector? An Empirical Analysis

Recent development in the field of face detection highlights the benefits from large scale training samples, which can be cheaply collected through internet. However, these large training sets are usually constructed in a rather arbitrary manner. In this paper, we empirically investigate the fundamental question of how the training set effects the performance of a given state of the art face detector. In particular, we construct a very large training set containing over 340K face images and study the effect of five common factors of variations (i.e., lighting, expression, blurring, contrast change and noise) which may change face appearance largely. Our results show that noise factor has the most significant influence on the performance of the detector while others (e.g., lighting, expression) are of much less importance. Based on these, we propose a new method to construct an effective training set with much small size for face detection, without significantly reducing the performance.