End-to-End Cascade CNN for Simultaneously Face Detection and Alignment

Abstract

Real-world face detection and alignment demand an advanced discriminative model to address challenges by pose, lighting and expression. Recent studies have utilized the relation between face detection and alignment to make models computationally efficiency, but they ignore the connection between each cascade CNNs. In this paper, we combine detection, calibration and alignment in each cascade structure and propose an End-to-End cascade Online Hard Example Mining (OHEM) for training, which expert in accelerating convergence. Experiments on FDDB and AFLW demonstrate considerable improvement on accuracy and speed.