Joint Face Detection and Alignment Using Multitask Cascaded Convolutional Networks

The paper addresses face detection and alignment in unconstrained environments, where varying poses, illuminations, and occlusions make both tasks difficult. It proposes a deep cascaded multitask framework that leverages the inherent correlation between detection and alignment, using three stages of carefully designed deep convolutional networks to predict face and landmark locations in a coarse-to-fine manner. An online hard sample mining strategy is introduced to further improve performance in practice.

By jointly modeling the two related tasks, the framework boosts the performance of each. It achieves accuracy superior to state-of-the-art techniques on challenging face detection benchmarks, including WIDER FACE, and on a facial landmark alignment benchmark, all while keeping real-time performance, which matters for deploying face analysis in practical systems.