Simultaneous Face Detection and Head Pose Estimation: A Fast and Unified Framework

Tingfeng Li, Xu Zhao

Abstract: In this paper, we present a fast and unified framework for simultaneous face detection and 3D pose (pitch, yaw, roll) estimation of unconstrained faces using deep convolutional neural networks (CNN). Face detection is implemented with region-based framework as previous work like Faster RCNN. We model the pose estimation as a classification and regression problem: first divide continuous head poses into several discrete clusters, then adjust poses within each class with a class-specific regressor to achieve more accurate results. All classification and regressions for the two tasks are trained and tested simultaneously in one unified network. Our approach runs at 10 fps, which is the fastest implementation among the recently proposed methods as far as we know. Moreover, it is able to predict pose without using any 3D information. Extensive evaluations on several challenging benchmarks such as AFLW and AFW demonstrate the effectiveness of the proposed method with competitive results.