TRFH: towards real-time face detection and head pose estimation

Abstract Nowadays, face detection and head pose estimation have a lot of application such as recognition, aiding in gaze modeling attention. For these two tasks, it is usually to design different models. However, the model often depends on region interest (ROI) detected advance, which means that serial detector needed. Even lightest will slow down whole forward inference time cannot achieve real-time performance when detecting multiple people. We can see both need features, so shared feature map be used between them. In this paper, multi-task learning proposed solve problems simultaneously. directly detect location center point bounding box face; at location, we calculate size attitude. evaluate our model’s AFLW. The has great competitiveness with multi-stage attribute analysis model, performance.