Face SSD: A Real-time Face Detector based on SSD
Bin Ye, Yunlin Shi, Huijun Li, Liuchuan Li, Shuo Tong · 2021
Face detection has made substantial progress in recent years. In many applications, face detectors must run on mobile devices or embedded devices. Due to the limited computing resource in such scenarios, it is usually difficult to meet the need of both accuracy and speed. Most detectors cannot detect small face accurately or the detection speed will slow down. To address this challenge, we propose a novel face detector in which the basic framework is a single shot multibox detector (SSD), and name it Face SSD. In order to accelerate the detection speed of Face SSD and improve the detection accuracy, we make contributions in the following three aspects: we improved the structure of ShuffleNet V2 and use it as the backbone network; then we proposed a modified prediction module (MPM) to improve recall of small faces; finally we introduced a scale-equitable face detection framework to match face better. Experimental results show that Face SSD runs 40 frames per second (FPS) on CPU and 110 FPS on GPU. We train Face SSD on WIDER FACE - a face detection benchmark. The detector has also achieved excellent performance on FDDB, a benchmark for face detection in unconstrained settings.