Faceboxes: A CPU real-time and accurate unconstrained face detector

Although tremendous strides have been made in face detection, one of the remaining open issues is to achieve CPU real-time speed as well as maintain high performance, since effective models for face detection tend to be computationally prohibitive. To address this issue, we propose a novel face detector, named FaceBoxes, with superior performance on both speed and accuracy. Specifically, the proposed method has a lightweight yet powerful network that consists of the Rapidly Digested Convolution Layers (RDCL) and the Multiple Scale Convolution Layers (MSCL). The former is designed to enable FaceBoxes to achieve CPU real-time speed, while the latter aims to enrich the features and discretize anchors over different layers to handle faces of various scales. Besides, we propose a new anchor densification strategy to make different types of anchors have the same density on the image, which significantly improves the recall rate of small faces. Finally, we present a Divide and Conquer Head (DCH) to boost the prediction ability of the detection layer using above strategy. As a consequence, the proposed detector runs at 28 FPS on the CPU and 254 FPS using a GPU for VGA-resolution images. Moreover, the speed of FaceBoxes is invariant to the number of faces. We evaluate the proposed method on several face detection benchmarks including AFW, PASCAL face, FDDB, WIDER FACE and achieve state-of-the-art performance among CPU real-time methods. (C) 2019 Elsevier B.V. All rights reserved.