Detecting Face with Densely Connected Face Proposal Network

Accuracy and efficiency are two conflicting challenges for face detection, since effective models tend to be computationally prohibitive. To address these two conflicting challenges, our core idea is to shrink the input image and focus on detecting small faces. Reducing the image resolution can significantly improve the detection speed, but it also results in smaller faces that need to pay more attention. Specifically, we propose a novel face detector, dubbed the name Densely Connected Face Proposal Network (DCFPN), with high accuracy as well as CPU real-time speed. Firstly, we subtly design a lightweight-but-powerful fully convolution network with the consideration of efficiency and accuracy. Secondly, we present a dense anchor strategy and a scale-aware anchor matching scheme to improve the recall rate of small faces. Finally, a fair L1 loss is introduced to locate small faces well. As a consequence, our proposed method can detect faces at 30 FPS on a single 2.60 GHz CPU core and 250 FPS using a GPU for the VGA-resolution images. We achieve state-of-the-art performance on the common face detection benchmark datasets.