Real-Time and Accurate Face Detection Networks Based on Deep Learning

In this paper, we propose a real-time and accurate face detection network based on MobileNets, SSD (Single Shot MultiBox Detector) and FPN (Feature Pyramid Networks). Our network adopts MobileNets and SSD as the basic networks and introduces FPN to fuse the multi-layer information, which makes the features extracted from the network more abundant and plays a positive role in face detection. In addition, the design of the network loss function draws on the loss function form of MultiBox, divides the whole loss function into two parts of confidence loss and localization loss, and expands the binary cross entropy loss of MultiBox into the loss function in the form of multi-category focal loss, which solves the unbalanced phenomenon of category during the training process.