Robust Face Detector with Fully Convolutional Networks

Abstract

Many of the exist face detection algorithms are based on the generic object detection methods and have achieved desirable results. However, these methods still struggle in solving the problem of partial occluded face detection. In this paper, we introduce a simple and effective face detector which uses a fully convolutional networks (FCN) for face detection in a single stage. The proposed FCN model is used for pixel-wise prediction instead of anchor mechanism. In addition, we also apply a long short term memory (LSTM) architecture to enhance the contextual infomation of feature maps, making the model more robust to occlusion. Besides, we use a light-weighted neural network PVANet as the backbone, which greatly reduces the computational burden. Experimental results show that the proposed method achieves competitive results with state-of-the-art face detectors on the common face detection benchmarks, including the FDDB, WIDER FACE and MAFA datasets, what’s more, it is much more robust to the detection of occluded faces.