125 / 2022-10-20 16:05:36

A single-stage robust face detector based on multi-scale feature fusion

Face detection is an important research content in the field of computer vision. With the development of deep learning network model, traditional face detection has achieved good accuracy. However, due to the influence of occlusion, illumination and scale change, the research of robust face detection is still a challenging task. In this paper, a single-stage robust face detector based on multi-scale feature fusion is proposed, which combines the advantages of multi-task learning to locate the pixelated faces of people of various scales, thus reducing the influence of environment and size factors and improving the detection performance. In the training process, RegNet+ is used as the backbone network, and an adaptive parallel sampling method of hole convolution is adopted, which can realize the parallel sampling of ASPP hole convolution and FPN pyramid, and has better robustness to face detection in special environment. A large number of experimental results show that the accuracy of our face detector can reach 95.592% on WIDER FACE test set, which is about 1% higher than the average accuracy of the existing method Tinaface.