Small-Scale Face Detection Method Based on YOLO5Face Redistribution
Aiming at the problem of low accuracy of small-scale face detection in complex scenes,a small-scale face detection method is proposed to improve YOLO5Face.Based on YOLO5Face,the method introduces improved CBAM attention in the shallow layer of the network and computationally redistributes the model to improve the accura-cy of small-scale face detection in complex scenes while reducing the number of model parameters;The data enhance-ment method of fusion mixup was adopted to fully train the small-scale face detection branch of the model;Softmax loss was used as the classification loss according to the face detection characteristics to Maximize the difference of fea-tures between classes.The experimental results on each subset of WiderFace show that the improved model meets the real-time performance and has higher accuracy of small-scale face detection compared with the mainstream face de-tection methods,in which the detection accuracy of Hard subset is improved by 2 percentage points compared with YOLO5Face.
Face detectionSmall-scaleComputation redistributionClassification loss
NETL
NSTL
万方数据
