A lightweight network face detection based on YOLOv5 Lightweight model face detection based on YOLOv5 combined with Mobilenetv2

Face detection is a mainstream way of human identification today, but the current object detection network requires a large number of parameters and calculations, in response to this problem, we propose a lightweight improved face detection method based on YOLOv5, which can be easily deployed in embedded devices and mobile devices. We replaced YOLOv5s backbone with a more lightweight MobileNet, replacing standard convolution with deep separable convolution; Then we also use the SIoU loss function to speed up the convergence speed of the prediction box and the target box, improving the inference rate. Compared with the baseline model, the YOLOv5-MS model reduces the number of parameters by 91.4% and increases FPS by 62.5%, which achieves a good balance between detection accuracy and rate.