Mask Wearing Detection Model based on Deformable Detr

This paper presents a lightweight Deformable DETR detection method under the condition that the current wearing mask data set has a small number of samples, and the hardware conditions are limited. Traditional target detection has many problems, such as invalid and slow pixel candidate frame, too much RPN output redundancy, too few adjustable super parameters, and complex model adjustment. In this paper, a deformable DETR target detection method is proposed, and the effect of small target detection is good. Compared with other models, it achieves better performance, reduces the training time by 10 times, and greatly improves its fast convergence, computing, and storage efficiency.