Mask Detection Algorithm Based on Improved Lightweight YOLOv3
Lingyun Bi, Lixia Deng, Hongquan Li, Haiying Liu · 2022 IEEE 5th International Conference on Information Systems and Computer Aided Education (ICISCAE) · 2022
In the object detection task, in order to reduce the reasoning time of the mask detection task, reduce the computing cost, and avoid a certain degree of computing resource waste caused by the full convolution network adopted by YOLOv3, this paper proposes an Mdwneck network, which improves the Bottleneck in the original algorithm and changes its stacking mode to achieve a more lightweight effect. The experimental results show that the mAP index of the improved algorithm is only 0.8% lower than that of the original algorithm, and there is almost no difference in the actual detection scenario. At the same time, the computing cost is greatly reduced and the expected effect is achieved.