Yolov4 High-Speed Train Wheelset Tread Defect Detection System Based on Multiscale Feature Fusion

Table 5

mAP of MFF-Yolov4 on WT-DET: use L2 normalization.

The connected layer comes from:					mAP (%)
F1	F2	F3	F4	F6	Without L2 norm	L2 norm
√		√	√	√	77.19	81.86
	√	√	√	√	83.85	86.21
√	√	√	√	√	76.70	75.45