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

Table 2

Detection results on WT-DET.

Method	Recall (%)	Precision (%)	mAP (%)	Fps	F1
SSD [19]	69.70	97.87	77.39	47.11	0.81
CenterNet [37]	65.15	95.56	84.82	63.95	0.77
Faster R-CNN [16]	72.73	48.48	73.36	2.05	0.58
RetinaNet [20]	75.76	90.91	82.67	45.28	0.83
RFB Net [38]	74.24	89.09	78.38	39.96	0.81
M2Det [39]	75.76	98.04	84.57	39.69	0.85
Our method	78.79	94.55	86.25	37.05	0.86