A Model for Surface Defect Detection of Industrial Products Based on Attention Augmentation

Table 1

Comparison with state-of-the-art methods in three different datasets.

Methods	(Internal) mAP	(RSDDs) mAP	(BSData) mAP	(NEU-DET) mAP	(NRSD-MN) mAP
YOLOv3	91	76.60	77.63	54.7	63.09
YOLOv4	95.97	79.27	81.38	70.35	75.4
YOLOv5l	95.53	80.43	80.39	72.7	78.71
YOLOv5x	98.45	85.45	81.79	74.07	80.15
Our method	98.52	86.74	82.00	74.67	81.09