Table 2 Performance of different components in the DY-YOLO Network.

From: Smartphone screen surface defect detection using dynamic large separable kernel attention and multi-scale feature bi-directional path aggregation network

Dynamic-C2f

Dynamic-LSKA

HSF-BPAN

Avg_F1

Avg_Precision (%)

Avg_Recall (%)

mAP@0.5 (%)

mAP@0.5:0.95 (%)

Dataset

✔

  

0.9835

98.7

98.0

98.6

70.2

MSD

 

✔

 

0.9865

99.0

98.3

98.8

70.3

  

✔

0.9840

98.8

98.0

98.8

70.2

✔

 

✔

0.9785

98.5

97.2

98.4

70.7

 

✔

✔

0.9845

98.8

98.1

98.5

70.6

✔

✔

✔

0.9895

99.4

98.5

99.3

70.9

✔

  

0.4529

54.2

38.9

41.2

18.7

SSGD

 

✔

 

0.4503

47.5

42.8

41.9

19.2

  

✔

0.4183

43.6

40.2

40.2

19.0

✔

 

✔

0.4361

53.3

36.9

39.4

17.1

 

✔

✔

0.4337

51.8

37.3

41.7

18.6

✔

✔

✔

0.4919

52.6

46.2

46.0

20.2