Table 2 Ablation study without or with different attention mechanisms.

From: Enhanced photovoltaic panel defect detection via adaptive complementary fusion in YOLO-ACF

Model

P(%)\(\uparrow\)

R(%)\(\uparrow\)

mAP50(%)\(\uparrow\)

mAP50-95(%)\(\uparrow\)

Params(M)\(\downarrow\)

FLOPs(G)\(\downarrow\)

Weight(MB)\(\downarrow\)

Time(ms)\(\downarrow\)

FPS\(\uparrow\)

Without

81.7

86.8

89.1

60.4

7.0

15.9

14.5

7.1

140

C3CA

88.6

78.8

79.7

54.5

8.4

17.1

17.3

9.2

108

CA

81.3

89.6

91.2

62.0

5.9

14.9

12.2

7.4

135

C3CBAM

92.3

76.6

91.3

60.6

7.1

15.9

14.5

9.0

111

CBAM

89.1

79.5

81.9

56.6

7.6

16.7

15.6

6.6

151

C3ACF

89.0

92.3

92.7

62.4

7.3

15.2

15.0

9.4

106

ACF

84.5

92.3

92.6

62.1

6.1

14.2

12.7

6.8

147

  1. 1. Highlight the best performance in the table using bold font, and underline the second best performance
  2. 2. The \(\uparrow\) indicates the larger value achieved, the better performance is, while the \(\downarrow\) indicates. the smaller, the better