Table 6 Comparative analysis of different attention mechanisms integrated into the DY-YOLO architecture.

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

Model variant

Attention module

Params (M)

FLOPs (G)

MSD

SSGD

mAP@0.5

mAP@0.5:0.95

mAP@0.5

mAP@0.5:0.95

Variant A

SE

4.1

9.6

98.7

70.4

44.2

19.5

Variant B

CBAM

4.2

9.6

98.5

70.8

42.1

18.8

Variant C

Deformable_LKA

5.7

10.9

98.7

70.1

43.5

19.1

Variant D

LSKA

4.2

9.7

98.7

70.2

45.1

19.8

DY-YOLO (Ours)

Dynamic-LSKA

3.0

5.8

99.3

70.9

46.0

20.2

Back to article page