Table 5 Detection results with different settings of deformable convolution, squeeze-and-excitation attention, and enhanced pixel-wise squeeze-and-excitation attention modules

From: Fine grained representation learning for low resource Yi script detection and dataset construction

Backbone

DConv

SE

EPSA

F (%)

P (%)

R (%)

ResNet-18

   

90.3

90.2

90.5

ResNet-18

✓

  

91.0

91.3

90.7

ResNet-18

 

✓

 

91.9

92.4

91.7

ResNet-18

  

✓

93.5

97.2

90.8

ResNet-18

✓

✓

 

92.7

93.2

92.3

ResNet-18

✓

✓

✓

94.7

98.4

91.3

  1. "DConv” indicates deformable convolution, “SE” indicates Squeeze-and-Excitation Attention, and “EPSA” indicates Enhanced Pixel-wise Squeeze-and-Excitation Attention. The best and second-best results are Boldfaced and Underlined. “F”, “P”, and “R” indicate f-score, precision, and recall, respectively.
Back to article page