Skip to main content

Table 5 The performances (%) of models with different attention methods on our LDWBC test set

From: Accurate classification of white blood cells by coupling pre-trained ResNet and DenseNet with SCAM mechanism

Attention method

OA

AP

AR

AF1

No

97.55

92.15

93.36

92.61

CD

SE [51, 52]

97.75

91.87

96.21

93.92

ECA [53]

97.37

90.29

93.05

91.24

CAM [30]

98.06

93.51

95.06

94.17

SD

SAM [30]

97.75

91.78

95.49

93.56

TD

CAM // SAM [30]

97.73

91.73

94.27

92.90

CAM + SAM (CSAM) [30]

97.68

89.49

97.43

93.11

SAM + CAM (SCAM) [30]

97.84

91.61

96.38

93.82

  1. Best results are in bold; CD channel dimension, SD spatial dimension, TD two dimensions, // parallel, + sequential