Table 2 Performance comparison between CSA-Net and current methods on the ACA dataset

From: A coordinate-to-semantic attention network for multi-label ancient Chinese architecture image classification

Methods

mAP(%)

P(%)

R(%)

F1(%)

Inception-v4+softmax38,40

45.99

30.47

15.81

20.79

Inception-v4+LSEP39,40

48.23

36.78

18.84

24.90

Inception-v4+ranking40

53.69

45.81

23.73

31.23

Inception-v4+WARP40,41

53.38

46.33

23.99

31.57

AG2E42

-

31.97

23.59

27.15

VGG19+GCN-A20

63.02

61.02

32.60

42.49

VGG19+GCN-B20

64.03

58.80

34.49

43.48

MSRN44

68.56

63.25

67.67

65.39

GATN43

70.76

64.24

65.81

64.43

DA-GAT45

77.83

61.93

77.76

68.95

MMDSR14

83.26

69.64

74.53

71.31

NOAH(ResNet101)15

85.94

75.63

82.31

81.54

CSA-Net

88.75

77.68

88.00

82.52

  1. Best results are marked in bold, and “-” indicates that the metric is not reported.
Back to article page