Table 2 Segmentation performance comparison of our method with the state-of-the-art methods on the MT-Defect dataset.

FCN

99.48

20.04

49.92

37.60

85.49

80.36

62.15

PSPNet

99.30

38.90

61.40

52.80

83.10

70.50

67.66

DeepLabV3+

98.52

22.26

52.79

37.42

53.23

23.42

47.97

EMANet

99.57

23.40

71.75

44.49

84.90

81.37

67.40

FPN

98.94

0.28

5.46

1.32

27.63

63.26

32.82

ICNet

99.00

1.26

37.14

24.45

82.7

52.52

49.51

CGNet

98.86

1.08

22.79

0.47

52.09

54.51

38.30

STDC2

98.94

0.66

30.27

1.12

65.69

55.09

41.96

STDC1

99.13

1.10

28.10

1.33

66.57

62.97

43.20

BiSeNetV1

99.38

14.83

65.02

31.11

86.12

74.09

61.08

BiSeNetV2

99.32

11.28

57.53

37.20

83.28

72.74

60.22

Fast-SCNN

99.23

1.56

52.97

1.81

80.27

68.07

50.65

DDRNet

99.47

34.98

63.38

37.29

83.93

77.62

66.11

FDSNet*

–

–

–

–

–

–

63.9

RTFormer

99.51

32.42

64.13

36.85

84.21

78.39

65.91

Trans4Trans

99.28

34.47

63.97

36.11

85.01

76.08

65.82

PIDNet

98.58

58.77

55.93

38.60

79.10

66.44

66.23

SegFormer

98.03

54.62

61.21

38.51

81.89

75.69

68.32

SeaFormer

99.51

57.68

62.53

37.96

81.21

74.15

68.84

SCTNet

99.41

58.81

60.78

39.81

80.8

75.91

69.25

ConvNext

99.57

58.26

74.84

40.69

89.40

83.69

74.41

SegNext

99.44

54.24

69.32

48.60

87.70

74.89

72.36

PoolFormer

99.44

57.80

73.20

46.96

88.67

75.39

73.58

DDSNet

99.56

55.31

71.75

61.86

88.72

81.86

76.51

Ours

99.67

62.23

79.33

70.85

90.91

85.74

81.45