Table 2 PotsdamIRRG \(\rightarrow\) VaihingenIRRG quantitative results for cross-domain segmentation. Significant values are in bold.

From: Unsupervised domain adaptation for remote sensing semantic segmentation with the 2D discrete wavelet transform

Task

Methods

 

Clutter

Imp. surfaces

Car

Tree

Low vegetation

Building

Overall

PotsdamIRRG

\(\downarrow\)

VaihingenIRRG

Source-Only

IOU\(\%\)

1.83

30.48

17.93

52.46

16.63

59.76

29.85

F1\(\%\)

3.23

46.15

26.79

68.61

28.22

74.57

41.76

CycleGAN

IOU\(\%\)

3.11

43.73

9.85

58.44

33.28

49.54

32.99

F1\(\%\)

4.80

60.3

17.25

73.58

49.49

65.9

45.22

DualGAN

IOU\(\%\)

3.58

52.67

19.74

62.24

39.8

62.83

40.14

F1\(\%\)

5.66

68.53

32.19

76.63

56.53

76.86

52.73

FADA

IOU\(\%\)

10.83

62.26

39.82

64.13

43.22

72.22

48.75

F1\(\%\)

19.54

76.74

56.96

78.18

60.35

83.87

62.6

ResidualGAN

IOU\(\%\)

7.76

70.09

50.27

60.53

46.19

77.71

52.09

F1\(\%\)

11.42

82.28

66.44

75.22

62.78

87.38

64.25

MemoryAdaptNet

IOU\(\%\)

10.86

67.82

44.30

50.27

46.06

76.92

49.37

F1\(\%\)

19.59

80.82

61.40

66.91

63.07

86.95

63.12

MBATA-GAN

IOU\(\%\)

3.78

64.92

38.51

51.33

40.19

74.67

45.57

F1\(\%\)

7.29

78.73

55.61

67.85

57.34

85.50

58.72

Ours

IOU\(\%\)

7.95

72.64

56.67

65.63

49.92

83.42

56.04

F1\(\%\)

11.15

84.05

72.05

79.16

66.33

90.92

67.28