Table 7 Model performance on the ArCH dataset was evaluated using FB-IoU (%)

From: Learning discriminative universal background knowledge for few-shot point cloud semantic segmentation of architectural cultural heritage

Method

1-way

2-way

 

1-shot

5-shot

1-shot

5-shot

 

S0

S1

Avg

S0

S1

Avg

S0

S1

Avg

S0

S1

Avg

ProtoNet41

40.18

38.11

39.14

48.15

46.91

47.53

22.91

24.32

23.61

37.63

34.77

36.20

AttProtoNet41

39.11

33.62

36.36

45.90

44.58

45.24

24.75

24.04

24.39

38.61

37.53

38.07

MPTI23

44.28

38.69

41.48

49.69

50.73

50.21

24.12

25.91

25.01

40.19

36.70

38.44

AttMPTI23

43.89

39.17

41.53

51.22

47.82

49.52

25.80

23.65

24.72

30.87

33.57

32.22

QGPA28

47.01

41.92

44.46

57.88

54.13

56.00

45.52

38.07

41.79

50.86

49.38

50.12

QGE29

44.41

38.01

41.21

43.12

38.45

40.78

30.31

31.08

30.69

31.53

30.73

31.13

DPA32

44.70

50.46

47.58

52.83

57.73

55.28

35.49

31.71

33.59

46.87

44.94

45.90

Ours

51.21

54.02

52.61

59.95

58.08

59.01

48.17

39.28

43.72

53.90

51.12

52.51

  1. Si denotes the split i, which is used to test the model. The best results are masked in bold.
Back to article page