Skip to main content

Table 3 Semantic segmentation results on the S3DIS dataset (no RGB)

From: Pseudo-labelling-aided semantic segmentation on sparsely annotated 3D point clouds

Method

OA (%)

AF (%)

Per-class I-scores(%)

   

(i)

(ii)

(iii)

(iv)

(v)

(vi)

(vii)

(viii)

(ix)

(x)

(xi)

Pointwise [1]

18.8

16.8

15.3

0.1

18.0

27.5

7.5

8.4

49.9

18.3

18.4

21.0

0.9

CRF-reg [22]

35.6

25.6

−∗1

−∗2

48.9

53.3

2.6

−∗1

94.9

11.3

39.2

31.2

0.2

Seg-aided [17]

42.1

34.0

52.6

−∗2

72.4

47.1

−∗2

−∗2

94.1

26.6

49.5

31.0

1.1

Supervised baseline

34.8

30.5

27.5

47.9

23.5

26.7

19.2

34.2

60.2

25.9

29.1

24.7

16.1

Ours no kdist

31.7

28.5

32.2

46.2

31.8

31.8

22.9

36.5

28.4

32.5

29.2

4.2

18.0

Ours with kdist

49.8

44.3

31.7

73.8

35.3

38.1

31.1

59.2

74.1

40.9

46.7

20.9

29.0

*1 no instances of class are predicted correctly; precision=0, recall=0 - F-score undefined, taken to be 0 for the average

*2 no instances of class are predicted at all; precision undefined, recall=0 - F-score undefined, taken to be 0 for average

  1. OA overall accuracy, AF average F-score. Classes are as follows: (i) door, (ii) floor, (iii) table, (iv) window, (v) beam, (vi) book-case, (vii) ceiling, (viii) clutter, (ix) chair, (x) board, (xi) wall