356
V. L. Kondarattsev et al.
Table 24.1 Comparison of metrics for deep learning models for semantic segmentation of point
clouds
Architecture
Dataset
Overall accuracy
Mean accuracy
Mean intersection
over union
SEGCloud
S3DIS
–
57.35
48.92
RSNet
S3DIS
–
59.42
51.93
RSNet
ScanNet
–
48.37
39.35
RSNet
ShapeNet-part
–
–
84.9
LDGCNN
ModelNet40
92.9
90.3
–
LDGCNN
ShapeNet-part
–
–
85.1
SpiderCNN
ModelNet40
92.4
–
–
SpiderCNN
ShapeNet-part
–
–
85.3
PointNet++
ModelNet40
90.7
–
–
PointNet++
ScanNet (with
voxelization)
84.5
–
–
MVCNN
ModelNet40
90.1
–
–
VoxNet
ModelNet40
–
83
–
SO-Net
ShapeNet-part
–
–
84.6
SO-Net
ModelNet40
90.8
–
–
RGCNN
ShapeNet-part
–
–
84.3
RGCNN
ModelNet40
90.5
87.3
–
3DMAX-Net
S3DIS
79.5
–
47.5
PointSIFT
S3DIS
88.72
–
70.23
PointSIFT
ScanNet
86.2
–
41.5
PointGrid
ModelNet40
92.0
88.9
–
PointCNN
ModelNet40
(pre-aligned)
92.5
88.8
–
PointCNN
ModelNet40
(unaligned)
92.2
88.1
–
PointCNN
ScanNet
85.1
–
–
PointCNN
S3DIS
88.1
–
65.39
PointCNN
ShapeNet-part
–
–
84.6
GAPNet
ModelNet40
92.4
89.7
–
GAPNet
ShapeNet-part
92
84.7
–
A-CNN
ModelNet40
92.6
90.3
–
A-CNN
ScanNet
85.4
–
–
A-CNN
S3DIS
87.3
–
–
A-CNN
ShapeNet-part
86
–
–
3P-RNN
S3DIS
86.9
73.6
56.3
(continued)