Semantic Segmentation of Three-Dimensional Point Cloud Based on Spatial Attention and Shape Feature

Wen Hao; Hongxiao Wang; Yang Wang

doi:10.3788/LOP202259.0828004

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 8 >Page 0828004 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 8, 0828004 (2022)

Semantic Segmentation of Three-Dimensional Point Cloud Based on Spatial Attention and Shape Feature

Wen Hao^1、2、*, Hongxiao Wang^1、2, and Yang Wang^1、2

Author Affiliations

¹School of Computer Science and Engineering, Xi'an University of Technology, Xi'an , Shaanxi 710048, China

²Shaanxi Key Laboratory for Network Computing and Security Technology, Xi'an , Shaanxi 710048, China

show less

DOI: 10.3788/LOP202259.0828004 Cite this Article Set citation alerts

Wen Hao, Hongxiao Wang, Yang Wang. Semantic Segmentation of Three-Dimensional Point Cloud Based on Spatial Attention and Shape Feature[J]. Laser & Optoelectronics Progress, 2022, 59(8): 0828004 Copy Citation Text

show less

Fig. 1. AMFF-DGCNN structure

Download full size

Fig. 2. Spatial attention module

Download full size

Fig. 3. Component segmentation effects of DGCNN,AMFF-DGCNN on ShapeNet dataset. (a) Ground truth; (b) DGCNN;(c) AMFF-DGCNN

Download full size

Fig. 4. Semantic segmentation results of DGCNN, AMFF-DGCNN on S3DIS dataset. (a) Input point cloud; (b) ground truth; (c) DGCNN segmentation results; (d) AMFF-DGCNN segmentation results

Download full size

Algorithm	mACC /%	OA /%
VoxNet^［22］	83.0	85.9
PointNet^［6］	86.0	89.2
PointNet++^［7］		90.7
ECC^［23］	83.2	87.4
SO-Net^［24］	87.2	90.9
RGCNN^［25］	87.3	90.5
DGCNN^［13］	88.9	91.2
Proposed algorithm	89.2	91.8

Table 1. Comparison of classification experimental results on ModelNet40 dataset

Algorithm	mIOU	Aero	Bag	Cap	Car	Chair	Earphone	Guitar	Knife	Lamp	Laptop	Motor	Mug	Pistol	Rocket	Skateboard	Table
PointNet^［6］	83.7	83.4	78.7	82.5	74.9	89.6	73.0	91.5	85.9	80.8	95.3	65.2	93.0	81.2	57.9	72.8	80.6
Kd-Net^［26］	82.3	80.1	74.6	74.3	70.3	88.6	73.5	90.2	87.2	81.0	94.9	57.4	86.7	78.1	51.8	69.9	80.3
GAPNet^［27］	84.7	84.2	84.1	88.8	78.1	90.7	70.1	91.0	87.3	83.1	96.2	65.9	95.0	81.7	60.7	74.9	80.8
SCN^［28］	84.6	83.8	80.8	83.5	79.3	90.5	69.8	91.7	86.5	82.9	96.0	69.2	93.8	82.5	62.9	74.4	80.8
RSNet^［29］	84.9	82.7	86.4	84.1	78.2	90.4	69.3	91.4	87.0	83.5	95.4	66.0	92.6	81.8	56.1	75.8	82.2
DGCNN^［13］	84.7	83.6	85.7	84.8	78.2	90.5	75.4	91.1	87.2	82.4	95.6	62.9	94.5	80.7	63.6	75.4	81.8
Proposed algorithm	84.9	84.1	80.2	85.5	78.7	90.8	74.5	90.7	88.3	82.1	95.7	67.0	93.8	83.0	56.3	73.7	81.9

Table 2. Comparison of experimental results of component segmentation on ShapeNet dataset

Algorithm	OA	mIOU	Ceiling	Floor	Wall	Beam	Column	Window	Door	Chair	Table	Bookcase	Sofa	Board	Clutter
PointNet^［6］		41.09	88.80	97.33	69.80	0.05	3.92	46.26	10.76	52.61	58.93	40.28	5.85	26.38	33.22
MS3_DVS^［30］		46.32	79.03	88.07	53.55	0.00	20.47	29.01	37.29	68.84	63.72	47.44	61.62	16.50	36.64
SegCloud^［31］		48.92	90.06	96.05	69.86	0.00	18.37	38.35	23.12	75.89	70.40	58.42	40.88	12.96	41.60
DGCNN^［13］	83.31	47.56	92.77	97.54	74.86	0.00	11.77	50.72	23.71	66.38	69.58	8.58	48.72	31.58	42.06
Proposed algorithm	84.60	51.56	92.81	97.78	78.12	0.00	24.91	51.92	31.05	69.11	73.91	16.21	52.68	39.51	42.18

Table 3. Semantic segmentation results of Area 5 on S3DIS dataset

Algorithm	OA	mIOU	Ceiling	Floor	Wall	Beam	Column	Window	Door	Chair	Table	Bookcase	Sofa	Board	Clutter
PointNet^［6］	78.60	47.60	88.00	88.70	69.30	42.40	23.10	47.50	51.60	42.00	54.10	38.20	9.60	29.40	35.20
RSNet^［29］		56.47	92.48	92.83	78.56	32.75	34.37	51.62	68.11	59.72	60.13	16.42	50.22	44.85	52.03
SPG^［32］	82.90	54.06	92.17	95.00	71.91	33.46	15.03	46.53	60.92	65.05	69.45	56.82	38.21	6.86	51.29
DGCNN^［13］	84.32	56.86	92.78	93.74	76.42	53.10	35.59	56.46	61.24	64.16	51.43	15.93	48.26	43.06	46.99
Octant-CNN^［33］	84.6	58.3	92.1	94.5	76.3	48.9	30.8	56.9	62.9	65.8	55.5	28	48.1	50.3	48.4
Contextual attention CNN^［34］	85.2	57.6
Proposed algorithm	85.61	59.90	92.86	94.73	78.27	54.04	43.25	59.06	62.60	67.39	59.08	18.28	51.29	47.40	50.51

Table 4. Semantic segmentation results of 6-fold cross-validation on S3DIS dataset

Model	Point feature	Geometry feature	Spatial attention	OA / %
A				83.3
B	√			84.15
C	√	√		84.3
D	√	√	√	84.6

Table 5. Module importance analysis

Wen Hao, Hongxiao Wang, Yang Wang. Semantic Segmentation of Three-Dimensional Point Cloud Based on Spatial Attention and Shape Feature[J]. Laser & Optoelectronics Progress, 2022, 59(8): 0828004

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information