Point Cloud Segmentation Algorithm Based on Density Awareness and Self-Attention Mechanism

Table 1. Experimental environment
View table
Table 1. Experimental environment
Environment Version
System Ubuntu 18.04
GPU Nvidia RTX 3090
CUDA 11.3
PyTorch 1.8.0

Table 2. Comparison of IoU of segmentation results for all categories on the S3DIS dataset

View table

Table 2. Comparison of IoU of segmentation results for all categories on the S3DIS dataset

Class	PointCNN^［13］	DSPoint^［25］	BAAF-Net^［26］	LLGF-Net^［24］	Ours
Ceiling	92.31	94.20	92.90	94.10	93.81
Floor	98.24	98.10	97.90	98.20	98.52
Wall	79.41	82.40	82.30	85.10	85.99
Beam	0	0	0	0	0
Column	17.60	19.10	23.10	30.60	40.31
Window	22.77	49.90	65.50	60.60	60.03
Door	62.09	66.20	64.90	73.50	68.17
Table	74.39	85.60	78.50	89.50	84.95
Chair	80.59	78.20	87.50	79.60	91.09
Sofa	31.67	67.90	61.40	72.30	57.66
Bookcase	66.67	59.00	70.70	63.10	77.21
Board	62.05	62.30	68.70	79.80	80.69
Clutter	56.74	59.90	57.20	57.60	59.93

Table 3. Comparison of evaluation indicators of segmentation results on the S3DIS dataset

View table

Table 3. Comparison of evaluation indicators of segmentation results on the S3DIS dataset

Method	mIoU /%	mAcc /%
PointNet^［11］	41.09	48.98
SegCloud^［8］	48.92	57.35
PointWeb^［27］	60.28	66.64
SegGCN^［28］	63.60	70.44
PointCNN^［13］	57.26	63.86
KPConv^［15］	67.10	72.80
LLGF-Net^［24］	68.00	74.40
Ours	69.11	75.09

Table 4. Comparison of evaluation indicators of segmentation results on the ScanNet dataset

View table

Table 4. Comparison of evaluation indicators of segmentation results on the ScanNet dataset

Method	mIoU /%	mAcc /%
PointNet^［11］	14.69	19.90
PointNet++^［12］	34.26	43.77
PointConv^［29］	55.60	—
RSNet^［30］	39.35	48.37
FG-Net^［31］	69.00	—
PointCNN^［13］	43.70	57.90
KPConv^［15］	68.40	—
Ours	72.52	80.63

Table 5. Comparison of IoU of segmentation results for all categories on the ScanNet dataset

View table

Table 5. Comparison of IoU of segmentation results for all categories on the ScanNet dataset

Class	PointNet++^［12］	PointCNN^［13］	PointASNL^［32］	KPConv^［15］	Ours
Wall	77.48	74.50	80.60	81.90	85.62
Floor	92.50	90.70	95.10	93.50	95.10
Bed	51.32	56.10	78.10	75.80	80.66
Chair	64.55	68.80	83.00	81.40	89.75
Sofa	52.27	60.10	75.10	78.50	81.52
Table	46.60	55.30	55.30	61.40	72.29
Door	2.02	7.50	53.70	59.40	68.74
Desk	12.69	28.80	47.40	60.50	62.86
Sink	30.23	41.90	67.50	69.00	64.58
Toilet	31.37	73.40	81.60	88.20	92.20
Cabinet	23.81	26.40	65.50	64.70	65.50
Picture	0	1.30	27.90	18.10	36.56
Counter	20.04	22.50	47.10	47.30	61.26
Curtain	32.97	36.10	76.90	77.20	74.09
Window	3.56	11.00	70.30	63.20	68.24
Bathtub	42.72	73.50	70.30	84.70	87.70
Bookshelf	52.93	38.90	75.10	78.40	81.57
Refridgerator	18.51	43.40	63.50	58.70	64.06
Shower curtain	27.43	40.60	69.80	80.50	60.46
Other furniture	2.20	23.60	47.50	45.00	57.58

Table 6. Experimental results of different K values
View table
Table 6. Experimental results of different K values
K value KNN Adaptive KNN
mIoU /% mAcc /% mIoU /% mAcc /%
8 65.10 72.01 65.54 72.76
16 68.62 74.38 69.11 75.09
32 68.11 74.03 68.83 74.36
64 65.93 74.52 66.79 73.15

Table 7. Ablation experimental results of different modules
View table
Table 7. Ablation experimental results of different modules
Algorithm Mod.1 Mod.2 mIoU /% OA /% mAcc /%
0 × × 66.58 — 73.00
1 √ × 67.26 89.70 73.13
2 × √ 66.79 89.67 73.57
3 √ √ 69.11 90.99 75.09

Table 8. Ablation experimental results of different submodules
View table
Table 8. Ablation experimental results of different submodules
Algorithm Smod.1 Smod.2 Smod.3 mIoU /% OA /% mAcc /%
0 × × √ 66.74 89.65 73.23
1 √ × √ 67.37 89.42 73.61
2 × √ √ 67.52 89.61 73.88
3 √ √ √ 69.11 90.99 75.09
4 √ √ × 68.32 89.86 74.69

Tools

Get Citation

Copy Citation Text

Bin Lu, Yawei Liu, Yuhang Zhang, Zhenyu Yang. Point Cloud Segmentation Algorithm Based on Density Awareness and Self-Attention Mechanism[J]. Laser & Optoelectronics Progress, 2024, 61(8): 0811004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Imaging Systems

Received: Jun. 2, 2023

Accepted: Jul. 31, 2023

Published Online: Mar. 13, 2024

The Author Email: Yawei Liu (1425655356@qq.com)

DOI:10.3788/LOP231450

CSTR:32186.14.LOP231450

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Experimental environment

Table 1. Experimental environment

Table 2. Comparison of IoU of segmentation results for all categories on the S3DIS dataset

Table 2. Comparison of IoU of segmentation results for all categories on the S3DIS dataset

Table 3. Comparison of evaluation indicators of segmentation results on the S3DIS dataset

Table 3. Comparison of evaluation indicators of segmentation results on the S3DIS dataset

Table 4. Comparison of evaluation indicators of segmentation results on the ScanNet dataset

Table 4. Comparison of evaluation indicators of segmentation results on the ScanNet dataset

Table 5. Comparison of IoU of segmentation results for all categories on the ScanNet dataset

Table 5. Comparison of IoU of segmentation results for all categories on the ScanNet dataset

Table 6. Experimental results of different K values

Table 6. Experimental results of different K values

Table 7. Ablation experimental results of different modules

Table 7. Ablation experimental results of different modules

Table 8. Ablation experimental results of different submodules

Table 8. Ablation experimental results of different submodules