COGCN Model Suitable for Point Cloud Classification and Segmentation Tasks

Weichao Chen; Lingchen Zhang; Ronghua Chi; Zhenbo Yang; Qi Liu; Hongxu Li

doi:10.3788/LOP242046

Laser & Optoelectronics Progress, Volume. 62, Issue 10, 1015005(2025)

COGCN Model Suitable for Point Cloud Classification and Segmentation Tasks

Weichao Chen¹, Lingchen Zhang², Ronghua Chi^2,3, Zhenbo Yang², Qi Liu¹, and Hongxu Li^2,3、*

Author Affiliations

¹Wuxi Institute of Technology, Nanjing University of Information Science and Technology, Wuxi 214101, Jiangsu , China

²School of Electronic and Information Engineering, Nanjing University of Information Science and Technology, Nanjing 210044, Jiangsu , China

³Jiangsu Province Engineering Research Center of Integrated Circuit Reliability Technology and Testing System, Wuxi University, Wuxi 214105, Jiangsu , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(15)

Fig. 1. Structure of CSEConv module

Download full size

Fig. 2. Structure of offset attention module

Download full size

Fig. 3. Structure of COGCN classification network

Download full size

Fig. 4. Structure of COGCN segmentation network

Download full size

Fig. 5. Structure of spatial transformation network

Download full size

Fig. 6. Visualization comparison of component segmentation by different methods

Download full size

Fig. 7. Visualization comparison of semantic segmentation by different methods

Download full size

Fig. 8. Visualization comparison of ablation experiments

Download full size

Fig. 9. Comparison of OA under different numbers of point clouds by different methods

Download full size

Table 1. OA of different models on the ModelNet40 dataset

View table

Table 1. OA of different models on the ModelNet40 dataset

Mechanism	Model	Input	OA
MLP	PointNet	1000 points+normal	89.2
	PointNet++	1000 points	90.7
	PointNN^［17］	1000 points	81.8
	MANet^［18］	1000 points	92.5
Conv	PointCNN	1000 points	92.5
	PointWeb	1000 points+normal	92.3
	PointConv	1000 points+normal	92.5
	KPConv	1000 points	92.9
	Hybrid-CNN^［19］	1000 points	92.6
	SACNN	1000 points	93.2
Attention	PCT^［20］	1000 points	93.2
Attention	3DCTN^［21］	1000 points	92.7
Graph	SpecGCN	1000 points	92.1
	DGCNN	1000 points	92.9
	GCN3D	1000 points	93.0
	Proposed	1000 points	93.2

Table 2. Segmentation IoU results of different models on the ShapeNet dataset

View table

Table 2. Segmentation IoU results of different models on the ShapeNet dataset

Category	PointNet	PointNet++	PCNN^［22］	DGCNN	3D-GCN^［14］	DT-Net^［23］	PointCutMix^［24］	AG-Net^［25］	GCN3D	SACNN	Proposed
mIoU	83.7	85.1	85.1	85.2	85.1	85.6	85.5	85.4	85.2	85.4	86.1
Airplane	83.4	82.4	82.4	84.0	83.1	83.0	82.6	84.1	83.3	84.5	85.9
Bag	78.7	79.0	80.1	83.4	84.0	81.4	85.9	83.2	78.8	76.4	84.2
Cap	82.5	87.7	85.5	86.7	86.6	84.3	83.7	86.0	83.2	81.8	87.4
Car	74.9	77.3	79.5	77.8	77.5	78.4	78.3	78.8	77.5	78.4	81.3
Chair	89.6	90.8	90.8	90.6	90.3	90.9	90.7	90.6	90.8	91.6	91.4
Earphone	73.0	71.8	73.2	74.7	74.1	74.3	72.5	76.9	75.8	76.8	78.3
Guitar	91.5	91.0	91.3	91.2	90.9	91.0	90.9	91.9	90.8	91.6	92.4
Knife	85.9	85.9	86.0	87.5	86.4	87.3	87.7	88.4	86.0	88.5	87.1
Lamp	80.8	83.7	85.0	82.8	83.8	84.7	84.3	82.3	83.8	85.5	84.2
Laptop	95.3	95.3	95.7	95.7	95.6	95.6	95.3	96.0	95.2	95.8	96.1
Motor	65.2	71.6	73.2	66.3	66.8	69.0	70.7	65.5	66.4	60.6	72.5
Mug	93.0	94.1	94.8	94.9	94.8	94.4	95.1	93.7	93.4	92.7	96.0
Pistol	81.2	81.3	83.3	81.1	81.3	82.5	82.4	84.2	81.3	80.5	83.1
Rocket	57.9	58.7	51.0	63.5	59.6	59.0	62.3	64.2	51.2	56.2	50.8
Skateboard	72.8	76.4	75.0	74.5	75.7	76.4	74.9	76.8	67.1	73.7	77.0
Table	80.6	82.6	81.8	82.6	82.8	83.5	83.4	80.6	83.0	82.8	83.1

Table 3. Semantic segmentation results of different models on the S3DIS dataset
View table
Table 3. Semantic segmentation results of different models on the S3DIS dataset
Method mIoU OA
PointNet 47.6 78.5
PointNet++ 53.5 —
SegCloud 48.9 —
ShapeContextNet^［26］ 52.7 81.6
DGCNN 56.1 84.1
RSNet^［27］ 51.9 —
AGNet 59.6 85.9
SDANet^［28］ 59.6 —
Proposed 60.3 85.7

Table 4. Ablation experimental results for classification network
View table
Table 4. Ablation experimental results for classification network
CSEConv Offset attention OA
√ — 91.7
— √ 90.3
√ √ 93.2

Table 5. Ablation experimental results for segmentation network
View table
Table 5. Ablation experimental results for segmentation network
CSEConv Offset attention mIoU
√ — 85.7
— √ 84.6
√ √ 86.1

Table 6. Comparison of OA under different numbers of point clouds by different methods
View table
Table 6. Comparison of OA under different numbers of point clouds by different methods
Number of point cloud COGCN DGCNN PointNet
256 91.0 47.1 81.8
512 91.7 86.5 86.7
768 92.4 91.9 88.5
1024 93.2 92.9 89.2

Tools

Get Citation

Copy Citation Text

Weichao Chen, Lingchen Zhang, Ronghua Chi, Zhenbo Yang, Qi Liu, Hongxu Li. COGCN Model Suitable for Point Cloud Classification and Segmentation Tasks[J]. Laser & Optoelectronics Progress, 2025, 62(10): 1015005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites