Attention-Enhanced Multiscale Dual-Feature Point Cloud Completion Method

Tianli Wang; Zequn Zhang; Jie Chen; Dunbing Tang; Lanlan Jiang; Lingfei Qian

doi:10.3788/LOP250484

Laser & Optoelectronics Progress, Volume. 62, Issue 16, 1637004(2025)

Attention-Enhanced Multiscale Dual-Feature Point Cloud Completion Method

Tianli Wang¹, Zequn Zhang^1、*, Jie Chen¹, Dunbing Tang¹, Lanlan Jiang², and Lingfei Qian³

¹College of Mechanical and Electrical Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, Jiangsu , China

²Nanjing Nanrui Information Communication Technology, Nanjing 210003, Jiangsu , China

³College of Economics and Management, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, Jiangsu , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Framework diagram of multiscale dual-feature point cloud completion network based on OAPTransform. (a) Dual-feature encoder; (b) pyramidal decoder

Download full size

Fig. 2. Schematic diagram of OAPMLP feature extraction layer

Download full size

Fig. 3. Schematic of the local magnification of C-DGCNN

Download full size

Fig. 4. Structure of the offset-position attention layer OAPTransform

Download full size

Fig. 5. Schematic diagram of point cloud completion at different scales

Download full size

Fig. 6. Structure of the discriminator

Download full size

Fig. 7. Loss value curve

Download full size

Fig. 8. Comparison of the completion effect of the proposed algorithm and PF-Net algorithm on some categories of ShapeNet dataset

Download full size

Table 1. Time consumed by the main methods at three scales
View table
Table 1. Time consumed by the main methods at three scales
Number of dimensions Time /s
C-DGCNN OAPTransform
2048 0.016 0.015
1024 0.006 0.005
512 0.003 0.002

Table 2. Memory consumed by the main methods at three scales
View table
Table 2. Memory consumed by the main methods at three scales
Number of dimensions Memory /MB
C-DGCNN OAPTransform
2048 894.34 1504.01
1024 447.84 496.01
512 223.65 184.01

Table 3. Comparison of CD values on the PCN dataset

View table

Table 3. Comparison of CD values on the PCN dataset

Model	CD /10³
Model	Average	Plane	Car	Chair	Lamp	Table	Boat	Couch	Cabinet
FoldingNet	13.80	9.64	12.62	15.58	14.41	12.97	14.45	15.03	15.68
TopNet	11.39	9.28	10.34	13.13	12.91	10.45	10.68	11.32	12.97
PoinTr	8.18	4.83	8.34	9.16	7.86	7.78	8.35	8.13	10.89
PCN	10.84	5.90	9.05	12.44	13.53	9.84	9.67	10.32	15.98
PF-Net	7.06	3.52	6.55	7.94	5.28	7.13	7.53	8.78	9.74
Ours	6.24	3.01	5.76	4.99	7.49	5.16	6.95	8.25	8.31

Table 4. Comparison of CD values on the Completion3D dataset

View table

Table 4. Comparison of CD values on the Completion3D dataset

Model	CD /10³
Model	Average	Plane	Car	Chair	Lamp	Table	Boat	Couch	Cabinet
FoldingNet	18.06	12.38	15.64	20.61	19.97	18.67	15.69	21.13	20.35
TopNet	14.66	7.68	13.72	20.07	14.52	16.76	9.89	16.27	18.36
PoinTr	8.37	3.57	9.15	8.40	8.69	7.54	9.17	11.15	9.26
PCN	11.09	7.26	10.05	15.61	10.88	11.44	8.57	12.48	12.39
PF-Net	7.63	4.05	6.08	8.68	8.49	6.69	6.32	10.11	10.65
Ours	7.09	4.12	6.35	8.57	6.96	5.03	6.51	9.98	9.18

Table 5. Comparison results of ablation experiments
View table
Table 5. Comparison results of ablation experiments
Model MDFF OAPTransform CD
A 6.08
B √ 5.51
C √ 5.45
D √ √ 5.28

Tools

Get Citation

Copy Citation Text

Tianli Wang, Zequn Zhang, Jie Chen, Dunbing Tang, Lanlan Jiang, Lingfei Qian. Attention-Enhanced Multiscale Dual-Feature Point Cloud Completion Method[J]. Laser & Optoelectronics Progress, 2025, 62(16): 1637004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites