Point Cloud Completion Network Based on Multiencoders and Residual-Transformer

Fig. 3. Entire completion process of the proposed model. (a) Original point cloud; (b) predicted missing part; (c) complete point cloud after merging Fig.3 (a) and Fig.3 (b); (d) point cloud after sampling; (e) complete point cloud after enhancement; (f) ground truth

Download full size

Fig. 4. Comparison of completion visualization effects of different methods on Shapenet-Part dataset

Download full size

Fig. 5. Comparison of completion visualization effects of ablation study on Shapenet-Part dataset

Download full size

Table 1. $P r e d \to G T / P r e d \leftarrow G T$ of different methods, results are magnified by 10000

View table

Table 1. $P r e d \to G T / P r e d \leftarrow G T$ of different methods, results are magnified by 10000

Category	MSN^［14］	PF-Net^［1］	FCAE^［26］	RP-MLP^［26］	Proposed method
Average	8.395/5.704	11.131/4.289	20.121/15.932	7.514/3.396	5.035/3.294
Airplane	3.113/3.500	4.604/2.997	6.980/7.047	2.892/2.212	2.164/2.641
Bag	15.393/7.942	21.350/7.184	34.774/29.045	14.232/5.748	7.127/3.340
Cap	16.824/5.842	28.984/3.719	52.914/37.094	17.016/2.527	6.950/3.065
Car	8.443/6.991	9.643/3.206	19.982/15.550	8.683/2.516	6.103/2.592
Chair	5.934/4.434	7.803/2.774	16.408/13.733	6.049/2.051	4.107/2.050
Guitar	5.226/3.365	3.195/3.068	3.276/3.185	1.822/2.381	1.693/2.168
Lamp	19.873/9.819	24.143/10.559	39.082/21.319	15.675/9.389	14.506/10.085
Laptop	3.543/3.910	4.879/2.262	9.252/9.623	2.832/1.430	2.110/1.195
Motorbike	6.802/7.178	6.473/5.028	14.094/11.592	6.197/4.162	5.188/3.701
Mug	8.199/6.340	10.538/3.774	25.223/22.119	7.464/3.054	5.481/3.316
Pistol	5.092/6.349	7.613/4.944	10.898/10.115	5.399/3.952	3.350/4.051
Skateboard	3.441/3.766	5.562/2.788	8.981/8.996	2.693/1.994	1.812/2.101
Table	7.257/4.709	9.919/3.454	19.712/17.699	6.722/2.737	4.865/2.513

Table 2. $P r e d \to G T / P r e d \leftarrow G T$ comparison of ablation study, results are magnified by 10000

View table

Table 2. $P r e d \to G T / P r e d \leftarrow G T$ comparison of ablation study, results are magnified by 10000

Method

Airplane

Bag

Cap

Car

Chair

Guitar

Lamp

Proposed method

2.164/

2.461

7.127/

3.340

6.950/

3.065

6.103/

2.592

4.107/

2.050

1.693/

2.168

14.506/

10.085

Proposed method without RT

3.208/

3.356

9.793/

6.232

11.605/

4.262

8.464/

3.730

6.407/

3.413

1.853/

3.214

18.528/

12.221

Method

Laptop

Motorbike

Mug

Pistol

Skateboard

Table

Average

Proposed method

2.110/

1.195

5.188/

3.701

5.481/

3.316

3.350/

4.051

1.812/

2.101

4.865/

2.513

5.035/

3.294

Proposed method without RT

2.975/

2.013

5.748/

5.525

7.922/

4.639

5.621/

6.031

7.391/

4.809

6.560/

4.103

7.390/

4.888

Table 3. $P r e d \to G T / P r e d \leftarrow G T$ comparison of ablation study, results are magnified by 10000

View table

Table 3. $P r e d \to G T / P r e d \leftarrow G T$ comparison of ablation study, results are magnified by 10000

Method

Airplane

Bag

Cap

Car

Chair

Guitar

Lamp

Proposed method

2.164/

2.461

7.127/

3.340

6.950/

3.065

6.103/

2.592

4.107/

2.050

1.693/

2.168

14.506/

10.085

Proposed method w/o Lo

5.055/

7.812

18.436/

4.450

9.441/

2.961

7.561/

3.938

12.959/

4.838

4.534/

3.363

15.442/

14.081

Method

Laptop

Motorbike

Mug

Pistol

Skateboard

Table

Average

Proposed method

2.110/

1.195

5.188/

3.701

5.481/

3.316

3.350/

4.051

1.812/

2.101

4.865/

2.513

5.035/

3.294

Proposed method w/o Lo

3.198/

4.156

3.229/

6.181

10.524/

4.145

4.440/

10.144

5.669/

6.232

7.862/

4.432

8.335/

5.903

Tools

Get Citation

Copy Citation Text

Hui Gao, Zhijing Yang, Wing-Kuen Ling, Jiangzhong Cao, Weijie Li. Point Cloud Completion Network Based on Multiencoders and Residual-Transformer[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0210012

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Dec. 22, 2021

Accepted: Mar. 14, 2022

Published Online: Jan. 6, 2023

The Author Email: Zhijing Yang (yzhj@gdut.edu.cn)

DOI:10.3788/LOP213314

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Pred→GT/Pred←GT of different methods, results are magnified by 10000

Table 1. Pred→GT/Pred←GT of different methods, results are magnified by 10000

Table 2. Pred→GT/Pred←GT comparison of ablation study, results are magnified by 10000

Table 2. Pred→GT/Pred←GT comparison of ablation study, results are magnified by 10000

Table 3. Pred→GT/Pred←GT comparison of ablation study, results are magnified by 10000

Table 3. Pred→GT/Pred←GT comparison of ablation study, results are magnified by 10000

Table 1. $P r e d \to G T / P r e d \leftarrow G T$ of different methods, results are magnified by 10000

Table 1. $P r e d \to G T / P r e d \leftarrow G T$ of different methods, results are magnified by 10000

Table 2. $P r e d \to G T / P r e d \leftarrow G T$ comparison of ablation study, results are magnified by 10000

Table 2. $P r e d \to G T / P r e d \leftarrow G T$ comparison of ablation study, results are magnified by 10000

Table 3. $P r e d \to G T / P r e d \leftarrow G T$ comparison of ablation study, results are magnified by 10000

Table 3. $P r e d \to G T / P r e d \leftarrow G T$ comparison of ablation study, results are magnified by 10000