Optics and Precision Engineering, Volume. 30, Issue 10, 1189(2022)
Single-view 3D object reconstruction based on NFFD and graph convolution
Figures & Tables(14)
Fig. 5. Comparison of 3D reconstruction results of ours， 3D-LMNet，Occupancy networks，DISN and PSGN
Fig. 6. Comparison of the 3D reconstruction results of ours， Pixel2Mesh， DISN and PSGN
Fig. 7. Comparison of 3D reconstruction results of ours， VGG-Pixel2Mesh，Occupancy networks，DISN and PSGN
Fig. 8. Influence of mixed attention module on local details of point cloud
Fig. 9. Real scene 3D reconstruction effect of the proposed method
Table 1. CD、EMD evaluation indicators
View tableView in ArticleTable 1. CD、EMD evaluation indicators
Item CD EMD PSGN 3D-LMNet Pix2point Ours PSGN 3D-LMNet Pix2point Ours Airplane 3.74 3.34 3.29 2.35 6.38 4.77 3.82 2.65 Bench 4.63 4.55 4.59 3.34 5.88 4.99 4.31 3.45 Cabinet 6.98 6.09 6.07 4.45 6.04 6.35 4.94 4.28 Car 5.20 4.55 4.39 3.32 4.87 4.10 3.61 2.91 Chair 6.39 6.41 6.48 3.41 9.63 8.01 6.45 3.53 Lamp 6.33 7.10 6.58 4.85 16.17 15.80 8.45 6.13 Monitor 6.15 6.40 6.39 4.49 7.59 7.13 5.94 4.15 Phone 4.56 4.63 4.27 3.46 5.07 5.43 3.77 3.43 Rifle 2.91 2.75 2.89 2.33 8.48 6.08 4.25 3.72 Sofa 6.98 5.85 5.85 4.13 7.42 5.65 5.03 3.81 Speaker 8.75 8.10 8.39 5.61 8.70 9.15 7.37 5.05 Table 6.00 6.05 6.26 4.03 8.40 7.82 6.05 4.29 Vessel 4.38 4.37 4.55 3.52 6.18 5.68 4.89 3.86 Mean 5.62 5.40 5.38 3.79 7.75 7.00 5.30 3.94
Table 2. IoU evaluation indicators
View tableView in ArticleTable 2. IoU evaluation indicators
Item PSGN 3D-R2N2 Ours 1view 3views 5views Airplane 0.601 0.513 0.549 0.561 0.71 Bench 0.550 0.421 0.502 0.527 0.69 Cabinet 0.771 0.716 0.763 0.772 0.74 Car 0.831 0.798 0.829 0.836 0.66 Chair 0.544 0.466 0.533 0.550 0.67 Lamp 0.462 0.381 0.415 0.421 0.61 Monitor 0.552 0.468 0.545 0.565 0.68 Phone 0.749 0.661 0.732 0.754 0.72 Rifle 0.604 0.544 0.593 0.600 0.71 Sofa 0.708 0.628 0.690 0.706 0.68 Speaker 0.737 0.662 0.708 0.717 0.73 Table 0.606 0.513 0.564 0.580 0.65 Vessel 0.611 0.513 0.596 0.610 0.63 Mean 0.640 0.560 0.617 0.631 0.68
Table 3. Evaluation indicators of GCN ablation experiments
View tableView in ArticleTable 3. Evaluation indicators of GCN ablation experiments
Item CD Deviation EMD Deviation FC GCN FC GCN Airplane 2.54 2.35 +0.19 2.68 2.65 +0.03 Bench 3.39 3.34 +0.05 3.51 3.45 +0.06 Cabinet 4.48 4.45 +0.03 4.13 4.28 -0.15 Car 3.43 3.32 +0.11 3.03 2.91 +0.12 Chair 3.73 3.41 +0.32 3.67 3.53 +0.14 Lamp 4.76 4.85 -0.09 5.76 6.13 -0.37 Monitor 4.56 4.49 +0.07 4.55 4.15 +0.40 Phone 3.74 3.46 +0.28 3.51 3.43 +0.08 Rifle 2.45 2.33 +0.12 3.93 3.72 +0.21 Sofa 4.17 4.13 +0.04 4.05 3.81 +0.24 Speaker 5.82 5.61 +0.21 4.78 5.05 -0.27 Table 3.97 4.03 -0.06 4.43 4.29 +0.14 Vessel 3.67 3.52 +0.15 3.95 3.86 +0.09 Mean 3.90 3.79 +0.11 4.00 3.94 +0.06
Table 4. Evaluation indicators of NFFD ablation experiments
View tableView in ArticleTable 4. Evaluation indicators of NFFD ablation experiments
Item CD Deviation EMD Deviation NFFD GCN NFFD GCN Bench 3.72 3.34 +0.38 4.08 3.45 +0.63 Monitor 4.68 4.49 +0.19 4.87 4.15 +0.72 Phone 3.94 3.46 +0.48 3.66 3.43 +0.23 Mean 4.11 3.76 +0.35 4.20 3.68 +0.52
Table 5. CD comparison of loss function ablation experiments
View tableView in ArticleTable 5. CD comparison of loss function ablation experiments
Item Remove isometric prior loss Remove symmetry loss All loss Bench 3.362 5 3.356 8 3.344 6 Rifle 2.349 1 2.345 9 2.332 7 Vessel 3.574 9 3.544 2 3.526 1 Mean 3.095 5 3.082 3 3.067 8
Tools
Get Citation
Copy Citation Text
Yuanfeng LIAN, Shoushuang PEI, Wei HU. Single-view 3D object reconstruction based on NFFD and graph convolution[J]. Optics and Precision Engineering, 2022, 30(10): 1189
Paper Information
Category: Information Sciences
Received: Nov. 10, 2021
Accepted: --
Published Online: Jun. 1, 2022
The Author Email: LIAN Yuanfeng (lianyuanfeng@cup.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20