Laser & Optoelectronics Progress, Volume. 61, Issue 16, 1611010(2024)
Advances in Differentiable Rendering Based on Three-Dimensional Gaussian Splatting (Invited)
Figures & Tables(14)
![3D Gaussian splatting technology utilizes adaptive density control for densification[8]](/Images/icon/loading.gif){kind=icon}
Fig. 4. 3D Gaussian splatting technology utilizes adaptive density control for densification[8]
Fig. 7. Comparison of 2DGS, 3DGS and SuGaR on the surface mesh extraction task[45]
Fig. 9. Physical-based realistic point cloud rendering pipeline[49]. (a) Point cloud; (b) rendered normal map; (c) ambient occlusion map;(d) realistic relighting
Table 1. Quantitative comparison results of the improved schemes of 3DGS on the Mip-NeRF 360 dataset
View tableTable 1. Quantitative comparison results of the improved schemes of 3DGS on the Mip-NeRF 360 dataset
Improved scheme PSNR↑ SSIM↑ LPIPS↓ FPS↑ Memory↓ 3DGS 27.21 0.815 0.214 134 734 O-GS 27.17 0.836 0.210 ‒ ‒ GaussianPro 27.92 0.825 0.208 108 ‒ Pixel-GS 27.88 0.834 0.176 89 ‒ Mini-Splatting 27.54 0.832 0.175 83 < 50FreGS 27.85 0.826 0.209 ‒ ‒ Scaffold-GS 28.84 0.848 0.220 102 156 GES 26.91 0.794 0.250 186 377 Spec-Gaussian 28.01 0.812 0.222 70 245
Table 2. Quantitative comparison results of dynamic reconstruction methods on the D-NeRF synthetic dataset
View tableTable 2. Quantitative comparison results of dynamic reconstruction methods on the D-NeRF synthetic dataset
Method PSNR↑ SSIM↑ LPIPS↓ FPS↑ Size 4DGS 34.05 0.980 0.020 82 800 D3DGS 39.51 0.990 0.012 30 800 CoGS 37.90 0.983 0.017 ‒ 800 E3DGS 32.07 0.960 ‒ 150 800 4DG 34.09 0.980 0.020 114 800 NPG 33.86 0.974 0.042 20 800 SCGS 43.31 0.998 0.008 ‒ 400 DynMF 36.80 0.983 0.020 >300 400 GauFRe 34.80 0.982 0.020 50 400
Table 3. Quantitative comparison of large scene reconstruction results
View tableTable 3. Quantitative comparison of large scene reconstruction results
Method Residence Rubble PSNR↑ SSIM↑ LPIPS↓ PSNR↑ SSIM↑ LPIPS↓ MegaNeRF 22.08 0.628 0.489 24.06 0.553 0.516 SwitchNeRF 22.57 0.654 0.457 24.31 0.562 0.496 GridNeRF 23.77 0.802 0.137 24.13 0.767 0.207 Fed3DGS 20.00 0.665 0.344 20.62 0.588 0.437 CityGaussian 22.00 0.813 0.211 25.77 0.813 0.228 VastGaussian 24.25 0.852 0.124 26.92 0.823 0.132 Method Building Campus PSNR↑ SSIM↑ LPIPS↓ PSNR↑ SSIM↑ LPIPS↓ MegaNeRF 20.93 0.547 0.504 23.42 0.537 0.618 SwitchNeRF 21.54 0.579 0.474 23.62 0.541 0.609 GridNeRF ‒ ‒ ‒ 24.90 0.757 0.162 Fed3DGS 18.66 0.602 0.362 21.64 0.635 0.436 CityGaussian 21.55 0.778 0.246 ‒ ‒ ‒ VastGaussian 23.50 0.804 0.130 26.00 0.816 0.151
Table 4. Quantitative comparison of sparse view reconstruction results on the LLFF dataset
View tableTable 4. Quantitative comparison of sparse view reconstruction results on the LLFF dataset
Method Three views Six views Nine views PSNR↑ SSIM↑ LPIPS↓ PSNR↑ SSIM↑ LPIPS↓ PSNR↑ SSIM↑ LPIPS↓ DNGaussian 19.12 0.591 0.294 22.18 0.755 0.198 23.17 0.788 0.180 CoherentGS 20.33 0.725 0.180 ‒ ‒ ‒ ‒ ‒ ‒ FSGS 20.43 0.682 0.248 24.09 0.823 0.145 25.31 0.860 0.122 CoR-GS 20.45 0.712 0.196 24.49 0.837 0.115 26.06 0.874 0.089
Tools
Get Citation
Copy Citation Text
Jian Gao, Linzhuo Chen, Qiu Shen, Xun Cao, Yao Yao. Advances in Differentiable Rendering Based on Three-Dimensional Gaussian Splatting (Invited)[J]. Laser & Optoelectronics Progress, 2024, 61(16): 1611010
Paper Information
Category: Imaging Systems
Received: May. 27, 2024
Accepted: Jun. 27, 2024
Published Online: Aug. 12, 2024
The Author Email: Yao Yao (yaoyao@nju.edu.cn)
CSTR:32186.14.LOP241369
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20