Research map / 3D vision / Gaussian splatting
SSD-GS: Scattering and Shadow Decomposition for Relightable 3D Gaussian Splatting
Iris Zheng, Guojun Tang, Alexander Doronin, Paul D. Teal, Fang-Lue Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add subsurface scattering term to diffuse + specular (ablation) | SSD-GS · NRHints Pixiu (test) | PSNR: 20.5692 -> 24.8100 | Improved |
| add shadow term to the full model without it (ablation) | SSD-GS · NRHints Pixiu (test) | PSNR: 30.5952 -> 31.1213 | Improved |
| add subsurface scattering term to the full model without it (ablation) | SSD-GS · NRHints Pixiu (test) | SSIM: 0.7204 -> 0.9452 | Improved |
| add specular term to diffuse only (ablation) | SSD-GS · NRHints Pixiu (test) | PSNR: 20.1878 -> 20.5692 | Improved |
| use progressive schedule instead of joint training from the start (ablation) | SSD-GS · NRHints Pixiu (test) | PSNR: 31.0880 -> 31.1213 | Improved |
| order phases Diffuse, Shadow, Scatter, Specular instead of swapping last two (ablation) | SSD-GS · Average over 8 scenes (test) | PSNR: 32.0824 -> 33.2876 | Improved |
| introduce terms sequentially instead of merged after diffuse warm-up (ablation) | SSD-GS · Average over 8 scenes (test) | PSNR: 32.7400 -> 33.2876 | Improved |
| apply shadow term to SSS component in addition to diffuse and specular (ablation) | SSD-GS · Average over 9 scenes (test) | PSNR: 33.0039 -> 32.1271 | Hurt |
| add camera pose and light direction optimization (ablation) | SSD-GS · SSS-GS synthetic (test average) | PSNR: 37.4409 -> 38.3542 | Improved |
| replace GS3 with SSD-GS on Lego and Hotdog test scenes | SSD-GS · GS3 synthetic Lego and Hotdog (test) | PSNR: Lego 30.5617 vs 30.4664; Hotdog 32.7244 vs 32.1330 | No clear effect |
| use SSD-GS instead of RNG and GS3 on NRHints real scenes | SSD-GS · NRHints FurScene (test) | PSNR: 28.2228 -> 30.7349 | Improved |
Closest papers by meaning
- Subsurface Scattering for Gaussian Splatting (NeurIPS 2024 · similarity 0.92)
- GI-GS: Global Illumination Decomposition on Gaussian Splatting for Inverse Rendering (ICLR 2025 · similarity 0.92)
- MetaGS: A Meta-Learned Gaussian-Phong Model for Out-of-Distribution 3D Scene Relighting (NeurIPS 2025 · similarity 0.90)
- MaterialRefGS: Reflective Gaussian Splatting with Multi-view Consistent Material Inference (NeurIPS 2025 · similarity 0.90)
- RNG: Relightable Neural Gaussians (CVPR 2025 · similarity 0.90)
- Eclipse: Disambiguating Illumination and Materials using Unintended Shadows (CVPR 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
