Research map / 3D vision / Gaussian splatting
Matryoshka Gaussian Splatting
Zhilin Guo, Boqiao Zhang, Hakan Aktas, Kyle Fogarty, Jeffrey Hu, Nursena Aslan, Wenzhao Li, Canberk Baykal, Albert Miao, Josef Bengtson, Chenliang Zhou, Weihao Xia and 2 more
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 |
|---|---|---|---|
| order Gaussians by opacity descending instead of ascending (ablation) | MGS · MipNeRF 360 bicycle (ablation) | AUC_fps: 30.53 -> 68.62 | Improved |
| order by opacity descending instead of SH energy descending (ablation) | MGS · MipNeRF 360 bicycle (ablation) | PSNR at 10% budget: 17.6 -> 22.2 | Improved |
| order by opacity descending instead of color variance (ablation) | MGS · MipNeRF 360 bicycle (ablation) | AUC_fps: 51.94 -> 68.62 | Improved |
| order by opacity descending instead of fixed random order (ablation) | MGS · MipNeRF 360 bicycle (ablation) | PSNR at 10% budget: 21.5 -> 22.2 | Improved |
| add full-set loss term to prefix loss (ablation) | MGS · MipNeRF 360 bicycle (ablation) | PSNR (full budget): 24.97 -> 25.43 | Improved |
| add full-set loss term to prefix loss (ablation) | MGS · MipNeRF 360 bicycle (ablation) | LPIPS (full budget): 0.253 -> 0.175 | Improved |
| use stochastic prefix sampling instead of MRL deterministic nesting (ablation) | MGS · MipNeRF 360 bicycle (ablation) | AUC_splats: 60.35 -> 60.96 | Improved |
| shift loss weight from prefix-heavy 9:1 to full-heavy 1:9 (ablation) | MGS · MipNeRF 360 bicycle (ablation) | PSNR at 10% prefix: 22.2 -> 21.1 | Hurt |
| shift loss weight from 9:1 to 1:9 (ablation) | MGS · MipNeRF 360 bicycle (ablation) | PSNR at full budget: +0.2 dB (25.5) | Improved |
| train for all budgets (MGS) versus full-set-only training | 3DGS-MCMC · MipNeRF 360 | PSNR: 28.40 -> 28.20 | Hurt |
| train for all budgets (MGS) versus full-set-only training | 3DGS-MCMC · Deep Blending | PSNR: 27.63 -> 28.41 | Improved |
| MGS instead of Octree-GS | continuous LoD · MipNeRF 360 | PSNR: 27.62 -> 28.20 | Improved |
Closest papers by meaning
- LODGE: Level-of-Detail Large-Scale Gaussian Splatting with Efficient Rendering (NeurIPS 2025 · similarity 0.93)
- CLoD-GS: Continuous Level-of-Detail via 3D Gaussian Splatting (ICLR 2026 · similarity 0.93)
- Octree-GS: Towards Consistent Real-time Rendering with LOD-Structured 3D Gaussians (TPAMI 2025 · similarity 0.93)
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 2025 · similarity 0.93)
- LOD-GS: Achieving Levels of Detail using Scalable Gaussian Soup (CVPR 2025 · similarity 0.93)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 2025 · similarity 0.93)
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.
