Research map / 3D vision / Gaussian splatting
Complexity Aware Continuous Level of Details for Gaussian Splatting
Radu Beche, Raoul de Charette
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Studies how to allocate capacity in a single 3D Gaussian Splatting set so that prefix truncations under varying compute budgets remain effective, motivated by image complexity correlating with reconstruction error.
Problem
A single Gaussian set must remain effective when truncated to a prefix of primitives under varying compute budgets, raising how capacity should be allocated as the budget decreases.
Approach
Observes that image complexity correlates with reconstruction error in both image and 3D Gaussian space, and that this structure degrades as budget decreases; the abstract is truncated so the specific method is not stated.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add complexity-aware densification, importance score and variance loss to 3DGS-MCMC (ablation) | 3DGS-MCMC · MipNerf360 (7 scenes) | AUC PSNR: 28.74 -> 29.09 | Improved |
| add complexity-aware importance score only (ablation) | 3DGS-MCMC · MipNerf360 (7 scenes) | AUC PSNR: 28.74 -> 28.94 | Improved |
| add complexity-aware densification only (ablation) | 3DGS-MCMC · MipNerf360 (7 scenes) | PSNR at 100% capacity: 29.55 -> 29.79 | Improved |
| add variance loss only (ablation) | 3DGS-MCMC · MipNerf360 (7 scenes) | AUC SSIM: 0.855 -> 0.855 | No clear effect |
| add variance loss to densification alone (ablation) | complexity densification · MipNerf360 (7 scenes) | AUC PSNR: 28.81 -> 28.75 | Mixed |
| add variance loss on top of densification + score (ablation) | KoalaGS · MipNerf360 (7 scenes) | AUC PSNR: 28.99 -> 29.09 | Improved |
| use complexity dropout 0.2 instead of none (0) (ablation) | complexity-biased relocation · Deep Blending (2 scenes) | AUC PSNR: 29.00 -> 29.51 | Improved |
| use complexity dropout 0.2 instead of pure opacity sampling (1) (ablation) | complexity-biased relocation · Tanks and Temples (2 scenes) | AUC PSNR: 23.88 -> 24.02 | Improved |
| use high dropout rates 0.6 or 0.8 (ablation) | complexity-biased relocation · Deep Blending / Tanks and Temples | AUC PSNR: 29.51 -> 28.94 (delta 0.6, DB) | Hurt |
| rank primitives by opacity plus scale over complexity instead of opacity alone, with complexity-aware training | 3DGS-MCMC · MipNerf360 (7 scenes), 10% capacity | PSNR: 20.58 -> 27.82 | Improved |
| use KoalaGS instead of MGS at 0.1% capacity | prefix LoD · Tanks and Temples (2 scenes) | PSNR: 12.10 -> 11.56 | Mixed |
| use KoalaGS ranking for progressive streaming vs MGS | streaming · Bonsai, 50 Mbps | time to 25 dB: 1.96 s -> 1.47 s | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- FlexGS — FlexGS: Train Once, Deploy Everywhere with Many-in-One Flexible 3D Gaussian Splatting (CVPR 2025) · better on 1 of 1 shared result
- LightGaussian — LightGaussian: Unbounded 3D Gaussian Compression with 15x Reduction and 200+ FPS (NeurIPS 2024) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Reframing Gaussian Splatting Densification with Complexity-Density Consistency of Primitives (NeurIPS 2025 · similarity 0.93)
- Matryoshka Gaussian Splatting (ECCV 2026 · similarity 0.91)
- Gaussian Herding across Pens: An Optimal Transport Perspective on Global Gaussian Reduction for 3DGS (NeurIPS 2025 · similarity 0.91)
- Optimized Minimal 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.90)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- Locality-aware Gaussian Compression for Fast and High-quality Rendering (ICLR 2025 · 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.
