Research map / 3D vision / Gaussian splatting
Revising Densification in Gaussian Splatting
Samuel Rota Bulò, Lorenzo Porzi, Peter Kontschieder
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 error-based densification, opacity correction, growth control and opacity regularization to 3DGS | 3DGS · Mip-NeRF 360 | LPIPS: 0.250 -> 0.223 | Improved |
| apply the same method to Mip-Splatting | Mip-Splatting · Tanks and Temples | PSNR: 23.65 -> 24.10 | Improved |
| apply the method on Deep Blending | 3DGS · Deep Blending | PSNR: 29.54 -> 29.50 | No clear effect |
| add opacity correction after cloning (ablation) | 3DGS · Mip-NeRF 360 | PSNR: 27.45 -> 27.65 | Improved |
| add opacity correction after cloning (ablation) | Ours (full) · Mip-NeRF 360 | PSNR: 27.04 -> 27.61 | Improved |
| add growth control (primitive budget) to gradient-based 3DGS (ablation) | 3DGS · Mip-NeRF 360 | PSNR: 27.45 -> 27.35 | Hurt |
| add growth control to the full method (ablation) | Ours (full) · Mip-NeRF 360 | PSNR: 27.54 -> 27.61 | Improved |
| add opacity regularization to 3DGS (ablation) | 3DGS · Mip-NeRF 360 | LPIPS: 0.250 -> 0.243 | Improved |
| add opacity regularization to the full method (ablation) | Ours (full) · Mip-NeRF 360 | PSNR: 27.28 -> 27.61 | Improved |
| use l1 instead of SSIM as the densification guiding error (ablation) | Ours · Mip-NeRF 360 | LPIPS: 3DGS 0.250 -> Ours-l1 0.232 | Improved |
| give standard gradient-based densification 10M primitives (ablation) | 3DGS · Mip-NeRF 360 flowers | qualitative blur: 4.2M -> 10M primitives | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Mip-Splatting — Mip-Splatting: Alias-free 3D Gaussian Splatting (CVPR 2024) · better on 3 of 4 shared results
- Mip-NeRF 360 — Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance Fields (CVPR 2022) · better on 1 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
- Improving Gaussian Splatting with Localized Points Management (CVPR 2025 · similarity 0.93)
- Steepest Descent Density Control for Compact 3D Gaussian Splatting (CVPR 2025 · similarity 0.92)
- Frequency-Aware Density Control via Reparameterization for High-Quality Rendering of 3D Gaussian Splatting (AAAI 2025 · similarity 0.92)
- AdpSplit: Error-Driven Adaptive Splitting for Faster Geometry Discovery in 3D Gaussian Splatting (NeurIPS 2026 · similarity 0.91)
- Metropolis-Hastings Sampling for 3D Gaussian Reconstruction (NeurIPS 2025 · similarity 0.91)
- ReorgGS: Equivalent Distribution Reorganization for 3D Gaussian Splatting (NeurIPS 2026 · similarity 0.91)
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.
