Research map / 3D vision / Gaussian splatting
REFINE: Super-efficient Pruning for 3D Gaussian Splatting via Rendering-Free Primitive Importance
Zhang Chen, Shuai Wan, Mengting Yu, Fuzheng Yang, Junhui Hou
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 rendering-free Hessian importance (REFINE) instead of parameter heuristics (LightGaussian) | 3DGS pruning · MipNeRF360 (ratio 0.5) | PSNR: 22.21 -> 26.61 | Improved |
| use REFINE instead of render-aware PUP at ratio 0.5 | 3DGS pruning · MipNeRF360 (ratio 0.5) | PSNR: 26.07 -> 26.61 | Improved |
| use REFINE instead of PUP at ratio 0.5 | 3DGS pruning · Tanks&Temples (ratio 0.5) | PSNR: 23.18 -> 22.97 | Hurt |
| use REFINE instead of PUP at extreme ratio 0.7 | 3DGS pruning · DeepBlending (ratio 0.7) | PSNR: 29.14 -> 27.93 | Hurt |
| use REFINE instead of PUP at ratio 0.7 | 3DGS pruning · MipNeRF360 (ratio 0.7) | PSNR: 25.12 -> 24.43 | Hurt |
| use REFINE instead of PUP for pruning cost | 3DGS pruning · MipNeRF360 (ratio 0.7) | GFLOPs: 9582.81 -> 3.14 | Improved |
| use REFINE instead of PUP for latency | 3DGS pruning · MipNeRF360 (ratio 0.7) | Time (s): 38.15 -> 2.55 | Improved |
| add view-dependent visibility term (ablation) | REFINE · Mip360 subset (ratio 0.5) | PSNR: 27.74 -> 28.35 | Improved |
| add geometric projection term (ablation) | REFINE · Mip360 subset (ratio 0.5) | PSNR: 27.24 -> 28.35 | Improved |
| use content-adaptive lambda instead of equal lambda (ablation) | REFINE · Mip360 subset (ratio 0.5) | PSNR: 27.82 -> 28.35 | Improved |
| fine-tune after pruning, REFINE vs PUP at ratio 0.7 | 3DGS pruning · Mip360 kitchen (fine-tuned, ratio 0.7) | PSNR: 29.11 -> 30.55 PSNR; LPIPS 0.137 -> 0.152 | Mixed |
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.
- MesonGS — MesonGS: Post-training Compression of 3D Gaussians via Efficient Attribute Transformation (ECCV 2024) · better on 4 of 5 shared results
- LightGaussian — LightGaussian: Unbounded 3D Gaussian Compression with 15x Reduction and 200+ FPS (NeurIPS 2024) · better on 4 of 4 shared results
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
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 2025 · similarity 0.93)
- 3D Gaussian Rendering Can Be Sparser: Efficient Rendering via Learned Fragment Pruning (NeurIPS 2024 · similarity 0.93)
- LP-3DGS: Learning to Prune 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.92)
- PUP 3D-GS: Principled Uncertainty Pruning for 3D Gaussian Splatting (CVPR 2025 · similarity 0.92)
- ReorgGS: Equivalent Distribution Reorganization for 3D Gaussian Splatting (NeurIPS 2026 · similarity 0.92)
- Gaussian Herding across Pens: An Optimal Transport Perspective on Global Gaussian Reduction for 3DGS (NeurIPS 2025 · similarity 0.92)
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.
