Research map / 3D vision / Gaussian splatting
Beyond Heuristics: Learnable Density Control for 3D Gaussian Splatting
Zhenhua Ning, Xin Li, Jun Yu, Guangming Lu, Yaowei Wang, Wenjie Pei
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 learnable densification (LD) policy to FastGS* (ablation) | FastGS* · Mip-NeRF 360 | PSNR: 27.73 -> 28.14 | Improved |
| add learnable pruning (LP) on top of learnable densification (ablation) | FastGS* + LD · Mip-NeRF 360 | PSNR / #Gaussians: 28.14 -> 28.30 PSNR; 2.48 -> 2.17 M Gaussians | Improved |
| add sensitivity-based reward instead of rendering-weight reward (ablation) | LeGS · Mip-NeRF 360 | PSNR: 28.01 -> 28.30 | Improved |
| use maintain-action value instead of PPO critic network (ablation) | LeGS · Mip-NeRF 360 | number of Gaussians (M): 2.63 -> 2.17 | Improved |
| use maintain-action value instead of PPO critic network (ablation) | LeGS · Mip-NeRF 360 | PSNR: 28.17 -> 28.30 | No clear effect |
| closed-form sensitivity score instead of naive re-rendering (ablation) | LeGS reward computation · Mip-NeRF 360 (Bicycle) | extra computing time (ms): 30.5 -> 0.9 | Improved |
| learned RL policy instead of FastGS heuristic density control | FastGS backend · Mip-NeRF 360 | PSNR: 27.87 -> 28.30 | Improved |
| learned RL policy instead of FastGS heuristic density control | FastGS backend · Deep Blending | PSNR: 30.17 -> 30.35 | Improved |
| increase FastGS Gaussian count via lower gradient threshold (FastGS*) | FastGS · Mip-NeRF 360 | PSNR: 27.87 -> 27.73 | Hurt |
| learned policy vs heuristic: proportion of locally optimal actions | LeGS vs FastGS · Mip-NeRF 360, iteration 15000 | locally optimal action rate (%): 35.60% -> 64.85% | Improved |
| add RL policy training and sensitivity reward to FastGS* | FastGS* · Mip-NeRF 360 | total training time (s): 599 -> 745 | Hurt |
| LeGS vs FastGS* peak GPU memory | LeGS · Mip-NeRF 360 | peak GPU memory (GB): 8.99 -> 12.57 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Perceptual-GS — Perceptual-GS: Scene-adaptive Perceptual Densification for Gaussian Splatting (ICML 2025) · 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
- Gradient-Direction-Aware Density Control for 3D Gaussian Splatting (ICLR 2026 · similarity 0.92)
- Frequency-Aware Density Control via Reparameterization for High-Quality Rendering of 3D Gaussian Splatting (AAAI 2025 · similarity 0.91)
- AdpSplit: Error-Driven Adaptive Splitting for Faster Geometry Discovery in 3D Gaussian Splatting (NeurIPS 2026 · similarity 0.91)
- Steepest Descent Density Control for Compact 3D Gaussian Splatting (CVPR 2025 · similarity 0.91)
- DC4GS: Directional Consistency-Driven Adaptive Density Control for 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.90)
- H3D-DGS: Exploring Heterogeneous 3D Motion Representation for Deformable 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.89)
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.
