Beyond Heuristics: Learnable Density Control for 3D Gaussian Splatting

2026-05-01

AI Summary

This work addresses the limitations of existing 3D Gaussian Splatting (3DGS) methods, which rely on handcrafted heuristics for density control and struggle to adapt to complex geometric scenes. To overcome this, the paper introduces reinforcement learning into 3DGS density optimization for the first time, proposing a learnable policy network and designing a precise reward function grounded in sensitivity analysis. By deriving a closed-form solution, the authors reduce the computational complexity of reward evaluation from O(N²) to O(N), substantially improving optimization efficiency. Experimental results demonstrate that the proposed method consistently outperforms state-of-the-art approaches across the Mip-NeRF 360, Tanks & Temples, and Deep Blending benchmarks, achieving a superior trade-off between reconstruction quality and computational efficiency.

Abstract

While 3D Gaussian Splatting (3DGS) has demonstrated impressive real-time rendering performance, its efficacy remains constrained by a reliance on heuristic density control. Despite numerous refinements to these handcrafted rules, such methods inherently lack the flexibility to adapt to diverse scenes with complex geometries. In this paper, we propose a paradigm shift for density control from rigid heuristics to fully learnable policies. Specifically, we introduce \textbf{LeGS}, a framework that reformulates density control as a parameterized policy network optimized via Reinforcement Learning (RL). Central to our approach is the tailored effective reward function grounded in sensitivity analysis, which precisely quantifies the marginal contribution of individual Gaussians to reconstruction quality. To maintain computational tractability, we derive a closed-form solution that reduces the complexity of reward calculation from $O(N^2)$ to $O(N)$. Extensive experiments on the Mip-NeRF 360, Tanks \& Temples, and Deep Blending datasets demonstrate that \textbf{LeGS} significantly outperforms state-of-the-art methods, striking a superior balance between reconstruction quality and efficiency. The code will be released at https://github.com/AaronNZH/LeGS

Problem

• 3D Gaussian Splatting
• density control
• heuristics
• scene reconstruction
• geometric complexity

Innovation

• Learnable Density Control
• 3D Gaussian Splatting
• Reinforcement Learning
• Sensitivity Analysis
• Closed-form Reward