Gradient-Driven Natural Selection for Compact 3D Gaussian Splatting

2025-11-21

AI Summary

To address the high storage and computational overhead of 3D Gaussian Splatting (3DGS) caused by its large number of Gaussian primitives, this paper proposes a fully automatic, nature-inspired pruning framework. Departing from manual thresholds or auxiliary learnable parameters, our method is the first to introduce a survival-competition mechanism into 3DGS optimization: a rendering-quality-driven gradient constructs a regularization gradient field that models “survival pressure,” enabling autonomous primitive retention decisions; opacity decay coupled with a bounded opacity prior further accelerates convergence. Under a stringent 15% parameter budget, our approach achieves a PSNR gain exceeding 0.6 dB over baseline methods, establishing new state-of-the-art performance for compact 3DGS reconstruction.

Technology Category

• Computer Vision: 3D Computer Vision
• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

3DGS employs a large number of Gaussian primitives to fit scenes, resulting in substantial storage and computational overhead. Existing pruning methods rely on manually designed criteria or introduce additional learnable parameters, yielding suboptimal results. To address this, we propose an natural selection inspired pruning framework that models survival pressure as a regularization gradient field applied to opacity, allowing the optimization gradients--driven by the goal of maximizing rendering quality--to autonomously determine which Gaussians to retain or prune. This process is fully learnable and requires no human intervention. We further introduce an opacity decay technique with a finite opacity prior, which accelerates the selection process without compromising pruning effectiveness. Compared to 3DGS, our method achieves over 0.6 dB PSNR gain under 15% budgets, establishing state-of-the-art performance for compact 3DGS. Project page https://xiaobin2001.github.io/GNS-web.

Problem

• Reduces storage and computational costs of 3D Gaussian Splatting
• Eliminates manual criteria or extra parameters for pruning Gaussians
• Automates selection of Gaussians to retain using gradient-driven optimization

Innovation

• Natural selection pruning framework driven by optimization gradients
• Opacity decay technique with finite opacity prior
• Fully learnable process requiring no human intervention