Gradient-Driven Natural Selection for Compact 3D Gaussian Splatting

TL;DR

This paper introduces a learnable, gradient-driven pruning method for 3D Gaussian splatting that autonomously selects primitives to optimize rendering quality while reducing storage and computation.

Contribution

We propose a natural selection inspired pruning framework that uses a regularization gradient field to automatically determine which Gaussians to keep or prune, eliminating manual criteria.

Findings

Achieves over 0.6 dB PSNR gain at 15% budget compared to 3DGS

Fully learnable pruning process without human intervention

State-of-the-art performance for compact 3D Gaussian splatting

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.