Gradient-Driven Natural Selection for Compact 3D Gaussian Splatting

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Deng, Xiaobin, Yu, Qiuli, Diao, Changyu, Li, Min, Xu, Duanqing
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917096598798336
author Deng, Xiaobin
Yu, Qiuli
Diao, Changyu
Li, Min
Xu, Duanqing
author_facet Deng, Xiaobin
Yu, Qiuli
Diao, Changyu
Li, Min
Xu, Duanqing
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2511_16980
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Gradient-Driven Natural Selection for Compact 3D Gaussian Splatting
Deng, Xiaobin
Yu, Qiuli
Diao, Changyu
Li, Min
Xu, Duanqing
Computer Vision and Pattern Recognition
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.
title Gradient-Driven Natural Selection for Compact 3D Gaussian Splatting
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2511.16980