Provable Pruning for Efficient 3D Gaussian Splatting via Coresets

2026-07-02

AI Summary

This work addresses the challenge of deploying 3D Gaussian splatting models on resource-constrained devices due to their large parameter count, compounded by the lack of provable guarantees on rendering quality in existing compression methods. We present the first theoretical framework for constructing 3D Gaussian coresets with multiplicative approximation guarantees. Leveraging sensitivity analysis for importance sampling, our method builds a weighted subset of Gaussians that provably approximates the original scene’s objective function under a given rendering resolution, while explicitly revealing the dependency between approximation feasibility and resolution. Without fine-tuning or with only minimal fine-tuning, our approach significantly outperforms current heuristic pruning strategies, achieving state-of-the-art performance—particularly under aggressive compression regimes.

Abstract

3D Gaussian Splatting (3DGS) enables high-quality real-time novel-view synthesis, but practical scenes often contain millions of Gaussians, making compression essential for deployment on limited hardware. Existing reduction methods are effective but mostly heuristic: they provide no multiplicative approximation guarantee for the rendered objective, and thus rely heavily on costly post-pruning finetuning to recover quality. We ask a basic question: can a 3DGS scene be provably replaced by a much smaller weighted subset (coreset) while preserving the objective of interest? We first show that, in the unrestricted setting, no non-trivial multiplicative 3DGS coreset exists. We then show that multiplicative guarantees are not impossible, but resolution-dependent. For a prescribed rendering resolution, such as representative views or grids of views/rays, we provide the first weighted coreset construction theorem for 3DGS. The construction samples Gaussians by sensitivity: provable importance scores measuring each Gaussian's role in the full-scene objective. Finally, under explicit validity and log-transmittance stability assumptions, we turn this objective guarantee into a rendering guarantee. Empirically, our method is strongest where deployment needs it most: aggressive compression with no or minimal recovery compute. In prune-only and very short finetuning regimes, it achieves state-of-the-art performance, showing that principled importance estimation can be both theoretically meaningful and practically useful. Open-source code is available at https://github.com/waseem-m/3dgs_provable_coresets.

Problem

• 3D Gaussian Splatting
• coreset
• pruning
• approximation guarantee
• rendering compression

Innovation

• provable coreset
• 3D Gaussian Splatting
• sensitivity sampling
• rendering approximation guarantee
• compression without finetuning