Camera-Agnostic Pruning of 3D Gaussian Splats via Descriptor-Based Beta Evidence

Peter Fasogbon; Ugurcan Budak; Patrice Rondao Alface; Hamed Rezazadegan Tavakoli

cs.CVMar 23, 2026

Camera-Agnostic Pruning of 3D Gaussian Splats via Descriptor-Based Beta Evidence

Authors: Peter Fasogbon, Ugurcan Budak, Patrice Rondao Alface, Hamed Rezazadegan Tavakoli

Organizations: Nokia Technologies, Espoo, Finland · Nokia Technologies, Antwerp, Belgium

Abstract

The pruning of 3D Gaussian splats is essential for reducing their complexity to enable efficient storage, transmission, and downstream processing. However, most of the existing pruning strategies depend on camera parameters, rendered images, or view-dependent measures. This dependency becomes a hindrance in emerging camera-agnostic exchange settings, where splats are shared directly as point-based representations (e.g., .ply). In this paper, we propose a camera-agnostic, one-shot, post-training pruning method for 3D Gaussian splats that relies solely on attribute-derived neighbourhood descriptors. As our primary contribution, we introduce a hybrid descriptor framework that captures structural and appearance consistency directly from the splat representation. Building on these descriptors, we formulate pruning as a statistical evidence estimation problem and introduce a Beta evidence model that quantifies per-splat reliability through a probabilistic confidence score. Experiments conducted on standardized test sequences defined by the ISO/IEC MPEG Common Test Conditions (CTC) demonstrate that our approach achieves substantial pruning while preserving reconstruction quality, establishing a practical and generalizable alternative to existing camera-dependent pruning strategies.

Figures & tables

Figure 1: Overview of the proposed Gaussian splat pruning method on an example scene from the ISO/IEC MPEG CTC dataset [ 8 ] , rendered from a fixed camera viewpoint. The image columns show the non-pruned baseline and increasing pruning thresholds $\tau=0.1,0.3,0.50,$ and $0.7$ , which remove about $10\%$ , $30\%$ , $50\%$ , and $70\%$ of splats, respectively.

Figure 2: Method overview of the proposed camera-agnostic Gaussian splat pruning framework. Given trained 3D Gaussian splats, local descriptor statistics are computed and transformed into Beta-distributed evidence to model confidence and uncertainty. A one-shot confidence-aware ranking then removes redundant splats, balancing expected pruning likelihood and uncertainty to produce the final pruned representation.

Dataset	Metric	Camera-Aware Methods						Our Method
		ConfSplatPrune			LightGSPrune			BetaDescPrune
		Low	Mid	High	Low	Mid	High	Low	Mid	High
bartender (tracked)	PSNR (↑)	34.84	31.03	30.10	34.80	31.15	30.05	33.50	31.03	30.00
	SSIM (↑)	0.991	0.990	0.985	0.992	0.990	0.983	0.990	0.982	0.973
	LPIPS (↓)	0.015	0.028	0.031	0.015	0.029	0.030	0.019	0.034	0.049
breakfast (tracked)	PSNR (↑)	34.01	29.13	25.07	33.03	29.12	24.72	31.12	28.24	26.12

Table 1: Comparison with prior pruning methods at low (

10\%

), medium (

20\%

), and high (

30\%

) pruning ratios on four MPEG CTC scenes. All metrics are computed between the pruned render and the non-pruned render (PSNR in dB, higher PSNR/SSIM better, lower LPIPS better). Methods are evaluated on held-out MPEG CTC camera views.

Dataset	Metric	No Descriptors		With Descriptors
		No Beta	Beta	No Beta	Beta
bartender (tracked)	PSNR (↑)	24.91	29.23	23.88	31.07
	SSIM (↑)	0.955	0.979	0.948	0.982
	LPIPS (↓)	0.055	0.035	0.074	0.034
breakfast (tracked)	PSNR (↑)	19.98	28.90	17.94	34.47
	SSIM (↑)	0.933	0.978	0.903	0.979

Table 2: Ablation over descriptor use and Beta modelling, at mid pruning (

20\%

). All metrics are computed between the pruned and non-pruned renders (higher PSNR/SSIM better, lower LPIPS better). Bold marks the best value per row.

Prune %	No Descriptors			With Descriptors			$\Delta$ PSNR
	PSNR ( $\uparrow$ )	SSIM ( $\uparrow$ )	LPIPS ( $\downarrow$ )	PSNR ( $\uparrow$ )	SSIM ( $\uparrow$ )	LPIPS ( $\downarrow$ )	(dB)
$30\%$	35.75	0.967	0.039	37.38	0.974	0.030	$-1.63$
$50\%$	32.79	0.944	0.061	33.68	0.948	0.054	$-0.89$
$70\%$	28.05	0.871	0.115	29.07	0.882	0.103	$-1.02$

Table 3: Pruning-fidelity across three keep-fractions on the

14

-scene MPEG CTC evaluation subset (

527

image pairs per cell; all held-out views at

1024

px longest side). Same evaluation criteria as Tables 1 – 2 : PSNR in dB, SSIM in

[0,1]

, LPIPS (VGG).

Figure 3: Qualitative comparison of camera-agnostic pruning on three MPEG CTC scenes ( bartender (tracked) , breakfast (tracked) , cinema (tracked) ) at approximately 20% pruning. The columns show (left to right) the input rendering, descriptor-only pruning without Beta modelling (No Beta), and uncertainty-aware pruning with Beta evidence (Beta). Incorporating Beta evidence improves preservation of fine structures and reduces visible pruning artefacts.

3D Gaussian Pruning

3D Gaussian pruning removes redundant primitives from 3D Gaussian Splatting scenes to cut storage, memory, and rendering cost while preserving visual quality.

See all 44 papers

Prune Wisely, Reconstruct Sharply: Compact 3D Gaussian Splatting via Adaptive Pruning and Difference-of-Gaussian Primitives
Feb 27, 2026Haoran Wang, Guoxi Huang, Fan Zhang +23D Gaussian Splatting 3D Gaussian Pruning
School of Computer Science, University of Bristol, Bristol, UK
REFINE: Super-efficient 3D Gaussian Splatting Pruning via Rendering-Free Primitive Importance
Jun 8, 2026Zhang Chen, Shuai Wan, Mengting Yu +23D Gaussian Splatting 3D Gaussian Pruning
School of Electronics and Information, Northwestern Polytechnical University · School of Telecommunication Engineering, Xidian University · Department of Computer Science, City University of Hong Kong
SplatQ: Interaction-Aware Pruning of 3D Gaussian Splatting via Quadratic Selection
Oct 9, 2026Hongbin Ye, Haixu Liu, Shenwei Kang +133D Gaussian Pruning 3D Gaussian Splatting

Feb 27, 2026J/K move · Enter open · S save

Prune Wisely, Reconstruct Sharply: Compact 3D Gaussian Splatting via Adaptive Pruning and Difference-of-Gaussian Primitives

Haoran Wang, Guoxi Huang, Fan Zhang, David Bull+1

School of Computer Science, University of Bristol, Bristol, UK

3D Gaussian Splatting 3D Gaussian Pruning 3DGS Compression

Open paper AskPDF arXiv

Recent significant advances in 3D scene representation have been driven by 3D Gaussian Splatting (3DGS), which has enabled real-time rendering with photorealistic quality. 3DGS often requires a large number of primitives to achieve high fidelity, leading to redundant representations and high resource consumption, thereby limiting its scalability for complex or large-scale scenes. Consequently, effective pruning strategies and more expressive primitives that can reduce redundancy while preserving visual quality are crucial for practical deployment. We propose an efficient, integrated reconstruction-aware pruning strategy that adaptively determines pruning timing and refining intervals based on reconstruction quality, thus reducing model size while enhancing rendering quality. Moreover, we introduce a 3D Difference-of-Gaussians primitive that jointly models both positive and negative densities in a single primitive, improving the expressiveness of Gaussians under compact configurations. Our method significantly improves model compactness, achieving up to 90% reduction in Gaussian-count while delivering visual quality that is similar to, or in some cases better than, that produced by state-of-the-art methods. Code will be made publicly available.

Abstract

3D Gaussian Pruning

Explore similar work

Prune Wisely, Reconstruct Sharply: Compact 3D Gaussian Splatting via Adaptive Pruning and Difference-of-Gaussian Primitives

REFINE: Super-efficient 3D Gaussian Splatting Pruning via Rendering-Free Primitive Importance

SplatQ: Interaction-Aware Pruning of 3D Gaussian Splatting via Quadratic Selection