BudSplat: Feed-forward 3D Gaussian Splatting under a Rendering Budget

Muyu Xu; Fangneng Zhan; Yu Wei; Ling Shao; Shijian Lu

cs.CVOct 28, 2026

BudSplat: Feed-forward 3D Gaussian Splatting under a Rendering Budget

Authors: Muyu Xu, Fangneng Zhan, Yu Wei, Ling Shao, Shijian Lu

Organizations: Nanyang Technological University · Hong Kong University of Science and Technology · UCAS-Terminus AI Lab

Abstract

Feed-forward 3D Gaussian Splatting has recently emerged as a promising alternative to per-scene optimization, enabling fast novel view synthesis by predicting Gaussian primitives directly from input images. However, it quickly faces a bottleneck of rendering efficiency while scaling from sparse to dense multi-view inputs, which offer more geometric constraints but produce a rapidly growing set of candidate Gaussians. Existing solutions cannot reliably distinguish redundant primitives, leading to many redundant primitives from repeatedly observed regions and increased rendering cost without much gain in visual quality. We propose BudSplat, a compact feed-forward 3DGS framework that learns to keep the most valuable Gaussians under a limited primitive count. BudSplat has two novel designs. The first is a detail-aware importance score that is extracted from multi-level visual tokens, providing an explicit cue for regions where pruning is likely to harm perceptual quality. The score can be injected into voxel fusion to preserve high-value local contributors during multi-view aggregation. The second is quality-aware Gaussian pruning that selects a compact set of primitives by jointly considering rendering contribution and detail importance. The quality-aware pruning enables local quota redistribution that helps avoid over-allocating primitives to large redundant regions. Experiments on RE10K, DL3DV, and ACID show that BudSplat achieves a favorable trade-off among rendering quality and efficiency.

Figures & tables

Figure 1: Overview of BudSplat. With dense geometry, confidence maps, and multi-layer tokens as predicted from multi-view RGB images by VGGT, the detail-aware decomposition first extracts residual token cues and predicts patch-level detail scores. The predicted scores are then propagated to the importance-preserving voxel fusion together with confidence and voxel support, retaining important local contributors during aggregation. Finally, the quality-aware pruning ranks the fused Gaussians by opacity, detail importance, and support-aware redundancy correction, followed by patch-level quota redistribution to select a compact set of retained Gaussian for rendering.

Dataset	Method	16 Views				32 Views				72 Views
Dataset	Method	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#GS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#GS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#GS $\downarrow$
RE10K	VolSplat (posed)	23.63	0.838	0.200	1605K	22.06	0.791	0.245	3210K	OOM	OOM	OOM	OOM
	GGN	19.57	0.738	0.276	425K	19.11	0.725	0.285	586K	OOM	OOM	OOM	OOM
	AnySplat	21.72	0.724	0.187	1298K	22.69	0.756	0.174	1363K	23.48	0.789	0.164	3461K
	WorldMirror	22.37	0.768	0.194	1072K	22.85	0.786	0.190	1782K	22.92	0.791	0.189	3034K
	Ours ρ=5%	22.48	0.768	0.193	65K	23.08	0.791	0.181	97K	23.52	0.803	0.179	173K

Table 1: Main quantitative comparison under increasing input views. We report PSNR, SSIM, LPIPS, and the number of rendered Gaussians (#GS). For our method, Ours ρ denotes retaining a

\rho

-fraction of fused Gaussians after pruning. Pink and orange cells indicate the best and second-best results under each dataset and view setting, following the metric direction indicated by the arrows. OOM denotes out-of-memory during inference.

Figure 2: Qualitative comparison on challenging low-overlap target views. Red boxes highlight regions where baselines suffer from incomplete Gaussian coverage, blurred local structures, or uneven rendering quality. Our method allocates Gaussian budget more effectively, producing stable renderings and preserving perceptually important details with fewer Gaussians.

Variant	DT	IVF	QAP	PQ	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
Budgeted AnySplat-style baseline					13.23	0.522	0.686
w/o detail-token importance		✓	✓	✓	20.04	0.592	0.476
w/o importance-preserving voxel fusion	✓		✓	✓	22.47	0.768	0.214
w/o quality-aware pruning score	✓	✓		✓	22.93	0.784	0.209
w/o patch quota redistribution	✓	✓	✓		22.55	0.776	0.214
Full model	✓	✓	✓	✓	23.02	0.791	0.201

Table 2: Ablation of the main design modules under the same Gaussian budget. DT denotes detail-token importance estimation. IVF denotes importance-preserving voxel fusion. QAP denotes quality-aware pruning. PQ denotes patch-level quota redistribution.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Dataset	Method	20 Views				24 Views				48 Views
Dataset	Method	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#GS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#GS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	#GS $\downarrow$
RE10K	VolSplat (posed)	23.17	0.824	0.221	2007K	22.94	0.814	0.231	2408K	OOM	OOM	OOM	OOM
	GGN	19.43	0.735	0.278	444K	19.33	0.732	0.280	492K	OOM	OOM	OOM	OOM
	AnySplat	22.21	0.737	0.178	1346K	22.35	0.745	0.178	1550K	23.07	0.769	0.167	2574K
	WorldMirror	22.15	0.740	0.194	1340K	22.72	0.773	0.192	1453K	22.93	0.790	0.187	2345K
	Ours ρ=5%	22.71	0.775	0.208	67K	23.02	0.791	0.201	78K	23.23	0.797	0.200	128K

Appendix

Table 3: Additional main quantitative comparison 20, 24, 48 input views. We report PSNR, SSIM, LPIPS, and the number of rendered Gaussians (#GS). For our method, Ours ρ denotes retaining a

\rho

Figure 3: Target-view visualization of Gaussian density allocation and RGB rendering before and after pruning. Top row: projected Gaussian density before pruning, after pruning, removed density, and retention ratio. Bottom row: normal RGB rendering before pruning, after pruning, absolute difference, and reference image. The density diagnostic shows that pruning mainly removes redundant Gaussian support from low-frequency and highly overlapped regions, while preserving more support around high-frequency structures. The RGB difference map further shows that the remaining rendering changes are concentrated around edges, boundaries, and fine details rather than smooth regions.

3DGS Acceleration

3D Gaussian Splatting acceleration reduces the time, memory, and compute needed to train and render Gaussian scene representations.

See all 64 papers

EffGS: Efficient and High-Fidelity Gaussian Splatting
Sep 30, 2026Changbai Li, Shuo Yang, Yichen Yang +23D Gaussian Splatting 3D Gaussian Pruning
Beihang University · Nanyang Technological University
ATSplat: Compact Feed-forward 3D Gaussian Splatting with Adaptive Token Expansion
Jul 22, 2026In Cho, Jeonghwan Cho, Mijin Yoo +23D Gaussian Splatting Adaptive Gaussian Densification
Yonsei University, South Korea · National University of Singapore, Singapore
GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens
Apr 16, 2026Roni Itkin, Noam Issachar, Yehonatan Keypur +33D Gaussian Splatting 3DGS Acceleration
The Hebrew University of Jerusalem · Westlake University

Sep 30, 2026J/K move · Enter open · S save

EffGS: Efficient and High-Fidelity Gaussian Splatting

Changbai Li, Shuo Yang, Yichen Yang, Shuwei Shao+1

Beihang University · Nanyang Technological University

3D Gaussian Splatting 3D Gaussian Pruning 3DGS Acceleration Adaptive Gaussian Densification Novel View Synthesis 3DGS Training

Open paper AskPDF arXiv

3D Gaussian Splatting (3DGS) enables real-time novel view synthesis, but existing general-purpose acceleration methods suffer severe rendering quality degradation when extended to more complex, large-scale scenes. To address this issue, we propose EffGS, a more general acceleration framework that improves training and rendering efficiency while maintaining reconstruction quality comparable to or better than vanilla 3DGS across bounded and large-scale scenes. EffGS combines frequency-aware guidance, localized density control, and adaptive primitive scale modulation. First, an importance scoring mechanism combines pixel-wise reconstruction errors with a difference-of-Gaussians mask scheduled over training to provide stage-dependent spatial guidance. Second, localized densification and pruning restricts density modifications to Gaussians with valid projected footprints in the sampled views. Third, learnable per-Gaussian scale modulation adjusts effective primitive extent during optimization while retaining the Compact Box rasterization rule. Extensive experiments on bounded and large-scale scene datasets demonstrate a favorable balance between reconstruction quality, training time, and primitive count. Component ablations and matched-primitive-budget comparisons further support the effectiveness of the framework.

Abstract

3DGS Acceleration

Explore similar work

EffGS: Efficient and High-Fidelity Gaussian Splatting

ATSplat: Compact Feed-forward 3D Gaussian Splatting with Adaptive Token Expansion

GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens