Gaussian Splatting-based Volumetric Video Compression with Sparse 4D Anchors

Ge Gao; Siyue Teng; Chanqgi Wang; Fan Zhang; Nantheera Anantrasirichai; Jui Chiu Chiang; Wen-Hsiao Peng; David Bull

cs.CVSep 27, 2026

Gaussian Splatting-based Volumetric Video Compression with Sparse 4D Anchors

Authors: Ge Gao, Siyue Teng, Chanqgi Wang, Fan Zhang, Nantheera Anantrasirichai, Jui Chiu Chiang, Wen-Hsiao Peng, David Bull

Organizations: Visual Information Lab, University of Bristol, UK · National Chung Cheng University, Taiwan · National Yang Ming Chiao Tung University, Taiwan

Abstract

Immersive video communication requires photorealistic, render-efficient, and compact dynamic scene representations. 3D Gaussian Splatting (3DGS) offers a promising representation, but dynamic 3DGS remains difficult to compress due to dense primitives and spatiotemporal redundancy. Anchor-based formulations improve compactness with sparse scaffolds that share geometry and appearance across primitives. However, existing designs often rely on deforming a single canonical scaffold and condition each primitive on its associated anchor in isolation, limiting their ability to handle non-local dynamics and disocclusion while under-exploiting inter-anchor correlations, particularly in motion- or texture-dense regions. To address these limitations, we propose SAGA, a volumetric video codec built upon Sparse Anchor-assisted GAussian splatting representations. SAGA represents dynamic 3D scenes using hierarchically organized sparse 4D anchors, where coordinate-based INR decoders generate fine anchors and Gaussian primitives from inter-anchor interpolations, enabling compact parameter sharing across spatiotemporal structures. For long-range dependencies among unstructured anchors, we further introduce fixed-size memory slots with orthogonality-informed updates for accurate entropy-context modeling. Experiments show that SAGA achieves strong rate-distortion performance against GIFStream, with PSNR BD-rate reductions of 80.39% and 83.94% on Neu3D and MPEG MIV, respectively.

Figures & tables

Panel of Figure 1 — Figure 1 : ( Left ) Visual comparison on the examples from Neu3D and MPEG MIV datasets. ( Right ) RD performance on Neu3D, Panoptic Sports, and MPEG MIV in terms of PSNR.

Figure 2 : Overview of SAGA. In the representation stage, coarse anchors provide shared spatiotemporal context for fine anchors through localized kernel interpolation, while fine anchors generate Gaussian primitives via INR decoders and residual hash refinement. In the compression stage, anchors, hash grids, and neural decoder parameters are quantized and entropy coded with a hierarchical context model, where a recurrent memory bank captures long-range dependencies across sequential coding groups for improved compression efficiency.

	Neu3D Li et al. (2022)						Panoptic Sports Joo et al. (2017)
Method	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	Train $\downarrow$	FPS $\uparrow$	Size (MB) $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	Train $\downarrow$	FPS $\uparrow$	Size (MB) $\downarrow$
3DGStream Sun et al. (2024) (CVPR’24)	32.65	0.947	0.201	0.806	218	67	21.11	0.720	0.448	0.213	389	97
4DGS Wu et al. (2024a) (CVPR’24)	31.57	0.993	0.0572	120	228	202	28.68	0.911	0.157	–	–	974
E-D3DGS Bae et al. (2024) (ECCV’24)	31.42	0.945	0.037	–	–	137	25.61	0.896	0.172	–	50	297.9
STG Li et al. (2024d) (CVPR’24)	32.05	0.946	0.050	0.77	140	200	25.09	0.900	0.181	0.41	265	181
GIFStream Li et al. (2025) (CVPR’25)	31.75	0.938	0.051	0.991	95	50	28.03	0.9160	0.0868	0.731	141	53

Table 1 : Novel View Synthesis. We report PSNR, SSIM, LPIPS Zhang et al. (2018) , training time in hours, decoding FPS, and model size on Neu3D and Panoptic Sports. Missing entries are denoted by “–” when results or implementations are unavailable. For a fair comparison, we report the raw model size for GIFStream and the proposed SAGA, rather than the size after quantization and entropy coding.

	Neu3D Li et al. (2022)				Panoptic Sports Joo et al. (2017)				MPEG MIV Boyce et al. (2021)
Method	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	FPS $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	FPS $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	FPS $\uparrow$
TMIV-24.0 Boyce et al. (2021)	-15.27%	-19.81%	-13.39%	35	-1.88%	-18.95%	-20.77%	45	+1.12%	-4.23%	-3.48%	42
4DGC Hu et al. (2025a) (CVPR’25)	-62.39%	-41.59%	-61.35%	110	–	–	–	–	-32.38%	-36.13%	-42.61%	115
GIFStream Li et al. (2025) (CVPR’25)	-80.39%	-88.21%	-85.01%	95	-75.88%	-77.31%	-76.95%	141	-83.94%	-86.20%	-86.74%	100
SAGA (Ours)	0.00%	0.00%	0.00%	89	0.00%	0.00%	0.00%	117	0.00%	0.00%	0.00%	91

Table 2 : 4D Volumetric Video Compression. Comparison on Neu3D, Panoptic Sports, and MPEG MIV, evaluated in terms of PSNR, SSIM, LPIPS, and decoding FPS. Each BD-rate value is computed using the corresponding benchmark codec as the anchor. We only report baselines for which BD-rate can be reliably computed (i.e., rate-distortion ranges overlap sufficiently for BD-rate calculations).

Figure 5 : RD comparison on Neu3D, Panoptic Sports and MPEG MIV in SSIM and LPIPS.

Component	Before EC MB (%)	After EC MB (%)	Runtime
Anchor	8.84 (61.2)	0.98 (66.2)	N/A
Hash	2.16 (15.0)	0.33 (22.3)	2.7ms
MLPs	3.44 (23.8)	0.17 (11.5)	2.6ms
Entropy	N/A	N/A	4.6ms
Total	14.44	1.48	9.9ms

Table 3 : Ablation, memory-slot analysis, and bitstream/runtime breakdown on Neu3D. BD-rate is computed in terms of PSNR using SAGA as the anchor.

Explore similar work

ACA-GS: Adaptive-Capacity Anchored Gaussian Splatting for Compact Dynamic Radiance Fields
Aug 5, 2026Seunghyeon Song, Joo Chan Lee, Chanung Park +4Gaussian Splatting 4D Reconstruction
Sungkyunkwan University Suwon, Gyeonggi-do Republic of Korea · Electronics and Telecommunications Research Institute Daejeon, Republic of Korea · Yonsei University Seoul, Republic of Korea
CoSAG: Compact Semantic Anchor Gaussians via Training-Free Rate-Distortion Coding
Jul 11, 2026Yuang Jia, Jinlong Wang, Junhong Lin +23D Scene Understanding Gaussian Primitives
SECE, Peking University · University of Electronic Science and Technology of China
ATGS: Anchored Temporal Gaussian Splatting for Long Volumetric Video Representation
Aug 31, 2026Jiahao Wu, Jie Liang, Die Hu +6Volumetric Video Gaussian Primitives
Guangdong Provincial Key Laboratory of Ultra High Definition Immersive Media Technology, Shenzhen Graduate School, Peking University, Pengcheng Laboratory, China · Pengcheng Laboratory, China · Shenzhen Graduate School, Peking University, Pengcheng Laboratory, China +1

Aug 5, 2026, cs.CVJ/K move · Enter open · S save

ACA-GS: Adaptive-Capacity Anchored Gaussian Splatting for Compact Dynamic Radiance Fields

Seunghyeon Song, Joo Chan Lee, Chanung Park, Jun Young Jeong+3

Sungkyunkwan University Suwon, Gyeonggi-do Republic of Korea · Electronics and Telecommunications Research Institute Daejeon, Republic of Korea · Yonsei University Seoul, Republic of Korea

Gaussian Splatting 4D Reconstruction Capacity Expressivity

Open paper AskPDF arXiv

Recent advances in 4D Gaussian Splatting (4DGS) enable high-fidelity, real-time spatiotemporal rendering, but expose a fundamental trade-off between motion expressiveness and storage efficiency. While anchor-based designs achieve compactness through anchor-level parameter sharing, their rigid uniform parametrization enforces fixed Neural Gaussian counts and feature budgets per anchor. Consequently, insufficient fidelity is addressed by excessive anchor density, rather than lightweight, targeted increases in Neural Gaussian count or feature capacity, resulting in memory waste. To overcome this rigidity, we introduce an adaptive-capacity anchor-based framework that dynamically allocates the representational capacity based on local spatiotemporal demands. Adaptive Anchor Cardinality varies the number of Neural Gaussians per anchor, concentrating primitives in regions of high geometric or motion complexity while suppressing redundancy. In parallel, Adaptive Anchor Feature Masking modulates anchor-level feature channels, assigning rich features to complex regions and lightweight representations to simpler ones. Experiments on MPEG, Panoptic Sports, and N3DV datasets demonstrate substantial storage reduction without degrading visual quality. Notably, on challenging MPEG sequences with complex motion, our method achieves up to 1.5x higher compression than state-of-the-art anchor-based methods while preserving comparable quality.

Abstract

Explore similar work

ACA-GS: Adaptive-Capacity Anchored Gaussian Splatting for Compact Dynamic Radiance Fields

CoSAG: Compact Semantic Anchor Gaussians via Training-Free Rate-Distortion Coding

ATGS: Anchored Temporal Gaussian Splatting for Long Volumetric Video Representation