LG-HCC: Local Geometry-Aware Hierarchical Context Compression for 3D Gaussian Splatting

Xuan Deng; Xiandong Meng; Hengyu Man; Qiang Zhu; Tiange Zhang; Debin Zhao; Xiaopeng Fan

cs.CVMar 30, 2026

LG-HCC: Local Geometry-Aware Hierarchical Context Compression for 3D Gaussian Splatting

Authors: Xuan Deng, Xiandong Meng, Hengyu Man, Qiang Zhu, Tiange Zhang, Debin Zhao, Xiaopeng Fan

Organizations: Harbin Institute of Technology, Peng Cheng Laboratory Shenzhen, China · Peng Cheng Laboratory Shenzhen, China · Harbin Institute of Technology Harbin, China · Harbin Institute of Technology, Peng Cheng Laboratory, Harbin Institute of Technology Suzhou Research Institute Harbin, China

Abstract

Although 3D Gaussian Splatting (3DGS) enables high-fidelity real-time rendering, its prohibitive storage overhead severely hinders practical deployment. Recent anchor-based 3DGS compression schemes reduce gaussian redundancy through some advanced context models. However, they overlook explicit geometric dependencies, leading to structural degradation and suboptimal ratedistortion performance. In this paper, we propose a Local Geometry-aware Hierarchical Context Compression framework for 3DGS(LG-HCC) that incorporates inter-anchor geometric correlations into anchor pruning and entropy coding for compact representation. Specifically, we introduce an Neighborhood-Aware Anchor Pruning (NAAP) strategy, which evaluates anchor importance via weighted neighborhood feature aggregation and then merges low-contribution anchors into salient neighbors, yielding a compact yet geometry-consistent anchor set. Moreover, we further develop a hierarchical entropy coding scheme, in which coarse-to-fine priors are exploited through a lightweight Geometry-Guided Convolution(GG-Conv) operator to enable spatially adaptive context modeling and rate-distortion optimization. Extensive experiments show that LG-HCC effectively alleviates structural preservation issues,achieving superior geometric integrity and rendering fidelity while reducing storage by up to 30.85x compared to the Scaffold-GS baseline on the Mip-NeRF360 dataset

Figures & tables

Figure 1. Trade-off between fidelity (PSNR) and rendering speed (FPS) for different methods on the BungeeNeRF dataset ( Xiangli et al., 2022 ) . Circle size denotes storage cost, where smaller circles correspond to lower storage.

Figure 2. Overview of the LG-HCC framework. Following Colmap initialization and densification, Neighborhood-Aware Anchor Pruning (NAAP) adaptively merges low-contribution anchors into salient neighbors via a local graph to ensure geometric consistency. The preserved anchors then undergo Hierarchical Geometry-Guided Context Modeling for efficient entropy coding. Finally, decoded attributes generate 3D Gaussians via learnable offsets for high-fidelity rasterization.

Figure 3. Neighborhood-Aware Anchor Pruning (NAAP) pipeline starts by constructing an anchor-based geometric graph, with nodes colored by average opacity (darker = higher). Importance scores (

\xi_{i}

) are computed via neighborhood aggregation. Low-score anchors are identified as redundant and associated with their nearest salient neighbor (Local Neighbor Preserver). Instead of direct removal, a Weighted Attribute Transformation merges attributes from pruned anchors (opacity

\bar{\alpha}_{i}

, scaling

s_{i}

, offset

o_{i}

) into the nearest retained anchor, yielding fused attributes (

\phi_{i}

O_{i}

S_{i}

, etc.) to preserve geometric and appearance in a compact structure.

Figure 4. This module exploits cross-level correlations by mapping Level 1 attributes to Level 2 query anchors via a deterministic mapping

\mathcal{M}

to obtain the preliminary attributes

\mathcal{A}_{pre}^{(2)}

. Then, a

k

-NN graph is constructed over

\mathcal{A}_{pre}^{(2)}

to form the graph-based geometric prior

\mathcal{G}_{geo}^{(2)}

. Within the GG-Conv (dashed box): (1) Geometry branch: Relative offsets

\Delta\mathbf{p}_{ij}

query a learnable 3D weight table via trilinear interpolation to generate dynamic weights

\mathbf{w}_{ij}

encoding the spatial layout. (2) Feature branch: Residual features

\Delta\mathbf{f}_{ij}^{(1)}

and offsets

\Delta\mathbf{p}_{ij}

are transformed by MLP

\phi

into embeddings

\mathbf{e}_{ij}

. Finally,

\mathbf{w}_{ij}

modulates

\mathbf{e}_{ij}

to produce the local geometry-aware feature

\mathbf{h}_{i}

, which serves as the final context

\mathbf{a}_{i(ctx)}

for Level 2 entropy coding.

Figure 5. Qualitative results of the proposed LG-HCC method compared to existing three compression methods on the drjohnson (DeepBlending), train (Tank&Temples) and bicycile (Mip-NeRF360) scenes, respectively.

Datasets	Mip-NeRF360 ( Barron et al., 2022 )				BungeeNeRF ( Xiangli et al., 2022 )				DeepBlending ( Hedman et al., 2018 )				Tank&Temples ( Knapitsch et al., 2017 )
Methods	PSNR(dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	size $\downarrow$	PSNR(dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	size $\downarrow$	PSNR(dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	size $\downarrow$	PSNR(dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	size $\downarrow$
3DGS ( Bagdasarian et al., )	27.49	0.813	c0.222	744.7	24.87	0.841	0.205	1616	29.42	0.899	0.247	663.9	23.69	0.844	0.178	431.0
Scaffold-GS ( Lu et al., 2024 )	27.50	0.806	0.252	253.9	26.62	0.865	0.241	183.0	30.21	0.906	0.254	66.00	23.96	0.853	0.177	86.50
Compact3DGS ( Lee et al., 2024 )	27.08	0.798	0.247	48.80	23.36	0.788	0.251	82.60	29.79	0.901	0.258	43.21	23.32	0.831	0.201	39.43
Compressed3D ( Navaneet et al., 2023 )	26.98	0.801	0.238	28.80	24.13	0.802	0.245	55.79	29.38	0.898	0.253	25.30	23.32	0.832	0.194	17.28
Morgen. et al. ( Morgenstern et al., 2024 )	26.01	0.772	0.259	23.90	22.43	0.708	0.339	48.25	28.92	0.891	0.276	8.40	22.78	0.817	0.211	13.05

Table 1. Comparison against existing compression approaches. Our method is evaluated under two settings: a high-fidelity mode and a high-compression mode , showcasing its versatility. Cells colored Red and yellow highlight the best-performing and second-best results, respectively, for both the low-bitrate and high-bitrate regimes. All size are reported in MB.

Figure 6. Rate-Distortion (RD) performance curves. Comparison with advanced compression techniques (HAC, HAC++, Context-GS, CAT-3DGS) on the Mip-NeRF360, BungeeNeRF, and Tank&Temples datasets. LG-HCC demonstrates robust coding efficiency, maintaining the upper envelope across a wide range of bitrates.

Methods	Storage Costs (MB) $\downarrow$					Fidelity			Rendering
Methods	Positions	Features	MLPs	Others	Total	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	FPS
Vanilla EM ( Wang et al., 2024b )	5.24	8.27	0.3162	11.24	25.07	25.07	0.7391	0.2683	135
GG-Conv-based EM (two level)	4.69	4.51	0.2381	6.31	15.75	25.16	0.7415	0.2704	170
GG-Conv-based EM (three level)	4.66	6.82	0.3162	7.54	19.34	25.10	0.7416	0.2703	160

Table 2. Comparison of storage breakdown (Positions, Features, MLP, others; "others" includes auxiliary structures and metadata (e.g., scalings, offsets, masks).), and rendering FPS on the "Bicycle" scene between Context-GS (vanilla EM, i.e., Entropy Modeling) and our GG-Conv-based Hierarchical Geometry-Guided Context EM.

Method	PSNR (dB) $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	size ( MB) $\downarrow$
Ours w/o NAAP & GG-Conv	30.10	0.9062	0.2659	3.76
Ours w/o GG-Conv	30.13	0.9087	0.2656	3.72
Ours w/o NAAP	30.20	0.9095	0.2648	3.58
Ours (full)	30.31	0.9106	0.2636	3.47

Table 3. Quantitative ablation study on the Deep Blending dataset.

Method	Positions (MB) $\downarrow$	Features (MB) $\downarrow$
Vanilla Pruning	0.7531	1.1108
NAAP (Ours)	0.7689 (+2.1%)	1.0847 (-2.4%)

Table 4. Storage comparison between vanilla pruning (Context-GS ( Wang et al., 2024b ) ) and our LG-HCC method.

3DGS Compression

3D Gaussian Splatting compression reduces storage and rendering costs while preserving the quality of reconstructed views and scenes.

See all 80 papers

Towards Practical Compression of 3D Gaussian Splatting
Sep 24, 2026Pengpeng Yu, Yueru Chen, Fei Song +43DGS Compression Entropy Coding
Sun Yat-sen University, China · Pengcheng Laboratory, China · Academy of Broadcasting Science, National Radio and Television Administration, China +1
Compressing 3D Gaussian Splatting via Cross-Representation Priors
Sep 19, 2026Yezheng Zhang, Huanxiong Liang, Chuqin Zhou +23DGS Compression Gaussian Splatting Compression
Institute of Image Communication and Network Engineering, Shanghai Jiao Tong University, Shanghai 200240, China
MMGS: 10 $\times$ Compressed 3DGS through Optimal Transport Aggregation based on Multi-view Ranking
May 19, 2026Beizhen Zhao, Sicheng Yu, Ziran Yin +23DGS Compression 3D Gaussian Splatting
The Hong Kong University of Science and Technology (Guangzhou)

Sep 24, 2026J/K move · Enter open · S save

Towards Practical Compression of 3D Gaussian Splatting

Pengpeng Yu, Yueru Chen, Fei Song, Tai Qin+3

Sun Yat-sen University, China · Pengcheng Laboratory, China · Academy of Broadcasting Science, National Radio and Television Administration, China+1

3DGS Compression Entropy Coding Gaussian Splatting Compression

Open paper AskPDF arXiv

3D Gaussian Splatting (3DGS) enables high-quality novel-view synthesis but requires substantial storage. Existing compression methods often rely on spatial context modeling over irregular 3D representations, increasing the complexity of training and coding. Meanwhile, floating-point context inference can introduce numerical inconsistencies across platforms, causing entropy-decoding failures. To address these practical challenges, we propose COSA-GS, which constructs context without spatial aggregation through anchor-wise causal factorization. Specifically, we use geometry context derived from each anchor's coordinates to model a compact learnable anchor latent. The anchor latent is then fused with the geometry context to form an anchor context for attribute coding. The resulting context model features a simple architecture composed solely of linear transformations and activations. We train COSA-GS using rate--distortion optimization with adaptive Gaussian pruning. Further, we develop quantization-aware training and integer inference for the context model to achieve bit-exact consistency of entropy-decoded symbols across platforms. Experiments demonstrate that COSA-GS achieves state-of-the-art compression performance while retaining fast and consistent cross-platform decoding, providing a simple yet effective framework for practical 3DGS compression. Code is available at https://github.com/pengpeng-yu/COSA-GS.

Abstract

3DGS Compression

Explore similar work

Towards Practical Compression of 3D Gaussian Splatting

Compressing 3D Gaussian Splatting via Cross-Representation Priors

MMGS: 10×\times× Compressed 3DGS through Optimal Transport Aggregation based on Multi-view Ranking

MMGS: 10 $\times$ Compressed 3DGS through Optimal Transport Aggregation based on Multi-view Ranking