Locality-aware Gaussian Compression for Fast and High-quality Rendering

2025-01-10

AI Summary

To address the high storage overhead and low rendering efficiency of 3D Gaussian Splatting (3DGS), this paper proposes LocoGS—a lightweight neural field compression method leveraging local spatial coherence. Its core innovations include: (i) the first locality-aware 3D Gaussian representation; and (ii) a joint optimization framework integrating dense initialization, adaptive spherical harmonic bandwidth selection, and attribute-specific encoding to simultaneously achieve compact modeling and real-time rendering. On real-world datasets, LocoGS achieves 54.6×–96.6× higher compression ratios than standard 3DGS, while accelerating rendering by 2.1×–2.4×. Compared to state-of-the-art compression methods, it delivers an average 2.4× rendering speedup with comparable compression performance. LocoGS thus uniquely balances high compression ratio, high-fidelity reconstruction, and efficient rasterization—enabling scalable, high-performance neural rendering.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Learning on the Edge & Model Compression
• Search and Optimization: Local Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

We present LocoGS, a locality-aware 3D Gaussian Splatting (3DGS) framework that exploits the spatial coherence of 3D Gaussians for compact modeling of volumetric scenes. To this end, we first analyze the local coherence of 3D Gaussian attributes, and propose a novel locality-aware 3D Gaussian representation that effectively encodes locally-coherent Gaussian attributes using a neural field representation with a minimal storage requirement. On top of the novel representation, LocoGS is carefully designed with additional components such as dense initialization, an adaptive spherical harmonics bandwidth scheme and different encoding schemes for different Gaussian attributes to maximize compression performance. Experimental results demonstrate that our approach outperforms the rendering quality of existing compact Gaussian representations for representative real-world 3D datasets while achieving from 54.6$ imes$ to 96.6$ imes$ compressed storage size and from 2.1$ imes$ to 2.4$ imes$ rendering speed than 3DGS. Even our approach also demonstrates an averaged 2.4$ imes$ higher rendering speed than the state-of-the-art compression method with comparable compression performance.

Problem

• 3D image compression
• efficiency improvement
• storage reduction

Innovation

• 3DGS Framework
• Neural Network Image Compression
• Adaptive Spherical Pattern Bandwidth