SymGS : Leveraging Local Symmetries for 3D Gaussian Splatting Compression

2025-11-17

AI Summary

To address the prohibitively high memory consumption (often reaching gigabytes) of 3D Gaussian Splatting in complex scenes, this paper proposes the first symmetry-aware compression framework, leveraging inherent local and global mirror symmetries to eliminate redundant Gaussian primitives. The core innovation is a learnable mirror modeling mechanism that enables end-to-end optimization of symmetry relationships and supports plug-and-play integration with existing compression techniques (e.g., Hierarchical Adaptive Compression, HAC). Unlike prior methods, it requires no manual priors—automatically detecting and parameterizing symmetric structures. Under high-fidelity rendering constraints, our approach significantly improves compression efficiency: achieving an average compression ratio 1.66× higher than HAC, and up to 3× higher on large-scale scenes. When combined with multi-level compression strategies, the overall compression ratio reaches 108×, establishing a new paradigm for efficient and scalable neural rendering.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Learning on the Edge & Model Compression
• Cognitive Modeling & Cognitive Systems: Symbolic Representations

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

3D Gaussian Splatting has emerged as a transformative technique in novel view synthesis, primarily due to its high rendering speed and photorealistic fidelity. However, its memory footprint scales rapidly with scene complexity, often reaching several gigabytes. Existing methods address this issue by introducing compression strategies that exploit primitive-level redundancy through similarity detection and quantization. We aim to surpass the compression limits of such methods by incorporating symmetry-aware techniques, specifically targeting mirror symmetries to eliminate redundant primitives. We propose a novel compression framework, extbf{ extit{SymGS}}, introducing learnable mirrors into the scene, thereby eliminating local and global reflective redundancies for compression. Our framework functions as a plug-and-play enhancement to state-of-the-art compression methods, (e.g. HAC) to achieve further compression. Compared to HAC, we achieve $1.66 imes$ compression across benchmark datasets (upto $3 imes$ on large-scale scenes). On an average, SymGS enables $f{108 imes}$ compression of a 3DGS scene, while preserving rendering quality. The project page and supplementary can be found at extbf{color{cyan}{symgs.github.io}}

Problem

• Compressing memory-intensive 3D Gaussian Splatting scenes efficiently
• Eliminating redundant primitives using local symmetry-aware techniques
• Enhancing existing compression methods while preserving rendering quality

Innovation

• Introducing learnable mirrors to eliminate redundant primitives
• Leveraging local symmetries for 3D Gaussian compression
• Plug-and-play enhancement to existing compression methods