Smaller and Faster 3DGS via Post-Training Dictionary Learning

2026-05-28

AI Summary

This work addresses the challenge of deploying 3D Gaussian Splatting (3DGS) models on resource-constrained devices due to their high memory footprint. Existing compression approaches often compromise rendering quality or require retraining the model. To overcome these limitations, we propose the first post-training compression framework based on dictionary learning, which enables sparse coding and efficient compression of 3DGS parameters without modifying or retraining the original model. By introducing dictionary learning into 3DGS compression for the first time, our method achieves a favorable balance of generality, high compression ratio, and fidelity preservation. Evaluated across 13 benchmark scenes, the approach yields an average compression ratio of 3.95× and accelerates rendering by over 23% while maintaining high visual quality.

Abstract

3D Gaussian Splatting (3DGS) is a promising neural scene representation for real-time rendering, but trained models often suffer from large memory footprints, limiting deployment on less powerful devices. Existing compression techniques often lead to architectures with several additional trainable parameters. While achieving outstanding compression ratios, they introduce noticeable drops in image quality. In this work, we introduce the first dictionary-learning-based compression framework for 3DGS. The proposed post-training compression pipeline can be deployed in virtually any 3DGS model without the need for re-training or modifications to existing 3DGS models. Our compression framework is straightforward to implement, yet provides significant compression capabilities, preserves image quality, and improves real-time rendering performance. Across 13 benchmark scenes, our approach achieves an average compression ratio of 3.95x, 3.10x, and 4.55x when applied to 3DGS, 3DGS-MCMC, and PixelGS, respectively. This yields consistent rendering speedups of 23.3%, 24.3%, and 25.3%, while maintaining image quality.

Problem

• 3D Gaussian Splatting
• model compression
• memory footprint
• image quality
• real-time rendering

Innovation

• dictionary learning
• post-training compression
• 3D Gaussian Splatting
• model compression
• real-time rendering