HEMGS: A Hybrid Entropy Model for 3D Gaussian Splatting Data Compression

TL;DR

This work introduces HEMGS, a hybrid entropy model for 3D Gaussian Splatting data that enables variable-rate lossy compression within a single model and enhanced lossless compression through a joint hyperprior-autoregressive framework. By integrating a variable-rate predictor, a scene-aware hyperprior, and an adaptive-context autoregressive network, HEMGS reduces storage while preserving rendering quality across multiple 3DGS benchmarks. The approach demonstrates state-of-the-art compression performance, achieving about 40% average size reduction over baselines and offering practical benefits for bandwidth-constrained transmission and storage. The results suggest robust applicability to anchor-based 3DGS representations, with future work extending to anchor-free data structures.

Abstract

In this work, we propose a novel compression framework for 3D Gaussian Splatting (3DGS) data. Building on anchor-based 3DGS methodologies, our approach compresses all attributes within each anchor by introducing a novel Hybrid Entropy Model for 3D Gaussian Splatting (HEMGS) to achieve hybrid lossy-lossless compression. It consists of three main components: a variable-rate predictor, a hyperprior network, and an autoregressive network. First, unlike previous methods that adopt multiple models to achieve multi-rate lossy compression, thereby increasing training overhead, our variable-rate predictor enables variable-rate compression with a single model and a hyperparameter $λ$ by producing a learned Quantization Step feature for versatile lossy compression. Second, to improve lossless compression, the hyperprior network captures both scene-agnostic and scene-specific features to generate a prior feature, while the autoregressive network employs an adaptive context selection algorithm with flexible receptive fields to produce a contextual feature. By integrating these two features, HEMGS can accurately estimate the distribution of the current coding element within each attribute, enabling improved entropy coding and reduced storage. We integrate HEMGS into a compression framework, and experimental results on four benchmarks indicate that HEMGS achieves about a 40% average reduction in size while maintaining rendering quality over baseline methods and achieving state-of-the-art compression results.

Figures (5)

• Figure 1: The overview of our 3DGS data compression framework, which integrates HEMGS with other additional coding components.
• Figure 2: (a) The details of our HEMGS, which comprises a variable-rate predictor, a hyperprior network, and an autoregressive network. (b) The autoregressive network utilizes adaptive receptive fields to reduce redundancies within each attribute. (c) The hyperprior network incorporates scene-agnostic and scene-specific architectures to reduce redundancies across attributes.
• Figure 3: The Rate-Distortion (RD) curves on three benchmark datasets, including Mip-NeRF360, Tank&Temples, and DeepBlending.
• Figure 4: The Rate-Distortion (RD) curves of our variable-rate HEMGS method for compressing 3DGS data using a single model on the Tank&Temples dataset. "HEMGS w/o Variable-rate" represents our HEMGS without using Variable-rate Predictor, which needs a separate model for each rate point.
• Figure 5: Visualization comparison between HAC and our newly proposed HEMGS on the "bicycle" scene from the Mip-NeRF360 dataset. PSNR, and SSIM of the rendered image as well as the size (MB) of the scene are reported.