RAVE: Rate-Adaptive Visual Encoding for 3D Gaussian Splatting

TL;DR

RAVE introduces a rate-adaptive compression framework for 3D Gaussian Splatting that delivers a continuous rate–distortion curve from a single training run, enabling on-the-fly bitrate adjustments without retraining. It ranks Gaussians by local gradient importance within anchor contexts to form target subsets for any desired rate, and uses codec-agnostic entropy coding to reconstruct the scene. The approach achieves competitive or state-of-the-art performance on major 3DGS benchmarks, notably attaining a continuous RD curve where prior methods require separate models for each rate. This work enables practical, bandwidth-aware deployment of high-quality 3D scene representations in immersive applications.

Abstract

Recent advances in neural scene representations have transformed immersive multimedia, with 3D Gaussian Splatting (3DGS) enabling real-time photorealistic rendering. Despite its efficiency, 3DGS suffers from large memory requirements and costly training procedures, motivating efforts toward compression. Existing approaches, however, operate at fixed rates, limiting adaptability to varying bandwidth and device constraints. In this work, we propose a flexible compression scheme for 3DGS that supports interpolation at any rate between predefined bounds. Our method is computationally lightweight, requires no retraining for any rate, and preserves rendering quality across a broad range of operating points. Experiments demonstrate that the approach achieves efficient, high-quality compression while offering dynamic rate control, making it suitable for practical deployment in immersive applications. The code will be provided open-source upon acceptance of the work.

Figures (4)

• Figure 1: We present RAVE, the first method enabling rate-adaptive visual encoding for 3DGS approaches. Unlike existing methods that require multiple separate training runs, RAVE produces a continuous rate–distortion curve in a single, efficient end-to-end training. This allows seamless adaptation of the bitrate to different constraints without retraining, offering both state-of-the-art quality and practical deployment.
• Figure 2: Overview of our method. We compute per-anchor gradient scores once and rank Gaussians by importance. For any target rate $\mathcal{R}^{\text{target}}$, the top-ranked Gaussians are selected to form $\mathcal{G}_{\text{target}}$, then compressed and decoded to reconstruct the scene. This enables multiple operating points and a continuous rate–distortion curve from a single trained model.
• Figure 3:
• Figure 4: Comparison with a global variant of our method, where the gradient is computed globally before each pruning stage, and with a naïve multi-anchors strategy ($50$ levels). Our method highlights the importance of context, while at the same time being superior to multi-anchor strategy.