Problem
Research questions and friction points this paper is trying to address.
3D Gaussian Splatting
compression
storage overhead
cross-platform decoding
entropy coding
Towards Practical Compression of 3D Gaussian Splatting
🤖 AI Summary
This study addresses the substantial storage overhead, compression complexity, and cross-platform decoding inconsistencies inherent in 3D Gaussian Splatting (3DGS) by proposing the COSA-GS framework. This method constructs contextual representations through causal anchor decomposition and employs a lightweight linear architecture that eliminates spatial aggregation. Furthermore, it incorporates quantization-aware training to guarantee bit-exact decoding consistency across platforms. Technically, the framework integrates anchor latent variable modeling, rate-distortion optimization, adaptive pruning, and integer inference to achieve highly efficient compression. Experimental results demonstrate that COSA-GS attains state-of-the-art compression performance while preserving fast and deterministic cross-platform decoding capabilities.
📝 Abstract
3D Gaussian Splatting (3DGS) enables high-quality novel-view synthesis but requires substantial storage. Existing compression methods often rely on spatial context modeling over irregular 3D representations, increasing the complexity of training and coding. Meanwhile, floating-point context inference can introduce numerical inconsistencies across platforms, causing entropy-decoding failures. To address these practical challenges, we propose COSA-GS, which constructs context without spatial aggregation through anchor-wise causal factorization. Specifically, we use geometry context derived from each anchor's coordinates to model a compact learnable anchor latent. The anchor latent is then fused with the geometry context to form an anchor context for attribute coding. The resulting context model features a simple architecture composed solely of linear transformations and activations. We train COSA-GS using rate--distortion optimization with adaptive Gaussian pruning. Further, we develop quantization-aware training and integer inference for the context model to achieve bit-exact consistency of entropy-decoded symbols across platforms. Experiments demonstrate that COSA-GS achieves state-of-the-art compression performance while retaining fast and consistent cross-platform decoding, providing a simple yet effective framework for practical 3DGS compression. Code is available at https://github.com/pengpeng-yu/COSA-GS.
Innovation
Methods, ideas, or system contributions that make the work stand out.
3D Gaussian Splatting Compression
Anchor-wise Causal Factorization
Quantization-aware Training
Integer Inference
Rate-Distortion Optimization
🔎 Similar Papers
No similar papers found.
