EvoGS: Constructing Continuous-Layered Gaussian Splatting with Evolution Tree for Scalable 3D Streaming

2026-06-05

AI Summary

Existing 3D Gaussian splatting streaming methods rely on discrete hierarchical structures, which often lead to error accumulation across layers, severe Gaussian redundancy, and discontinuous visual quality transitions. This work proposes EvoGS, the first approach to establish a continuous hierarchical representation based on an evolutionary tree. Leveraging a wavelet-inspired parent-to-child refinement mechanism, EvoGS progressively generates fine details while correcting ancestral errors at each level. The method substantially enhances representation sparsity and compression efficiency: Gaussian redundancy is reduced from over 65% to below 25%, while transmission bandwidth and GPU memory consumption drop to 1/2.4 and 1/5.5 of baseline levels, respectively. These improvements enable smooth, real-time adaptive streaming and rendering.

Abstract

Streaming 3D Gaussian Splatting requires highly scalable, progressive representations. Existing progressive methods rely on \textit{discrete layering}, accumulating separate splat sets for each level of detail. This structural independence between layers inherently leads to error accumulation, severe splat redundancy, and uncontrolled quality transitions. We propose EvoGS, the first \textit{continuous-layering} representation. Organized as an Evolution Tree, EvoGS generates finer details via an explicit, wavelet-inspired parent-child refinement. This empowers child nodes to structurally correct ancestral errors, yield inherently sparse and highly compressible inter-layer signals. Extensive experiments show EvoGS eliminates splat redundancy from over 65\% to under 25\%. Compared to state-of-the-art baselines, it reduces transmission payload and GPU VRAM footprint by up to 2.4$\times$ and 5.5$\times$, respectively, and achieves smooth quality transitions optimal for real-time adaptive streaming. Project page: https://yuang-ian.github.io/evogs/

Problem

• Gaussian Splatting
• progressive representation
• discrete layering
• splat redundancy
• quality transition

Innovation

• continuous-layering
• Evolution Tree
• Gaussian Splatting
• progressive representation
• wavelet-inspired refinement