QuARC-GS slashes dynamic scene streaming storage by 11x with quantization-aware 4D coding
New Gaussian splatting framework streams 4D video using 11x less per-frame data
Deep Dive
A new framework called QuARC-GS uses quantization-aware 4D scene optimization for online dynamic scene reconstruction, achieving ultra-high compression while maintaining reconstruction speed and quality. It represents a scene with a single canonical frame and highly compressed per-frame residuals, using quantization-aware anchor deformation and change-gated densification. Experiments show competitive reconstruction quality and training speed while cutting per-frame storage by up to 11x compared to the state-of-the-art.
Key Points
- QuARC-GS compresses 4D Gaussian splatting scenes to a canonical frame + per-frame residuals, cutting storage up to 11x vs SOTA.
- Quantization-aware anchor deformation suppresses insignificant motion updates while preserving key deformations.
- Change-gated densification only adds Gaussians where genuine temporal changes occur, reducing redundant data.
Why It Matters
QuARC-GS makes volumetric video streaming practical for AR/VR and free-viewpoint experiences by slashing bandwidth and storage costs dramatically.