Research map / 3D vision / Gaussian splatting
Gaussian Splatting-based Volumetric Video Compression with Sparse 4D Anchors
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes SAGA, a Gaussian-splatting volumetric video codec using hierarchical sparse 4D anchors with INR decoders for inter-anchor interpolation and an orthogonality-updated memory bank for entropy context modeling.
Problem
Compact, render-efficient compression of dynamic 3D Gaussian Splatting scenes for immersive volumetric video, where single canonical scaffolds and isolated anchor conditioning limit handling of non-local dynamics, disocclusion, and inter-anchor redundancy.
Approach
Organize anchors hierarchically directly in spacetime (coarse and fine 4D anchors), decode fine anchors and Gaussian primitives from kernel-interpolated neighbor-anchor features via shared coordinate-based INR decoders, and use a fixed-size memory bank whose slots are updated with orthogonal write components for long-range entropy context.
Key results
- On Neu3D, SAGA achieves PSNR BD-rate reduction of 80.39% over GIFStream and 83.94% on MPEG MIV, per the abstract's claims.
- Against TMIV-24.0 on Neu3D, SAGA achieves 15.27% PSNR BD-rate savings; on MPEG MIV it has a marginal 1.12% PSNR BD-rate increase.
- Raw model size on Neu3D is 15 MB at 32.63 dB PSNR and 89 FPS decoding on an RTX 3090.
Surprising findings
- Memory slot count is non-monotonic in trade-off: 512 slots give -16.19% BD-rate but drop decoding to 36.6 FPS versus 64 default at 89.8 FPS.
Limitations
- Requires per-scene optimization, with notable encoding time for long or complex sequences.
- Fast motion, topology changes, and sparse camera coverage may challenge reconstruction.
- Anchors still account for the largest share of the bitstream after entropy coding.
Experimental setup
Datasets: Neu3D Panoptic Sports MPEG MIV CTC
Benchmarks: Neu3D Panoptic Sports MPEG MIV CTC
Metrics: PSNR SSIM LPIPS BD-rate FPS model size training time
Baselines: 3DGStream 4DGS STG E-D3DGS GIFStream MEGA TMIV-24.0 4DGC ADC-GS
Builds on: ScaffoldGS STG NVRC GIFStream uniform-noise quantization relaxation
Closest papers by meaning
- CodecSplat: Ultra-Compact Latent Coding for Feed-Forward 3D Gaussian Splatting (NeurIPS 2026 · similarity 0.91)
- Temporal Smoothness-Aware Rate-Distortion Optimized 4D Gaussian Splatting (NeurIPS 2025 · similarity 0.90)
- ContextGS : Compact 3D Gaussian Splatting with Anchor Level Context Model (NeurIPS 2024 · similarity 0.90)
- HiCoM: Hierarchical Coherent Motion for Dynamic Streamable Scenes with 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.90)
- 4DGC: Rate-Aware 4D Gaussian Compression for Efficient Streamable Free-Viewpoint Video (CVPR 2025 · similarity 0.90)
- Compressed 3D Gaussian Splatting for Accelerated Novel View Synthesis (CVPR 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.