Paper Overview
- Field: Computer Vision
- Authors: Seungmi Choi, Ji-Hyoun Choi, Suhyun Jo, Tae-Gwon Park, Hui Yong Kim
- Published: 2026-10-14
- arXiv: 2610.20590
- End-to-end rate-distortion training enables MFGS to learn compact multi-view features that jointly represent geometry and appearance, allowing the receiver to predict Gaussian attributes from the decoded features.
- Importance-guided Gaussian selection adapts to different rendering budgets: the receiver can adjust the number of Gaussians used for rendering from the same bitstream.
- 99.76-99.86% BD-rate savings compared to generate-then-compress pipelines
- 52-55% reduction in total GMACs
- Approximately 12x end-to-end speedup
Summary
Feed-forward 3D Gaussian Splatting (3DGS) methods enable efficient reconstruction of 3D scenes from sparse multi-view images in a single forward pass, without scene-specific optimization. However, existing feed-forward scene-coding pipelines first generate a complete Gaussian representation and then encode its geometry and appearance attributes separately, often involving non-differentiable operations.
The resulting "generate-then-compress" pipeline requires complex attribute-specific codec designs and limits rate-distortion performance by restricting the exploitation of cross-attribute redundancy.
The authors propose MFGS (Multi-view Feature coding for Gaussian Splatting), an end-to-end feed-forward framework for scene coding. Instead of compressing an explicit Gaussian representation, MFGS extracts multi-view features from the input images and encodes them into a compact bitstream.
Key Contributions
Results
On RE10K, ACID, and DL3DV benchmarks, MFGS achieves:
*Auto-collected on 2026-10-16.*