Reproduce “D-FCGS: Feedforward Compression of Dynamic Gaussian Splatting for Free-Viewpoint Videos” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenkang Zhang, Yan Zhao, Qiang Wang, Zhixin Xu, Li Song, Zhengxue Cheng
A feedforward compression framework, D-FCGS, enables efficient, zero-shot encoding of dynamic Gaussian Splatting for Free-Viewpoint Videos. This approach achieves over 17x P-frame compression while preserving visual fidelity across diverse scenes without per-scene optimization.
Key insights
- Decoupling the compression process from reconstruction allows for a generalizable, pre-trained codec that operates in a zero-shot manner on diverse dynamic scenes.
- Representing inter-frame motion sparsely via control points significantly reduces data volume for compression while a dual prior-aware entropy model efficiently encodes these sparse motions.
- Integrating a video-like I-P coding structure with online color refinement helps to maintain high visual fidelity and mitigate artifacts despite aggressive compression ratios.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.