Reproduce “Robust Dual Gaussian Splatting for Immersive Human-centric Volumetric 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
Yuheng Jiang, Zhehao Shen, Yu Hong, Chengcheng Guo, Yize Wu, Yingliang Zhang, Jingyi Yu, Lan Xu
Robust Dual Gaussian Splatting (DualGS) is introduced for representing and rendering human-centric volumetric videos. DualGS achieves high-fidelity real-time playback while reducing data storage to approximately 350KB per frame, enabling practical deployment on memory-constrained devices.
Key insights
- Disentangling motion and appearance into separate Gaussian sets (joint and skin) allows for robust tracking of complex human movements while preserving fine visual details.
- A coarse-to-fine optimization strategy, combining motion prediction and temporal regularization, ensures both robust tracking and temporal consistency, preventing artifacts.
- Exploiting the disentangled representation enables highly efficient, specialized compression techniques for different attribute types (motion, appearance, SHs), leading to extreme data reduction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.