Reproduce “Topology-Aware Optimization of Gaussian Primitives for 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, Chengcheng Guo, Yize Wu, Yu Hong, Shengkun Zhu, Zhehao Shen, Yingliang Zhang, Shaohui Jiao and 4 more
TaoGS introduces a method for high-fidelity volumetric video of dynamic human performances, adeptly handling topological changes with an adaptive Gaussian primitive representation. It achieves superior rendering quality, evidenced by a PSNR of 37.264, and enables efficient streaming through a 40x compression ratio.
Key insights
- Disentangling motion and appearance through sparse motion Gaussians and dense appearance Gaussians provides a modular and efficient way to manage both scene dynamics and high-fidelity textures.
- A spatio-temporal tracking strategy, combined with photometric cues, effectively identifies and integrates newly observed regions due to topological changes, preventing redundant Gaussian densification.
- Organizing dynamic Gaussian attributes with a Global Gaussian Lookup Table and a lifespan-aware sorting strategy significantly enhances compatibility with existing video codecs, leading to high compression rates for streaming.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.