Reproduce “Real-Time Human Reconstruction and Animation using Feed-Forward Gaussian Splatting” 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

Devdoot Chatterjee, Zakaria Laskar, C.V. Jawahar

HumanGS builds a reusable 3D Gaussian avatar from multi-view images and known SMPL-X vertices. It samples features at projected body vertices to predict a canonical asset, then uses skinning and Gaussian rendering to animate it without a network pass. On AvatarReX animation, it reports comparable image quality and 0.048 vs. 0.287 seconds per-frame synthesis against HumanRAM; tests do not establish arbitrary motion robustness.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.