Reproduce “Compact Motion Embeddings for Feed-forward Unified Human Animation” 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
Seonwook Park, Hyunsoo Cha, Amrita Mazumdar, Shengze Wang, Tianye Li, Hanbyul Joo, Koki Nagano, Shalini De Mello
FUHA transfers a driver’s expression and pose to a reference person while separating motion from appearance. It compresses each driver frame into a small code, trains it with pose-versus-appearance contrast, and renders each frame in one feed-forward pass. On an upper-body test, FUHA-T reaches 0.723 reference identity similarity, versus 0.630 for the strongest evaluated non-FUHA baseline, on that upper-body test.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.