Reproduce “Synthetic Prior for Few-Shot Drivable Head Avatar Inversion” 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

Wojciech Zielonka, Stephan J. Garbin, Alexandros Lattas, George Kopanas, Paulo Gotardo, Thabo Beeler, Justus Thies, Timo Bolkart

SynShot creates drivable 3D head avatars by learning a reusable prior from synthetic data and adapting it to a person from a few real images. Its identity and expression branches generate Gaussian primitives with part-specific sampling, then tune to the subject. In qualitative comparisons, three-view SynShot handles novel views and expressions better than monocular baselines trained on about 3,000 frames, but synthetic appearance gaps constrain adaptation.

Summary by alphaXiv.

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