Reproduce “ELITE: Efficient Gaussian Head Avatar from a Monocular Video via Learned Initialization and TEst-time Generative Adaptation” 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
Kim Youwang, Lee Hyoseok, Subin Park, Gerard Pons-Moll, Tae-Hyun Oh
ELITE, developed by researchers from POSTECH, KAIST, UNIST, University of Tübingen, and Max Planck Institute for Informatics, introduces an efficient framework for generating high-fidelity, animatable human head avatars from a single monocular video. The system integrates a learned 3D Gaussian Splatting initialization with a novel rendering-guided single-step diffusion enhancer for test-time generative adaptation, achieving superior visual quality and identity preservation while reducing avatar synthesis time to 20 minutes per identity.
Key insights
- A synergistic coupling of 3D data priors (for geometric and initial appearance stability) and 2D generative priors (for diverse visual augmentation) can overcome the individual limitations of each approach.
- Using degraded 3D avatar renderings as strong conditioning for a generative model enables much faster image generation (60x speed-up) and prevents identity hallucination compared to generating from pure noise.
- Efficient, learned 3D Gaussian initialization, combined with targeted real and synthetic image supervision, facilitates robust generalization for animatable avatars from limited monocular video inputs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.