Reproduce “EDGAR: Efficient Diffusion-Driven Gaussian Avatar Refinement via Synthetic Failure” 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
Kwan Yun, Minhyeok Bang, Colette Suhjung Koo, Junyong Noh
EDGAR uses subject-reference-guided diffusion to refine a Gaussian head avatar, then transfers corrected images into the shared 3D avatar. This targets recoverable appearance artifacts left by limited capture. In RGBAvatar self-reenactment with four input videos, it lowers L1 error to 0.0364 and LPIPS to 0.2121; results vary by pipeline and metric, and structural failures may remain.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.