Reproduce “GAS: Generative Avatar Synthesis from a Single Image” 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
Yixing Lu, Junting Dong, Youngjoong Kwon, Qin Zhao, Bo Dai, Fernando De la Torre
GAS uses a generalizable human NeRF and SMPL geometry to guide a video diffusion model in synthesizing novel views and poses from a single image. A task switcher separates static viewpoint changes from animation. On THuman, GAS's novel-view PSNR is 26.77 versus 23.89 for fine-tuned Champ. Dense trajectories can blur frames, and fine details may be lost.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.