Reproduce “ZARA: Zero-shot Animatable and Relightable 3D Avatar reconstruction 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

Yixiao Chen, Dongdong Weng, Chengfeng Zhao, Tao Gong, Zhihe Zhao, Zhe Zhu, Yue Liu, Yuan Liu

ZARA adds predicted albedo, roughness, and normals to a pretrained single-image human-avatar model, enabling animatable 3D Gaussians to be rendered under new lighting. Its automatic material-prior pipeline feeds a simplified physically based renderer. On held-out Human4DiT-3D, it scores 20.667 PSNR and 0.0427 LPIPS, versus 14.949 and 0.1353 for LHM plus UniRelight. These image-space results depend partly on estimated material priors.

Summary by alphaXiv.

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