Reproduce “CORGI: Consistency-Aware 3D Dog Reconstruction from a Single Image in the Wild” 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
Yuxiao Wu, Weile Li, Boyi Zhu, Yumeng Liu, Youcheng Cai, Ligang Liu
A framework named CORGI reconstructs high-fidelity, animatable 3D dog models from a single in-the-wild image, eliminating the need for 3D supervision. It achieved the highest PSNR (14.309) on a synthetic dataset and superior FID (13.477) and CLIP-cosine similarity (0.800) on real-world dog images, producing detailed and pose-driven animated outputs.
Key insights
- Disentangling canonical 3D geometry from view-dependent generative inconsistencies is essential for accurate reconstruction from synthesized multi-view inputs.
- Combining a parametric animal prior (D-SMAL) with deformable 3D Gaussians allows for both anatomical consistency and the capture of intricate subject-specific details.
- A self-supervised, progressive refinement loop using a geometrically-conditioned diffusion model can effectively rectify residual 3D distortions and recover high-frequency textures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.