Reproduce “AnimalLift: Reconstructing Animatable 3D Animals from a Single Image by Learning Canonical Shape, Texture, and Fur Maps” 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
Chunyi Sun, Ruyi Zha, Weijian Deng, Junlin Han, Dylan Campbell, Stephen Gould
AnimalLift reconstructs animatable 3D animal assets with explicit fur from a single image by learning canonical shape, texture, and fur maps within a unified framework. The method generates production-ready models that are directly compatible with standard digital content creation pipelines, reducing required cleanup effort for animation by over 90% compared to baselines.
Key insights
- Both surface appearance and complex fur can be effectively represented within a shared, UV-aligned canonical space, facilitating disentanglement from pose and ensuring animation compatibility.
- Procedural data generation is crucial for training models to reconstruct highly structured, animation-ready 3D assets when real-world aligned supervision is unavailable.
- A joint, feed-forward architecture allows for efficient, single-image prediction of canonical geometry, detailed textures (via flow matching), and explicit fur, streamlining the asset creation pipeline.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.