Reproduce “Fishbone: From One 3D Asset to a Million Controllable Edits” 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
Yumeng He, Xiaoying Wang, Peihao Li, Yanjia Huang, Joe Masterjohn, Jiajun Wu, Leonidas Guibas, Yin Yang and 2 more
The Fishbone framework introduces an automated, geometry-aware control representation for general 3D meshes, deriving a coupled rib-spine structure to enable controllable parametric deformation, reduced-space dynamics, and animation from a single asset. The method allows for real-time interactive mesh editing with median runtimes of approximately 1 ms per edit and improved robot learning grasp success rates by up to 1.8% on unseen objects through data augmentation.
Key insights
- A biological structure-inspired rib-spine representation can be automatically derived for general 3D meshes, providing a low-dimensional, geometry-aware control space for both local and global deformations.
- An adaptive geodesic heat method and geometry-aware spine point extraction are critical for robustly segmenting and constructing semantically meaningful structural components that adapt to diverse mesh topologies.
- Coupling this derived structural representation with mesh skinning enables efficient, real-time parametric deformation, reduced-space dynamics, and keyframe animation, unifying multiple applications under a consistent control scheme.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.