Rigging data is more useful when it captures not just a character’s shape, but also how its joints are organized and how motion deforms its surface. These paired signals let researchers study skeleton prediction and skinning together, including the extra bones needed to articulate garments.
What Makes It Useful
- Each sample links mesh geometry, joint positions and hierarchy, and sparse skinning weights, supporting tasks from joint regression to full rig generation.
- The hosted archives list 39,073 characters across four subsets, with different skeleton conventions and mesh densities. This variety can expose models to more than one rig style, but requires care when mixing coordinate systems and normalizing scale.
- A body-joint mask distinguishes main-body joints from additional bones, such as skirt or clothing bones. Researchers can isolate conventional body rigging or explicitly study garment-aware rigs.
- The data is packaged as NumPy files inside ZIP archives, so the Hugging Face dataset viewer cannot preview it automatically; downloading and parsing the archives is necessary.
Good Fit—and Trade-offs
A good fit for researchers training or evaluating humanoid auto-rigging, joint-position prediction, or skinning-weight models, especially when garment articulation matters. Look elsewhere if you need a directly browsable dataset, consistent units across all sources, or commercial/production rights: source coordinate conventions vary, and the license limits use to academic research. Also check dataset versions carefully: the hosted archive counts differ from the paper’s reported 38,088-character Human-Mix curation, so results may not be directly comparable.
