Key points
- RigNet predicts skeleton topology and skin weights end-to-end from a 3D mesh with no class-specific priors.
- It generalises across humanoid, stylised and creature meshes from one model.
- It spawned Blender add-ons and is the named baseline for automatic-rigging research.
Summary
Xu, Zhou, Kalogerakis, Landreth and Singh (2020) presented RigNet, the first end-to-end learned method for automatic character rigging that predicts skeleton joint positions, connectivity and skin weights directly from a 3D mesh input without class-specific templates or priors. RigNet uses a graph neural network operating on the mesh surface to jointly predict where joints should be placed and how vertices should be weighted to each joint. The approach generalises across humanoid, stylised and creature meshes, producing rigs usable in standard animation software. RigNet was published at SIGGRAPH 2020, spawned Blender community add-ons, and became the named baseline for all subsequent automatic-rigging research, filling a previously empty pipeline stage in the collection.
Related items
Source
Source: ACM Transactions on Graphics (SIGGRAPH 2020) ↗ (Research)
Cite this item
Xu, Z., Zhou, Y., Kalogerakis, E., Landreth, C. & Singh, K. (2020). ‘RigNet: Neural Rigging for Articulated Characters’, ACM Transactions on Graphics (SIGGRAPH 2020). Available at: https://arxiv.org/abs/2005.00559
Your reference manager can also read this page directly: with the Zotero (or Mendeley) browser connector installed, save it straight to your library. Whole-collection exports: RIS, BibTeX, CSL-JSON.