RigNet: Neural Rigging for Articulated Characters

How to read this

Category

Method: neural auto-rigging from meshes

Contributions

• An end-to-end network that predicts a skeleton (joint placement and topology) from an input 3D character mesh
• Estimates surface skin weights based on the predicted skeleton
• Operates directly on the mesh with no assumptions about shape class or structure

Context

Advances learned rigging in the line of Liu et al.'s NeuroSkinning and the classic Baran-Popovic automatic rigging, predicting both skeleton and weights rather than weights alone.Builds on: NeuroSkinning: Automatic Skin Binding for Production Characters with Deep Graph Networks · Automatic Rigging and Animation of 3D Characters

Correctness

Validated three ways (quantitative comparison to animator rigs, qualitative posing/animation at multiple levels of detail, and ablations of algorithm choices); as a learned rigger trained on a rigged-model collection, output quality and topology fidelity remain bounded by that training data.

Clarity

Accessible at the pipeline level; the mesh-based architecture rewards a second pass.

How to read it

First pass for the two-stage skeleton-then-weights pipeline and the evaluation against animator rigs; second pass on the mesh architecture and the ablation study if you care about why design choices matter.