Model card: document only the sources actually used

Model card: neutral eval-set wording

Document v2.0.0: training data, canonicalisation, eval results

v2.0.0: surface-sampled synthetic + mined CC0 rigs — 31.5% → 94.7% on rig-truth eval

Restore original segmentation model (revert in-progress rework that didn't improve quality)

v1.2.0: flat-PointNet architecture (runs correctly in linked ONNX Runtime), trained on synthetic + CC0 real rigs

Card: add Versions changelog (v1.1.0)

Improve model: T/A-pose data + upright aug + PointNet++ local features

Add model card

Add meshseg.onnx

initial commit