research
Curvature Turned a Mesh Into Tokens
TokenMatch learned partial and full 3D correspondences with adaptive patches and sub-second feed-forward inference.
Summary
TokenMatch learned partial and full 3D correspondences with adaptive patches and sub-second feed-forward inference.
Trained only on a partial-shape dataset, the transformer generalized to full-shape benchmarks without fine-tuning and reported strong geodesic-error and overlap results across six suites.
Why it matters
TokenMatch learned partial and full 3D correspondences with adaptive patches and sub-second feed-forward inference.
Limits and context
- Trained only on a partial-shape dataset, the transformer generalized to full-shape benchmarks without fine-tuning and reported strong geodesic-error and overlap results across six suites.
Key claims
TokenMatch learned partial and full 3D correspondences with adaptive patches and sub-second feed-forward inference.
Qualification: Trained only on a partial-shape dataset, the transformer generalized to full-shape benchmarks without fine-tuning and reported strong geodesic-error and overlap results across six suites.
Evidence: source-2026-09-06-020
Sources
- arXiv preprint 2609.04202arXiv · primary research
Corrections
No corrections have been recorded for this story.