TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPermanent story

research

Curvature Turned a Mesh Into Tokens

TokenMatch learned partial and full 3D correspondences with adaptive patches and sub-second feed-forward inference.

Published Updated Story ID: mp-2026-09-06-018
Read the complete editionStory JSON

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

  1. 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

  1. arXiv preprint 2609.04202arXiv · primary research

Corrections

No corrections have been recorded for this story.