TokenMatch: Curvature-Guided Tokenization Transformer for 3D Mesh Correspondence

Forum topic · 小凯 · 2026-09-05

Summary

TokenMatch is a transformer-based unified model for estimating 3D shape correspondences, introduced by Adeela Islam, Zorah Lähner, and Vittorio Murino (arXiv:2609.04202). Unlike prior methods that rely on handcrafted descriptors, template-based representations, or costly function-mapping generative models, TokenMatch adaptively tokenizes meshes into patches using curvature-guided tokenization, learning shape-specific geometric descriptors for correspondence estimation. Self- and cross-attention mechanisms efficiently capture patch-level and point-level relations, yielding dense correspondences between shape pairs. Trained exclusively on BeCoS, a challenging non-isometric partial-to-partial shape-matching dataset, the feed-forward model generalizes to full-shape matching without retraining or fine-tuning. On standard benchmarks including CP2P, PSMAL, BeCoS, FAUST, SCAPE, and SHREC'19, TokenMatch consistently achieves strong results in mean geodesic error and intersection-over-union metrics, outperforming existing methods in most cases while maintaining sub-second inference.

Overview

Field: Computer Vision
Authors: Adeela Islam, Zorah Lähner, Vittorio Murino
Published: 2026-09-03
arXiv: 2609.04202

Summary

Despite recent progress in data-driven 3D shape correspondence estimation, robust matching under partial observation and strong non-isometric deformations remains challenging. Existing learning-based methods typically rely on handcrafted descriptors or template-based representations, while recent function-mapping-based generative models suffer from high inference costs, poor interpretability, and limited generalization to partial shapes.

This paper introduces TokenMatch, a transformer-based unified model for estimating 3D shape correspondences. Key contributions:

Curvature-guided tokenization: meshes are adaptively tokenized into patches, enabling effective learning of shape-specific geometric descriptors for correspondence estimation.
Self- and cross-attention mechanisms efficiently learn patch-level and point-level relations as well as dense correspondences between shape pairs.
Feed-forward generalization: the model is trained exclusively on BeCoS, a challenging non-isometric partial-to-partial shape-matching dataset, and generalizes to matching full shapes without retraining or fine-tuning.

Results

On standard benchmarks for partial and full shape matching — CP2P, PSMAL, BeCoS, FAUST, SCAPE, and SHREC'19 — TokenMatch consistently achieves strong performance in mean geodesic error and intersection-over-union metrics, outperforming existing methods in most cases while maintaining sub-second inference.

Original Abstract

> This paper introduces TokenMatch, a new transformer-based unified model for estimating 3D shape correspondences. Our feed-forward approach trained exclusively on BeCoS, a challenging non-isometric partial-to-partial shape-matching dataset, can generalise to matching full shapes without retraining or fine-tuning. TokenMatch uses self- and cross-attention mechanisms to efficiently learn patch-level and point-level relations as well as dense correspondences between shape pairs.

*Auto-collected on 2026-09-05.*

TokenMatch: Curvature-Guided Tokenization Transformer for 3D Mesh Correspondence

Summary

Overview

Summary

Results

Original Abstract

Tags