cs.CVNov 10, 2026

Spatial-Spectral Coupling as Attention: Scaling Shape Correspondence to High-Resolution Surfaces

Authors: Bence Demeter, Kilian Ehrenreich, Amelie Kleber, Ioana Oprisiu, Philipp Spohn, Christian Wachinger, Fabian Bongratz, Lennart Bastian

Organizations: Technical University of Munich, Germany · Munich Center for Machine Learning, Germany · Imperial College London, United Kingdom

Abstract

Establishing dense correspondence between high-resolution 3D shapes remains challenging due to the spatial--spectral coupling at the core of many functional-map methods, which grows quadratically with mesh resolution. We observe that this coupling is equivalent to a scaled dot-product attention formulation, with learned descriptors serving as queries and keys and spectral basis functions as values. This connection makes efficient attention implementations directly applicable to shape matching and an effective mechanism for scaling spatial-spectral coupling. We show that optimized quadratic attention implementations can scale modern deep functional map pipelines to large mesh sizes, improving both runtime and memory consumption, while sub-quadratic methods introduce errors, particularly for non-isometric shape pairs, where an exact coupling proves most necessary. These insights motivate Geo-FlashAttention, which restricts the exact softmax to geometry-selected surface patches and achieves near-linear empirical scaling while preserving correspondence accuracy of exact implementations. Across six standard benchmarks and three functional-map pipelines, rigorous analysis shows that efficient couplings preserve correspondence accuracy. At 60k60k vertices, these insights reduce peak training memory from 106.08 to 5.67,GiB, with Geo-FlashAttention being 1.5×1.5\times faster per training step than FlashAttention. This enables end-to-end learning for correspondence problems on large meshes such as cortical surfaces, where high resolution is required to preserve disease-relevant intricacies of the surface geometry. Project page available https://spatial-spectral-attention.github.io/

Figures & tables

Explore similar work

CardsList
  1. DeepShapeMatchingKit: Accelerated Functional Map Solver and Shape Matching Pipelines Revisited

    Apr 11, 2026Yizheng Xie, Lennart Bastian, Congyue Deng +3Functional MapsNon-Rigid Shape Correspondence

  2. Articulating then Matching: Zero-Shot Shape Matching for Uncurated Data

    Jun 28, 2026Qilong Liu, Qinfeng Xiao, Chenyuan Yi +2Non-Rigid Shape Correspondence3D Shape Matching

  3. TokenMatch: 3D Mesh Correspondence Transformer with Curvature-Guided Tokenisation

    Sep 3, 2026Adeela Islam, Zorah Lähner, Vittorio Murino +13D ViTsNon-Rigid Shape Correspondence