Research indexJournal article
Multi-resolution Partial Functional Maps
Xu, Zhang, Zhao
12 Sep 2025
NobleIDark:/48914/w1/20260525/3ED18FB2DOI10.1109/iscipt67144.2025.11265432
Multi-resolution Partial Functional Maps
Xu, Zhang, Zhao · 2025 · Subscription
NobleIDark:/48914/w1/20260525/3ED18FB2DOI10.1109/iscipt67144.2025.11265432
The full text is at the publisher
You can read the abstract here, or open it at the publisher.
Abstract
Non-rigid 3D shape correspondence plays a crucial role in fields such as computer graphics, computer vision, and robotics. Current deep learning-driven partial functional map methods usually employ a fixed number of Laplace-Beltrami operator (LBO) eigenfunctions. This design renders them susceptible to overfitting when handling extensive partial shape missing regions and incapable of preserving essential geometric details at low resolutions. To address these limitations, we introduce a Multi-resolution Partial Functional Maps (MPFM) framework to tackle the challenge of non-rigid partial shape correspondence. We integrate the multi-resolution spectral attention mechanism (LRMF-SA) into the Unsupervised Learning of Robust Spectral Shape Matching (RobustFMNet) framework. Specifically, within a predefined spectral range, a set of functional maps is computed, and a spectral attention network adaptively selects the most informative resolutions, thereby improving robustness for partial matching tasks. We conduct experiments on the PFAUST partial human database, and the results demon- strate that the proposed MPFM method outperforms existing approaches, particularly in challenging scenarios with severe missing regions and partial deformations.