Viganò, G., Longari, G., F. Pereira, L., Miolane, N., Melzi, S.
Abstract
We introduce GeomFuM, an open-source Python library for geometry processing and machine learning on functional maps, a compact and versatile representation for shape analysis and correspondence. This library provides object-oriented, modular, and tested implementations for spectral geometry, the study of shapes via the eigendecomposition of geometric operators, as well as functional maps and related algorithms. It includes tools for computing and learning functions and operators on geometric shapes and higher-level tasks such as shape matching, registration, and analysis. GeomFuM provides thoroughly tested object-oriented implementations and supports vectorized batch processing on multiple computational backends, including NumPy and PyTorch. The package integrates functional map theory with practical pipelines to enable research and development in 3D geometry, machine learning, geometric deep learning, and beyond. The source code is freely available under the MIT license at github.com/3diglab/geomfum.
Citation
Viganò, G., Longari, G., F. Pereira, L., Miolane, N., Melzi, S. (GeomFuM: A Python Package for Machine Learning with Functional Maps.
