// the take
A PyTorch implementation of the ICLR 2020 paper 'Deep Graph Matching Consensus' (DGMC), a neural method for matching nodes between two graphs via iterative neighborhood consensus. It's for researchers working on graph matching, correspondence problems, or geometric deep learning who want to reproduce or build on this specific paper.
Comes from the PyTorch Geometric author, so the module design and PyG integration are clean and idiomatic rather than a one-off research hack. It ships an actual test suite and readthedocs documentation, which is rare for paper-release repos. It includes working example scripts for four different benchmark datasets (Pascal VOC, WILLOW, PascalPF, DBP15K), so you can reproduce the paper's numbers rather than just reading about them.
Dead since September 2021 — no commits, no issue response, nothing. It pins to PyTorch Geometric >=1.5.0 and KeOps >=1.1.0, both of which have had breaking API changes since; getting this to actually run on a current stack means real dependency archaeology, not just a pip install. The Travis CI badge is decorative since Travis stopped free OSS builds around 2020-2021, so despite appearances nothing is actually being tested on new environments. It implements exactly one paper's method with no pretrained weights, so there's no shortcut to using this beyond training from scratch on the included research datasets.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →