PAPER 6
Gauge Equivariant Mesh CNNs: Anisotropic convolutions on geometric graphs
Pim de Haan (Qualcomm AI Research)
Maurice Weiler (QUVA Lab/University of Amsterdam)
Taco Cohen (Qualcomm AI Research)
Max Welling (Qualcomm AI Research)
ICLR 2021
Summary
Unlike traditional Convolutional Neural Networks (CNN), Gauge Equivariant CNNs (G-CNNs) can analyse image data on any curved space or geometry. It turns out that they can also be applied to define convolutions on meshes, as they generalize graph convolutional networks (GCNs) to apply anisotropic gauge equivariant kernels. The paper introduces a geometric message passing scheme defined by parallel transporting features over mesh edges.
Citation
@inproceedings{dehaan2021, title={Gauge Equivariant Mesh CNNs: Anisotropic convolutions on geometric graphs}, author={Pim de Haan and Maurice Weiler and Taco Cohen and Max Welling} booktitle={International Conference on Learning Representations}, year={2021}, url={https://openreview.net/forum?id=Jnspzp-oIZE} }
Results
The experiments validate the significantly improved expressivity of the proposed model over conventional GCNs and other methods.
Looking for more papers with code?
* Qualcomm AI Research is an initiative of Qualcomm Technologies, Inc.
