DeepSphere: a graph-based spherical CNN
Michaël Defferrard, Martino Milani, Frédérick Gusset, Nathanaël Perraudin
Code Available — Be the first to reproduce this paper.
ReproduceCode
- github.com/deepsphere/deepsphere-tf1Officialtf★ 15
- github.com/deepsphere/deepsphere-pytorchpytorch★ 165
- github.com/deepsphere/deepsphere-cosmo-tf1tf★ 135
- github.com/deepsphere/deepsphere-weatherpytorch★ 74
- github.com/deepsphere/deepsphere-cosmo-tf2tf★ 23
- github.com/smsharma/fermi-gce-flowspytorch★ 6
- github.com/deepsphere/paper-iclr20none★ 5
- github.com/smsharma/neural-global-astrometrypytorch★ 4
Abstract
Designing a convolution for a spherical neural network requires a delicate tradeoff between efficiency and rotation equivariance. DeepSphere, a method based on a graph representation of the sampled sphere, strikes a controllable balance between these two desiderata. This contribution is twofold. First, we study both theoretically and empirically how equivariance is affected by the underlying graph with respect to the number of vertices and neighbors. Second, we evaluate DeepSphere on relevant problems. Experiments show state-of-the-art performance and demonstrates the efficiency and flexibility of this formulation. Perhaps surprisingly, comparison with previous work suggests that anisotropic filters might be an unnecessary price to pay. Our code is available at https://github.com/deepsphere