DeepSphere: a graph-based spherical CNN

Michaël Defferrard, Martino Milani, Frédérick Gusset, Nathanaël Perraudin
12/30/2020
cs.LGcs.CVstat.ML

Official code: deepsphere/deepsphere-tf1 (14 stars, MIT, TensorFlow, last commit 2021-01-01).

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

View on arXivView PDF

Code Implementations(1)

deepsphere/deepsphere-tf1Official100%
TensorFlowMIT

DeepSphere: a graph-based spherical CNN (TensorFlow)

142Shell, PythonNov 7, 20195 years agoMIT
climate-sciencecosmological-inferencedeepspheregraph-neural-networkssphere+3 more

Similar Papers

DeepSphere: towards an equivariant graph-based spherical CNN

92% match1 repo142019

Learning SO(3) Equivariant Representations with Spherical CNNs

86% match1 repo3072017

DeepSphere: Efficient spherical Convolutional Neural Network with HEALPix sampling for cosmological applications

86% match1 repo1372018

Scaling Spherical CNNs

84% match1 repo1452023

Spherical Feature Pyramid Networks For Semantic Segmentation

84% match2023

Reducing SO(3) Convolutions to SO(2) for Efficient Equivariant GNNs

84% match2023

Cite this paper

@article{defferrard2020deepsphere,
  title  = {DeepSphere: a graph-based spherical CNN},
  author = {Michaël Defferrard and Martino Milani and Frédérick Gusset and Nathanaël Perraudin},
  year   = {2020},
  eprint = {2012.15000},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/2012.15000}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2012.15000.svg)](https://codeofpaper.com/paper/2012.15000)

Discussion