DeepSphere: a graph-based spherical CNN
This work addresses the problem of designing efficient and rotation-equivariant spherical CNNs for researchers and practitioners working with spherical data.
This paper introduces DeepSphere, a graph-based spherical Convolutional Neural Network (CNN) that balances efficiency and rotation equivariance. It achieves state-of-the-art performance on relevant problems, suggesting that anisotropic filters may not be necessary.
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