Scattering Networks on the Sphere for Scalable and Rotationally Equivariant Spherical CNNs
This work provides a solution for researchers and practitioners working with high-resolution spherical data, allowing them to apply spherical CNNs to previously unmanageable datasets.
This paper addresses the computational demands of spherical CNNs, which typically cannot scale beyond thousands of pixels. The authors develop spherical scattering networks that are computationally scalable and rotationally equivariant, enabling spherical CNNs to process high-resolution data of tens of megapixels and beyond.
Convolutional neural networks (CNNs) constructed natively on the sphere have been developed recently and shown to be highly effective for the analysis of spherical data. While an efficient framework has been formulated, spherical CNNs are nevertheless highly computationally demanding; typically they cannot scale beyond spherical signals of thousands of pixels. We develop scattering networks constructed natively on the sphere that provide a powerful representational space for spherical data. Spherical scattering networks are computationally scalable and exhibit rotational equivariance, while their representational space is invariant to isometries and provides efficient and stable signal representations. By integrating scattering networks as an additional type of layer in the generalized spherical CNN framework, we show how they can be leveraged to scale spherical CNNs to the high-resolution data typical of many practical applications, with spherical signals of many tens of megapixels and beyond.