Rotation-Invariant Autoencoders for Signals on Spheres

Abstract

Omnidirectional images and spherical representations of $3D$ shapes cannot be processed with conventional 2D convolutional neural networks (CNNs) as the unwrapping leads to large distortion. Using fast implementations of spherical and $SO(3)$ convolutions, researchers have recently developed deep learning methods better suited for classifying spherical images. These newly proposed convolutional layers naturally extend the notion of convolution to functions on the unit sphere $S^2$ and the group of rotations $SO(3)$ and these layers are equivariant to 3D rotations. In this paper, we consider the problem of unsupervised learning of rotation-invariant representations for spherical images. In particular, we carefully design an autoencoder architecture consisting of $S^2$ and $SO(3)$ convolutional layers. As 3D rotations are often a nuisance factor, the latent space is constrained to be exactly invariant to these input transformations. As the rotation information is discarded in the latent space, we craft a novel rotation-invariant loss function for training the network. Extensive experiments on multiple datasets demonstrate the usefulness of the learned representations on clustering, retrieval and classification applications.

References in corpus (16)

• Adam: A Method for Stochastic Optimization
• Shortcut Learning in Deep Neural Networks
• Deep Convolutional Inverse Graphics Network
• Invariant Risk Minimization
• On the Generalization of Equivariance and Convolution in Neural Networks to the Action of Compact Groups
• Gauge Equivariant Convolutional Networks and the Icosahedral CNN
• Scale-Invariant Convolutional Neural Networks
• Covariant Compositional Networks For Learning Graphs
• On Learning Sets of Symmetric Elements
• Gauge Equivariant Mesh CNNs: Anisotropic convolutions on geometric graphs
• Clebsch-Gordan Nets: a Fully Fourier Space Spherical Convolutional Neural Network
• Harmonic Networks: Deep Translation and Rotation Equivariance
• Theoretical Aspects of Group Equivariant Neural Networks
• FoldingNet: Point Cloud Auto-encoder via Deep Grid Deformation
• Temporal Transformer Networks: Joint Learning of Invariant and Discriminative Time Warping
• The general theory of permutation equivarant neural networks and higher order graph variational encoders

Cited by in corpus (1)

• You Only Hypothesize Once: Point Cloud Registration with Rotation-equivariant Descriptors