Abstract:We present a PDE-based framework that generalizes Group equivariant Convolutional Neural Networks (G-CNNs). In this framework, a network layer is seen as a set of PDE-solvers where geometrically meaningful PDE-coefficients become the layer's trainable weights. Formulating our PDEs on homogeneous spaces allows these networks to be designed with built-in symmetries such as rotation in addition to the standard translation equivariance of CNNs. Having all the desired symmetries included in the design obviates the need to include them by means of costly techniques such as data augmentation. We will discuss our PDE-based G-CNNs (PDE-G-CNNs) in a general homogeneous space setting while also going into the specifics of our primary case of interest: roto-translation equivariance. We solve the PDE of interest by a combination of linear group convolutions and non-linear morphological group convolutions with analytic kernel approximations that we underpin with formal theorems. Our kernel approximations allow for fast GPU-implementation of the PDE-solvers, we release our implementation with this article in the form of the LieTorch extension to PyTorch, available at <a class="link-external link-https" href="https://gitlab.com/bsmetsjr/lietorch" rel="external noopener nofollow">this https URL</a> . Just like for linear convolution a morphological convolution is specified by a kernel that we train in our PDE-G-CNNs. In PDE-G-CNNs we do not use non-linearities such as max/min-pooling and ReLUs as they are already subsumed by morphological convolutions. We present a set of experiments to demonstrate the strength of the proposed PDE-G-CNNs in increasing the performance of deep learning based imaging applications with far fewer parameters than traditional CNNs.

Lie Group Convolution Neural Networks with Scale-Rotation Equivariance

Enabling equivariance for arbitrary Lie groups

Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data

Rotation-Invariant Feature Learning Via Convolutional Neural Network with Cyclic Polar Coordinates Convolutional Layer

SIM2E: Benchmarking the Group Equivariant Capability of Correspondence Matching Algorithms

Lie Algebra Convolutional Neural Networks with Automatic Symmetry Extraction

B-Spline CNNs on Lie Groups

Equivariance-bridged SO(2)-Invariant Representation Learning using Graph Convolutional Network

Empowering Networks With Scale and Rotation Equivariance Using A Similarity Convolution

Rotation Equivariant Graph Convolutional Network for Spherical Image Classification

Warped Convolutional Networks: Bridge Homography to sl(3) algebra by Group Convolution

Lie Group Decompositions for Equivariant Neural Networks

Polar Coordinate Convolutional Neural Network: from Rotation-Invariance to Translation-Invariance

Group Equivariant Convolutional Networks

Rotation-Invariant Feature Learning in VHR Optical Remote Sensing Images Via Nested Siamese Structure with Double Center Loss.

RC-CNN: Representation-Consistent Convolutional Neural Networks for Achieving Transformation Invariance

Rotation-equivariant Correspondence Matching Based on a Dual-Activation Mixer

Transform-Invariant Convolutional Neural Networks for Image Classification and Search

Roto-translation equivariant convolutional networks: Application to histopathology image analysis

PDE-based Group Equivariant Convolutional Neural Networks

Scale-Equivariant Deep Learning for 3D Data