Abstract:In recent years, convolutional neural network has shown good performance in many image processing and computer vision tasks. However, a standard CNN model is not invariant to image rotations. In fact, even slight rotation of an input image will seriously degrade its performance. This shortcoming precludes the use of CNN in some practical scenarios. Thus, in this paper, we focus on designing convolutional layer with good rotation invariance. Specifically, based on a simple rotation-invariant coordinate system, we propose a new convolutional operation, called Rotation-Invariant Coordinate Convolution (RIC-C). Without additional trainable parameters and data augmentation, RIC-C is naturally invariant to arbitrary rotations around the input center. Furthermore, we find the connection between RIC-C and deformable convolution, and propose a simple but efficient approach to implement RIC-C using Pytorch. By replacing all standard convolutional layers in a CNN with the corresponding RIC-C, a RIC-CNN can be derived. Using MNIST dataset, we first evaluate the rotation invariance of RIC-CNN and compare its performance with most of existing rotation-invariant CNN models. It can be observed that RIC-CNN achieves the state-of-the-art classification on the rotated test dataset of MNIST. Then, we deploy RIC-C to VGG, ResNet and DenseNet, and conduct the classification experiments on two real image datasets. Also, a shallow CNN and the corresponding RIC-CNN are trained to extract image patch descriptors, and we compare their performance in patch verification. These experimental results again show that RIC-C can be easily used as drop in replacement for standard convolutions, and greatly enhances the rotation invariance of CNN models designed for different applications.

Rotation equivariant vector field networks

Deep Rotation Equivariant Network

Rotation Equivariance and Invariance in Convolutional Neural Networks

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

Harmonic Networks: Deep Translation and Rotation Equivariance

Empowering Networks With Scale and Rotation Equivariance Using A Similarity Convolution

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

RotEqNet: Rotation-equivariant network for fluid systems with symmetric high-order tensors

Vector Neurons: A General Framework for SO(3)-Equivariant Networks

Rotation-Invariant Feature Learning for Object Detection in VHR Optical Remote Sensing Images by Double-Net

Discrete Rotation Equivariance for Point Cloud Recognition

Designing rotationally invariant neural networks from PDEs and variational methods

RIC-CNN: Rotation-Invariant Coordinate Convolutional Neural Network

Equivariant Multi-View Networks

RRL:Regional Rotation Layer in Convolutional Neural Networks

Achieving Rotation Invariance in Convolution Operations: Shifting from Data-Driven to Mechanism-Assured

Rotation Equivariant Siamese Networks for Tracking

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

Efficient Rotation Invariance in Deep Neural Networks through Artificial Mental Rotation

H-NeXt: The next step towards roto-translation invariant networks

A rotation meanout network with invariance for dermoscopy image classification and retrieval