CyCNN: A Rotation Invariant CNN using Polar Mapping and Cylindrical Convolution Layers

TL;DR

CyCNN introduces a rotation-invariant CNN architecture using polar mapping and cylindrical convolution layers, significantly improving classification accuracy on rotated datasets without data augmentation.

Contribution

The paper presents CyCNN, a novel CNN model that achieves rotation invariance through polar mapping and cylindrical convolution layers, a significant advancement over traditional CNNs.

Findings

CyCNN outperforms conventional CNNs on rotated MNIST, CIFAR-10, and SVHN datasets.

Significant accuracy improvements are observed without data augmentation.

The implementation is publicly available for further research.

Abstract

Deep Convolutional Neural Networks (CNNs) are empirically known to be invariant to moderate translation but not to rotation in image classification. This paper proposes a deep CNN model, called CyCNN, which exploits polar mapping of input images to convert rotation to translation. To deal with the cylindrical property of the polar coordinates, we replace convolution layers in conventional CNNs to cylindrical convolutional (CyConv) layers. A CyConv layer exploits the cylindrically sliding windows (CSW) mechanism that vertically extends the input-image receptive fields of boundary units in a convolutional layer. We evaluate CyCNN and conventional CNN models for classification tasks on rotated MNIST, CIFAR-10, and SVHN datasets. We show that if there is no data augmentation during training, CyCNN significantly improves classification accuracies when compared to conventional CNN models. Our implementation of CyCNN is publicly available on https://github.com/mcrl/CyCNN.