Deep Rotation Equivariant Network

TL;DR

The paper introduces Deep Rotation Equivariant Network (DREN), which efficiently learns rotation-equivariant features by applying transformations on filters, significantly reducing computational overhead while improving performance on rotated image datasets.

Contribution

It proposes a novel architecture with cycle, isotonic, and decycle layers that apply rotation on filters instead of feature maps, enhancing efficiency and accuracy.

Findings

Over 2x speedup compared to previous methods

Reduced memory overhead in rotation-equivariant learning

Improved accuracy on Rotated MNIST and CIFAR-10 datasets

Abstract

Recently, learning equivariant representations has attracted considerable research attention. Dieleman et al. introduce four operations which can be inserted into convolutional neural network to learn deep representations equivariant to rotation. However, feature maps should be copied and rotated four times in each layer in their approach, which causes much running time and memory overhead. In order to address this problem, we propose Deep Rotation Equivariant Network consisting of cycle layers, isotonic layers and decycle layers. Our proposed layers apply rotation transformation on filters rather than feature maps, achieving a speed up of more than 2 times with even less memory overhead. We evaluate DRENs on Rotated MNIST and CIFAR-10 datasets and demonstrate that it can improve the performance of state-of-the-art architectures.