Rotated Ring, Radial and Depth Wise Separable Radial Convolutions

TL;DR

This paper introduces rotationally invariant convolution methods that improve neural network robustness to image rotations while reducing computational complexity through depthwise separable radial convolutions.

Contribution

It proposes trainable rotation-invariant convolutions and a depthwise separable radial convolution approach, enhancing rotation robustness with lower computational costs.

Findings

Achieves rotational invariance across different models and datasets

Depthwise separable radial convolutions reduce computational load

Rotationally invariant features impact accuracy positively

Abstract

Simple image rotations significantly reduce the accuracy of deep neural networks. Moreover, training with all possible rotations increases the data set, which also increases the training duration. In this work, we address trainable rotation invariant convolutions as well as the construction of nets, since fully connected layers can only be rotation invariant with a one-dimensional input. On the one hand, we show that our approach is rotationally invariant for different models and on different public data sets. We also discuss the influence of purely rotational invariant features on accuracy. The rotationally adaptive convolution models presented in this work are more computationally intensive than normal convolution models. Therefore, we also present a depth wise separable approach with radial convolution. Link to CUDA code https://atreus.informatik.uni-tuebingen.de/seafile/d/8e2ab8c3fdd444e1a135/