Rotation Equivariance and Invariance in Convolutional Neural Networks

  • 2018-05-31 03:13:41
  • Benjamin Chidester, Minh N. Do, Jian Ma
  • 24

Abstract

Performance of neural networks can be significantly improved by encodingknown invariance for particular tasks. Many image classification tasks, such asthose related to cellular imaging, exhibit invariance to rotation. We present anovel scheme using the magnitude response of the 2D-discrete-Fourier transform(2D-DFT) to encode rotational invariance in neural networks, along with a new,efficient convolutional scheme for encoding rotational equivariance throughoutconvolutional layers. We implemented this scheme for several imageclassification tasks and demonstrated improved performance, in terms ofclassification accuracy, time required to train the model, and robustness tohyperparameter selection, over a standard CNN and another state-of-the-artmethod.

 

Introduction (beta)

None

 

Conclusion (beta)

None