Abstract: In this article, we propose a set of transform-based neural network layers as an alternative to the $3\,\, {}\times {}3$ Conv2D layers in convolutional neural networks (CNNs). The proposed ...