A guide to convolution arithmetic for deep learning

TL;DR

This paper provides a comprehensive guide to convolution arithmetic, clarifying how input shape, kernel size, padding, strides, and layer types interact in deep learning architectures, with illustrative derivations for better understanding.

Contribution

It offers a detailed, intuitive explanation of convolutional and transposed convolutional layer relationships, aiding practitioners in designing and understanding CNN architectures.

Findings

Derived formulas for convolutional layer output shapes

Clarified relationships between convolutional and transposed convolutional layers

Provided illustrative examples for intuitive understanding

Abstract

We introduce a guide to help deep learning practitioners understand and manipulate convolutional neural network architectures. The guide clarifies the relationship between various properties (input shape, kernel shape, zero padding, strides and output shape) of convolutional, pooling and transposed convolutional layers, as well as the relationship between convolutional and transposed convolutional layers. Relationships are derived for various cases, and are illustrated in order to make them intuitive.