This guide, authored by Vincent Dumoulin and Francesco Visin, aims to explain the relationship between convolutional layers and transposed convolutional layers in deep learning models. It provides an intuitive understanding of how input shape, kernel shape, zero padding, strides, and output shape are related in convolutional, pooling, and transposed convolutional layers. The guide is applicable to various machine learning frameworks and focuses on 2-D discrete convolutions with square inputs, kernels, and strides, as well as zero padding. It covers: 1. **Introduction**: Explains the importance of convolutional neural networks (CNNs) and the complexity of understanding their output shapes. 2. **Convolution Arithmetic**: Analyzes the relationship between input and output shapes in convolutional layers, including no zero padding, unit strides, zero padding, and non-unit strides. 3. **Poolings**: Discusses pooling operations, which reduce the size of feature maps using functions like average or max pooling. 4. **Transposed Convolution Arithmetic**: Introduces transposed convolutions, which are useful for decoding layers in autoencoders or projecting feature maps to higher dimensions. It explains how transposed convolutions are related to direct convolutions and provides formulas for their output shapes. 5. **Miscellaneous Convolutions**: Explains dilated convolutions, which increase the receptive field without increasing the kernel size. The guide is supported by figures and animations available on GitHub, and it welcomes feedback from readers to improve its accuracy and clarity.This guide, authored by Vincent Dumoulin and Francesco Visin, aims to explain the relationship between convolutional layers and transposed convolutional layers in deep learning models. It provides an intuitive understanding of how input shape, kernel shape, zero padding, strides, and output shape are related in convolutional, pooling, and transposed convolutional layers. The guide is applicable to various machine learning frameworks and focuses on 2-D discrete convolutions with square inputs, kernels, and strides, as well as zero padding. It covers: 1. **Introduction**: Explains the importance of convolutional neural networks (CNNs) and the complexity of understanding their output shapes. 2. **Convolution Arithmetic**: Analyzes the relationship between input and output shapes in convolutional layers, including no zero padding, unit strides, zero padding, and non-unit strides. 3. **Poolings**: Discusses pooling operations, which reduce the size of feature maps using functions like average or max pooling. 4. **Transposed Convolution Arithmetic**: Introduces transposed convolutions, which are useful for decoding layers in autoencoders or projecting feature maps to higher dimensions. It explains how transposed convolutions are related to direct convolutions and provides formulas for their output shapes. 5. **Miscellaneous Convolutions**: Explains dilated convolutions, which increase the receptive field without increasing the kernel size. The guide is supported by figures and animations available on GitHub, and it welcomes feedback from readers to improve its accuracy and clarity.