Introduction to the Basics of CNNs

Chapter review

Image classification is one of the main tasks in the computer vision field. Most image classification architectures consist of convolutional layers besides fully connected ones. Therefore, it is essential to understand the logic behind convolution, how to apply it, and what to expect after this operation is applied. In addition to this operation, there are still some more terms and methodologies to use inside of convolutional layers, and all of these supplementary subjects will be covered during this chapter along with example Python codings to have better intuition.

We will learn the following in this chapter:

• 2D convolution operations
• Applying padding and stride to 2D convolutions
• Convolutional neural networks
• Comparison between fully and convolutional neural networks
• Pooling types (max pooling, average pooling)
• The difference between 2D and 3D convolution
• Python coding to apply convolution and pooling on 2D matrices
• Using PyTorch built-in functions to create CNN layer
• Creating the first batch of images from scratch
• Using the Matplotlib library to visualize the feature maps coming from CNN layers