We have built a few image classification models, but they used the MNIST dataset that was loaded from Keras or from CSV files. The data was always in tabular format. Obviously that is not how images are stored in most situations. This section looks at how to build an image classification model using a collection of image files. The first task is to acquire a set of image files. We are going to load the CIFAR10 data that is included in Keras and save the data as image files. We will then use those files to build a deep learning model. After this exercise, you will know how to create a deep learning image classification task with your own image files.
The deep learning model in this chapter is not a complex model. The focus is to show how the data pipeline for an image classification task is structured. We look at how to arrange the image files, how to use data augmentation and how callbacks can be used during training.