The first operation of the model is reading the images and standardizing them. In fact, we cannot work with images of variable sizes; therefore, in this first step, we'll load the images and reshape them to a predefined size (32x32). Moreover, we will one-hot encode the labels in order to have a 43-dimensional array where only one element is enabled (it contains a 1), and we will convert the color space of the images from RGB to grayscale. By looking at the images, it seems obvious that the information we need is not contained in the color of the signal but in its shape and design.
Let's now open a Jupyter Notebook and place some code to do that. First of all, let's create some final variables containing the number of ...
