How convolutional nets work?

I was preparing some material for an introductory class on Deep Learning using Tensorflow. For that, I created a small tutorial for image classification of faces with and without masks. Since it may be helpful for other professionals, I share it here. Please find the link for the Colab Notebook below:

This material shows you how you can view a deep network for image classification as a data compression method, as illustrated below. However, it is indeed a special type of compression: it is a semantical-based compression. How is it done? Check out the link above :-).

Furthermore, we go deep into the learned model to understand what is happening inside the Convolutional Neural Network when a new query image is inputted. CNN's should not be black boxes!

In this particular example, it is possible to see that the network could localize the mouths at the images where no facemasks were being used.

Finally, you can try the trained model in any particular set of images you want and see what it outputs for images of people with facemask....

... and without facemasks...

I hope you liked this educational article!