Teaching Neural Networks Using Comic Strips

There is a rich variety of information in comic strips: dialogues, images, face expressions and body postures. In this paper we describe our experience in using this variety to help students learning how to use neural networks. In particular, the scope is to generate dialogues between chosen characters, and this is achieved by a sequence of steps needed to build the proper dataset: face detection to recognize the characters in the comic, text recognition, and assigning texts to the characters. Finally, we use a Transformer network to generate a dialogue.