So far, in the previous chapters, we have learned about classifying images where we have hundreds/ thousands of example images to train on per class. In this chapter, we will learn about various techniques that will help in classifying an image even when there are very few training examples per class. We will start by training a model to predict a class, even though the images corresponding to the class are not present during training. Next, we will move on to a scenario where only a few images of the class we are trying to predict are present during training. We will code Siamese networks, which fall into the category of few-shot learning, and understand the working details of relation networks and prototypical networks.
We will learn about the following topics in this chapter:
- Implementing zero-shot learning
- Implementing few-shot learning
