Training neurons for image recognition?

Question

Python

Task: an image recognition program. There is a dataset of 100 different classes, each class has 1 image.
Question: What is the best thing to do, train a neuron? Or is there a better and easier option? Yes, I can augment each image 20 times in each class. But that doesn't make sense since the images for validation are exactly the same as those for training. For the test, the same thing, the main thing is that it distinguishes between classes.

Answer 1

1 image per class

Then you can disperse, there will be no kin. Thousands of examples are needed.