multi-class image classification problem with very low data for train or test

Question

the project I am talking about is a image detection machine learning project , I have 200k photos and each photo represent a one class, so we have 200k classes (for example