How Many classes in training loop are required?

[charurathour]

Problem
I am trying few shot learning but i don't have a lot of classes in similar domain i have only 2 classes, how many classes are required in training dataset to do few shot learning?

Considered solutions
I tried training with only 2 class dataset and then using it as feature extractor for prototypical network, which does not give good result, will it be still considered few shot or normal transfer learning, for what i have read in paper it will be considered simple shot. For doing meta learning how many classes are required?