Rotate four angles. Are these photos in four categories

[SplusCu]

Why is it necessary to rotate each type of photo at different angles in the experiment? Are these photos in four categories？

[fabian57fabian]

I think that, since the omniglot dataset was created by 20 different people and every character has 20 examples, we need more of them. So we do a simple augmentation by rotating 3 more times by 90 degrees. On top of that, using 5-shot 15-query you always use all the examples in each iteration, so it's not perfect.

In my own implementation i describe that: https://github.com/fabian57fabian/prototypical-networks-few-shot-learning