Doubt regarding Mini-Imagenet dataset

[Kadam-Tushar]

How train and test splits are made for mini-Imagenet? I guess there are 100 classes with 600 images of each class. For few-shot learning in literature people have divided dataset into 64 classes for training and rest for validation and testing. I didn't understood how will we translate this in partial label learning setup? Like how will we split dataset into train and test?