understanding the dataloader procedure about deepemd HOT 1 CLOSED

hello
1.query=15 means the number of test images each class in this mini-task (an episode). it can be changed. since the evaluation is by repeatedly sampling tasks, larger query number and sampling times can make the average result more stable and convincing. for example, if you have large gpu memory, you can set a larger query size for training and eval.
2. 80 is the number of query images plus the number of support images, i.e. 5 way * (1 support image + 15 query image). 640 is the dim of the feature embedding.
3.75 is the predicted scores of 75 query images. since each class has 15 query images, and you have 5 class, you have 75 testing images totally.

from deepemd.