search

liuquande / SRC-MT

[TMI'20] Semi-supervised Medical Image Classification with Relation-driven Self-ensembling Model
133 stars 20 forks source link

Chestxray14 labeled subset setup #3

Closed lfb-1 closed 4 years ago

lfb-1 commented 4 years ago

Hi, I am curious about the multi-label chestxray14 experiment setup. Because this dataset contains serious data imbalance problems, I wonder is your 20% labeled dataset a balanced subset or you just random sampling them?

Looking forward to your reply.

Cheers

liuquande commented 4 years ago

We random sample them and keep the the ratio of each class in the validation set the same as the original training set.

lfb-1 commented 4 years ago

Thanks for the clarification. I am closing the issue.