About ProtoNet's accuracy on miniImageNet
zhangyikaii opened this issue · comments
Thank you for sharing the nice repo.
But ProtoNet's accuracy on the miniImageNet was about 2% lower than the original published in any setup.
My question is: Are there any techniques/tricks/tuning methods for improving it? (But keep the methods(ProtoNet) in the paper)
(Notice that you said in your blog: as the authors provided the full set of hyperparameters, hence .... That's true)
Thank you very much.