Train-val splits for imagenet and places205
drimpossible opened this issue · 1 comments
drimpossible commented
Hi!
I wanted to reproduce the Imagenet and Places205 experiments in this paper. In the code, it seems like the training set of images was divided into 3 parts for k-fold cross validation -- are those numbers reported, or are the reported numbers on the validation split of Imagenet?
Sorry if you had specified this explicitly somewhere and I missed it.
drimpossible commented
Ah, that is switched off by default. Sorry, my bad.