Use of labels of published datasets

#1

Hi,

is it allowed to use the labels (i.e. ground truth factors) of the published datasets (mpi3d_toy and mpi3d_realistic) in any way for pretraining?

In [Announcement] Regarding Transfer Learning!, you stated that it is allowed to use any data to produce pretrained weights. Following this wording, using the labels is allowed.

Can you confirm this please?