Rename `self.train` -> `self.use_triplet_output` in `ImageFolderTriplet`

[ashlaban]

Renaming this member from train to use_triplet_output more clearly indicates that this is a switchable property (if one e.g. would want to eval both the triplet loss and the mAP "accuracy" metric on the same dataset).

DeepDIVA/datasets/image_folder_triplet.py

Line 103 in 4dccf3b

train : bool