STL10 `unlabeled` split has labels

vision
1

I am curious to know why the labels in the unlabeled split of STL10 are returned from the dataset.

I can suppress the labels in the training loop, but does it not intuitively make more sense to not send the labels in the unlabeled split of STL10?

>>> unlabeled_ds = STL10(
    root=".",
    split="unlabeled")

>>> unlabeled_ds.classes[unlabeled_ds[0][1]]
"truck"