GANomaly unsupervised or semi-supervised?

[retrok94]

Hello,
I studied the various papers and implementations of the Anomalib algorithms. One thing I found odd is that GANomaly is called semi-supervised in its paper while all others are unsupervised (and in the Anomalib paper its algorithms are also called unsupervised).

I found this comment from @samet-akcay here on the original GANomaly repository, which states that the test set is checked after every epoch.

My question is what exactly made the GANomaly algorithm semi-supervised originaly (was the test set check relevant for the model predictions?) and if that is still the case within Anomalib or if it is fully unsupervised now (except when using validation, which I am not).
I could not find anything in the code that would suggest the training is not unsupervised for this algorithm, so I would asume it is.