Seriously, though, synthesis using a recognition model can be a good reality check to remind us of the shortcomings of the model's "understanding" of the domain.
It looks like it's the 22MB file in the model folder.
This is just the resulting network weights. While useful, I believe this alone wouldn't let someone reproduce the experiment as easily as having the dataset.
I don't have the rights to the dataset, so unfortunately I won't be releasing it.
You should be able to scrape 30k+ labeled images relatively easily
How do you imagine going about this?
in the case of porn, there is a zillion sites of images with labeled categories. Web scrape those