Replying to
@lina I think I looked into a few datasets once (COCO?) and there was stuff about it in theory being pulled from appropriately CC0-ish stuff on flickr, but I don't think there was the kind of machine-readable accompanying database of every image, a link to the resource, it's licencing terms, etc etc that you'd want in order to mechanically track, audit, and verify this. I'm not sure there were even flickr reference numbers anywhere.