While I share your reservations, I think we should either give it a whirl or wait for someone else to try adversarial processing before we dismiss it out of hand.
My guess is though that if you had a training set of processed vs unprocessed pairs that you could train an “unmister” if you wanted to.
My guess is though that if you had a training set of processed vs unprocessed pairs that you could train an “unmister” if you wanted to.