Re: The adjusted pixels
That was not the point. The point was not to hide the deviation but to show that a small (<0.1%) image change catastophically destroyed the ability of the algorithm to catgorise the image and make it appear to be something else.
Biting the hand that feeds IT
