Raising the Cost of Malicious AI-Powered Image Editinggithub.com/MadryLab 2 pointsharijoe3 years ago1 commentSaveHideCopy link On HNComments−compressedgas3yThis is another example of an adversarial defense that will be useless once the models are trained on input images that have been scrambled like these have.
Comments
This is another example of an adversarial defense that will be useless once the models are trained on input images that have been scrambled like these have.