Raising the Cost of Malicious AI-Powered Image Editing (opens in new tab)

(github.com)

2 pointsharijoe3y ago1 comments

1 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.

j / k navigate · click thread line to collapse

1 comments

compressedgas3y ago

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.

j / k navigate · click thread line to collapse