8 ms·
The researchers identify the watermarks to be removed by finding pixel patterns that persist across a large number of images, as shown on this animation: https:
by cs702 9y ago
The researchers identify the watermarks to be removed by finding pixel patterns that persist across a large number of images, as shown on this animation: https://1.bp.blogspot.com/-cJwNoUxIBzM/WZTDpw3ru6I/AAAAAAAAB8s/N-KkDRo5NKkoXEqvXOlN1GUEWbpoyyZUQCLcBGAs/s640/image17.gif https://1.bp.blogspot.com/-cJwNoUxIBzM/WZTDpw3ru6I/AAAAAAAAB... . Their proposed solution is to randomly warp the watermarks.
Unfortunately, their solution could be quickly defeated with image-to-image generative adversarial convnets trained to... remove watermarks from image pairs. (That is, instead of training a model to change, say, image style or resolution, train it to remove artificially added watermarks.)
- gburt 9y agoThese seems likely to me too, but until someone demonstrates it, I'm less certain. GANs are notoriously unstable and it is still quite hard to produce useful models with them.
- HappyTypist 9y agoSounds like we need an adversarial convent for creating watermarks.