3 ms·

this technology could also be used to reveal AI-generated synthetic imagery Yeah, AI may one day be able to fool the human visual system but it will never be a

• antirez 6y ago
  LOL, I don't understand what they were thinking when writing this part. AI may be used, probably, in order to remove the watermark even right now, not one day, but recognizing the non uniform part, subtracting it and adding a different one. To create the training set of same-image taken with differnt-sensor is not hard.
• elwell 6y ago
  It's a game of cat and mouse of course.
  • hyko 6y ago
    Where the mouse is now the size of a horse. Ultimately this is a classic case of security through obscurity, it won’t protect us from deep fakes long term.
• aimor 6y ago
  I've seen a lot of attempts at neural network image generation that create constant artifacts in the images. It's not that image generators are unable to add a specific bias to their images in an attempt to look like a specific camera. It's the opposite: generators have their own signatures that these techniques can detect and track.
  • hyko 6y ago
    That part doesn’t really need ML of any kind; ultimately unless a watermark is based on a rigorous and proven cryptographic scheme, it can be faked. They are apparently not doing anything like that, it’s just “hey, these photo sites are all a bit different”. Well, that’s what they’re saying publicly anyway, could be complete bs and every sensor has a government sanctioned backdoor in it.