Wow, that is really close to the animation of the anonymizer hood in 'a scanner darkly' (<a href="https://www.imdb.com/title/tt0405296/" rel="nofollow">https://www.imdb.com/title/tt0405296/</a>).
Really cool project! I wonder how long until this sort of software is used in place of blurred-faces.<p>It would be really nice to shoot a public space and instead of worrying about release forms, just anonymize the footage.
I didn't know what GANs were or how they worked, but this video from yesterday explained it really well. So I'll recommend it for anyone else curious!<p>The clear explanation of what a GAN is from the first 8-10 minutes of the video. And there are quite a few interesting examples in there as well.<p><a href="https://www.youtube.com/watch?v=dCKbRCUyop8" rel="nofollow">https://www.youtube.com/watch?v=dCKbRCUyop8</a>
Great to see more research being done on the topic! :) At Brighter AI, we work on this technology since almost 2 years, offering a natural anonymization solution for faces as well as license plates (<a href="https://brighter.ai/video" rel="nofollow">https://brighter.ai/video</a>). Preventing facial recognition when collecting data in public is very important and to be fair, also other companies pursue this now.<p>Best regards from Berlin,
Thomas (thomas@brighter.ai)
I wish the authors wouldn't oversell the privacy claim:<p>> Github: "The DeepPrivacy GAN never sees any privacy sensitive information, ensuring a fully anonymized image."<p>> Abstract: "We ensure total anonymization of all faces in an image by generating images exclusively on privacy-safe information."<p>> Paper: "We propose a novel generator architecture to anonymize faces, which ensures 100% removal of privacy-sensitive information in the original face."<p>Changing a face anonymizes an image the same way that removing a name anonymizes a dataset -- poorly. This is cool, but it's not anonymization.
link to the arxiv paper: <a href="https://arxiv.org/abs/1909.04538" rel="nofollow">https://arxiv.org/abs/1909.04538</a>