Hi. I wrote the paper from University of Chicago. They invented a program which can ‘cloak’ an image with face so it won’t be able to be detected by image reverse search solutions such as Google Images, Facebook and other social networks face search mechanism and so on. The ‘cloaked’ images are undetectable by human eyes but it makes very hard for computers AI to detect a face and compare it with others in order to identify a user by face. I think it will be great if we can test that program and add it to Whonix.
It can be helpful to Whonix users to cloak the face images to save privacy.
Main page: Fawkes
Technical Paper: http://people.cs.uchicago.edu/~ravenben/publications/pdf/fawkes-usenix20.pdf
Source Code ( BSD-3-Clause License ): GitHub - Shawn-Shan/fawkes: Fawkes, privacy preserving tool against facial recognition systems. More info at https://sandlab.cs.uchicago.edu/fawkes
GNU/Linux binary: https://mirror.cs.uchicago.edu/fawkes/files/1.0/fawkes_binary_linux-v1.0.zip
I think your tool is awesome. How long do you guys plan on supporting it in the future in case adversarial detection techniques advance? We want to make sure that the tools we ship are future proof. Please package it for Debian so we can readily build it. Also git commit/tag signing is important for source integrity.
We would like to describe your tool in our documentation for our users. Can you contribute a simplified (for lay people) description of your program, its threat-model and usecases?
Also not important now, but unit tests to check functionality is very useful IMO.
Sorry for my bad English. I am not an inventor of this app, I’ve just read about it. I just test that app for myself and I found out that it could be great tool. For any questions you can write an email to them: firstname.lastname@example.org
Generally, there are three modes of ‘cloaking’ images. Low, mid and high. Low mode provides most undetectable by human eyes changes but a face still can be detected by face detection mechanisms. With high mode the changes of a face may be visible by human eyes too but it will be difficult for face detection to find out who is represented on image.