Low-level Fawkes dose not work. Inquiry about Adversarial Traning Versions and Testing Scripts

[csjunjun]

Dear Dr. Shawn, Sorry to interrupt you. My question is the same as "Doesn't protect against Face++ Recognition in my test #67" (https://github.com/Shawn-Shan/fawkes/issues/67). I also have the same problem as Harry and hope to test Face++. I would appreciate it if you could send me a script with a local facial recognition model as you send to Harry. My e-mail address is [email protected]. What is more, the feature extractors in Table 1 of your paper are Web-Incept, Web_Dense, VGG2_Incept and VGG2_Dense. But in this release code, there are only extractors_0 and extractors_2. Can you kindly tell me what architectures and datasets these extractors is based on？ and is that possible to release adversarial training versions? Since in this paper you claimed that the adversarial training models performance better in the wild.

Thank you so much for spending time on my questions! Jun.

[huojinyong]

I have exactly the same problem as pidanhua! The method provided in the code does have little effect, which may be related to the difference between the feature extractor in the code and that in the paper, or it may not use adversarial traning. If anyone can also send me the Local Facial recognition model and model uesd adversarial traning and feature extractor in the paper etc, I would be very grateful! My email is [email protected] Thanks a million!