About obfuscation attacks on SphereFace, I got only 28.4%ASR, while FGSM got around 95%ASR
About impersonation/target attacks on FaceNet, according to the accuracy.txt within the early 33 epochs, it's already above 20.85%(the accuracy result in paper. I change it to mode = 'target' in it for impersonation attacks and the max_perturbation to 8.0, then I run the train.py.What parts did I miss or do wrong steps in the process of my experiment？
About obfuscation attacks on SphereFace, I got only 28.4%ASR, while FGSM got around 95%ASR
About impersonation/target attacks on FaceNet, according to the accuracy.txt within the early 33 epochs, it's already above 20.85%(the accuracy result in paper. I change it to mode = 'target' in it for impersonation attacks and the max_perturbation to 8.0, then I run the train.py.What parts did I miss or do wrong steps in the process of my experiment？