This is a temp Pytorch implementation of PACA based on "Adversarial Example Detection beyond Image Space".

The folders image_test_disjoint and image_test are the training set and testing set, respectively. The arg 'adv_type' means the attack methods, e.g. PGD, DDN, CWL2.

ImageNet_PACA.py is the main script.

If you have any problem about the codes, mail me