29 lines (11 loc) · 523 Bytes

Cifar-Attack

How to use

using train.sh or train_pgd.sh to train a model to be attacked, where train_pgd.sh refers to PGD-based adversarial training
using attack.sh to make adversarial samples

Note

utils.py contains some tools to generate stronger adversarial samples
- Gaussian Ambiguity
- Logits Merge
examples contains some adversarial samples generated by using this repo

Special thanks to torchattacks