Open seopp opened 1 year ago
Hi, thanks for the great work. I'm trying to use autoattack.py to evaluate the robustness of my model. In my experiments, it seems that using autoattack.py in this repository versus autoattack.py in the auto-attack repository gives different results. I think I'm setting up the experiment incorrectly, as the two results should be the same. So I'm wondering what the difference is between autoattack.py in this repo and autoattack.py in the auto-attack repo. Thanks for providing such a great library.
Could you upload the code for me? In the latest version, the robust accuracy against AutoAttack was the same. Note that the randomness can result in a little bit different results.
❔ Any questions
Hi, thanks for the great work. I'm trying to use autoattack.py to evaluate the robustness of my model. In my experiments, it seems that using autoattack.py in this repository versus autoattack.py in the auto-attack repository gives different results. I think I'm setting up the experiment incorrectly, as the two results should be the same. So I'm wondering what the difference is between autoattack.py in this repo and autoattack.py in the auto-attack repo. Thanks for providing such a great library.