Attack saving and loading is not working

ain-soph / trojanzoo

TrojanZoo provides a universal pytorch platform to conduct security researches (especially backdoor attacks/defenses) of image classification in deep learning.

https://ain-soph.github.io/trojanzoo

GNU General Public License v3.0

274 stars 62 forks source link

Attack saving and loading is not working #189

Closed kshitijsachan closed 1 year ago

kshitijsachan commented 1 year ago

It seems like a lot of the attacks don't have saving implemented correctly (e.g. trojannet). This is making it hard for me to run backdoor_defense.py because it first loads the attacked model.

ain-soph commented 1 year ago

Please provide more details so that I could support you better.

Most backdoor attacks have default saving method implemented in the abstract class. https://github.com/ain-soph/trojanzoo/blob/58d309b399024be62782b285e7cc8cede147bbc8/trojanvision/attacks/abstract.py#L185-L192

ain-soph commented 1 year ago

Please provide further feedback. I'll close the issue if there is no response in a week.