Do you train with TRADES and ALP or one of them?

dvlab-research / LBGAT

Learnable Boundary Guided Adversarial Training (ICCV2021)

https://arxiv.org/abs/2011.11164

MIT License

34 stars 2 forks source link

Open headscott opened 1 month ago

headscott commented 1 month ago

When I run

python train_lbgat_cifar100.py -mark cifar100_lbgat0_0.031 -teacher_model ResNet18_cifar --beta 0

will the model be trained with ALP or TRADES? If not, how can I activate it?

jiequancui commented 1 month ago

Thanks for your interest in our work. For the TRADES baseline, we use their official code. Please refer to https://github.com/yaodongyu/TRADES.

headscott commented 1 month ago

Yeah, I know TRADES, but is this method also used, when I run the mentioned command? Or do I need special parameters?

headscott commented 1 month ago

And I saw in your paper, that you the Teacher Model is trained too, but I don't see, that the teacher starts its train mode by calling:

model_teacher.train()

I mean, there is no call of this type. So is the model_teacher pre-trained then?