Could you share the code to evaluate your model?

[Dzhange]

Hi, I'm trying to reproduce your experiment results recently, but encountered several technical problems(mostly from foolbox), I think it would be really helpful if you could share the evaluatation code. It would also prevent me from getting incorrect results from your model.

Thanks!

[tathagatv]

I believe sharing the torch, torchvision and foolbox versions will help. As of now I figured out that fmodel.batch_predictions() (used in attacks.ipynb) is only in foolbox v1.8.0 while the latest version is 3.3.1. But even after adjusting that, I am running into some other issue.

[xyli1905]

@lukas-schott Or would it be possible to share the parameters used in your experiments when running the attacks? They will be very useful in helping us to make sure our test codes are correct.

[jfsullivan1]

I believe sharing the torch, torchvision and foolbox versions will help. As of now I figured out that fmodel.batch_predictions() (used in attacks.ipynb) is only in foolbox v1.8.0 while the latest version is 3.3.1. But even after adjusting that, I am running into some other issue.

How did you get a previous version of foolbox? It would be useful to know for this particular use.

[tathagatv]

I believe sharing the torch, torchvision and foolbox versions will help. As of now I figured out that fmodel.batch_predictions() (used in attacks.ipynb) is only in foolbox v1.8.0 while the latest version is 3.3.1. But even after adjusting that, I am running into some other issue.

How did you get a previous version of foolbox? It would be useful to know for this particular use.

@jfsullivan1 this command will work pip install foolbox==1.8.0