Bugs when running Linf attack on pt_inception about square-attack HOT 4 CLOSED

Hi,

Thanks for the interest in our paper!
It's a bit hard to tell what can be a problem here without having your version of the code. But I think you should just print logits_list and see why it's not possible to vertically concatenate it. Most likely, the model.predict() function in your version of the code returns something unexpected while it should return a matrix of logits of size number_of_samples x number_of_classes.

I hope that helps. Feel free to reopen the issue if there are more questions on this!

from square-attack.

Hi,

Thanks for the interest in our paper!
It's a bit hard to tell what can be a problem here without having your version of the code. But I think you should just print logits_list and see why it's not possible to vertically concatenate it. Most likely, the model.predict() function in your version of the code returns something unexpected while it should return a matrix of logits of size number_of_samples x number_of_classes.

I hope that helps. Feel free to reopen the issue if there are more questions on this!

Thanks for your valuable advice. It seems the problem comes from loading ImageNet. If possible, could you show me how your files arranged in this path IMAGENET_PATH = "/scratch/maksym/imagenet/val_orig" (Line 29, data.py)? Are val images and labels in the same directory?
My file is arranged like this:

But I found after loading the dataset, the y_test is all zeros.
Thanks very much for your help!

from square-attack.

Hi,

So in the directory IMAGENET_PATH = "/scratch/maksym/imagenet/val_orig", you should have folders named like n01440764, n01739381, n01978287, etc. In each folder there are images of a particular class.

For further details, you can check the documentation of ImageFolder:
https://pytorch.org/vision/0.8/datasets.html#imagefolder

I hope that helps!

from square-attack.

Hi,

So in the directory IMAGENET_PATH = "/scratch/maksym/imagenet/val_orig", you should have folders named like n01440764, n01739381, n01978287, etc. In each folder there are images of a particular class.

For further details, you can check the documentation of ImageFolder:
https://pytorch.org/vision/0.8/datasets.html#imagefolder

I hope that helps!

Thanks very much for your help!

from square-attack.