Problem with creating attacks/ adversarial instances using pytorch (in example code)

[filbart]

Hello, I've run into a similiar issue as #194 while trying to run example code. I am recieving a following error while creating the attack ( adversarial = attack(image, label) )

RuntimeError: Expected object of type torch.LongTensor but found type torch.IntTensor for argument #2 'target'

I've checked and updated versions of foolbox, torch, torchvision, numpy and scipy. Also tried rewriting the example slightly (code below), still getting the same error.

import foolbox
import torch
import torchvision
import torchvision.models as models
import numpy as np

device = torch.device('cpu')

resnet18 = models.resnet18(pretrained=True).to(device).eval()

mean = np.array([0.485, 0.456, 0.406]).reshape((3, 1, 1))
std = np.array([0.229, 0.224, 0.225]).reshape((3, 1, 1))
fmodel = foolbox.models.PyTorchModel(
    resnet18, bounds=(0, 1), num_classes=1000, preprocessing=(mean, std), device=device)

image, label = foolbox.utils.imagenet_example(data_format='channels_first')
image = image / 255.  # because our model expects values in [0, 1]

print('label', label)
print('predicted class', np.argmax(fmodel.predictions(image)))

attack = foolbox.attacks.FGSM(fmodel)
adversarial = attack(image, label)

print('adversarial class', np.argmax(fmodel.predictions(adversarial)))

[samuelemarro]

Converting standard types to Torch tensors is a bit messy and often leads to bugs such as this one. As a workaround, replace attack(image, label) with attack(image, np.array(label, dtype=np.int64)).

[filbart]

Works like a charm, thanks!

[wielandbrendel]

Thanks for reporting and thanks for the work-around! I fixed this problem in master.