RuntimeError in ProjectedGradientDescentPyTorch due to Device Mismatch

[windshadow233]

Describe the bug
A RuntimeError occurs due to a device mismatch when using torch.ones(1) (created on the CPU by default) with values_tmp (on the GPU).

To Reproduce
Steps to reproduce the behavior:

1. Create an instance of PyTorchClassifier on GPU.
2. Create an instance of ProjectedGradientDescent to attack the PyTorchClassifier model.
3. Try to generate adversarial data with the attack.generate method.
4. Raise the error: "RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!"

Expected behavior
The adversarial data generation process should complete without errors.

Error Messages

Traceback (most recent call last):                                                                                                                                              
  [...]
  File "adversarial-robustness-toolbox/art/attacks/evasion/projected_gradient_descent/projected_gradient_descent.py", line 202, in generate
    return self._attack.generate(x=x, y=y, **kwargs)
  File "adversarial-robustness-toolbox/art/attacks/evasion/projected_gradient_descent/projected_gradient_descent_pytorch.py", line 223, in generate
    adv_x[batch_index_1:batch_index_2] = self._generate_batch(
  File "adversarial-robustness-toolbox/art/attacks/evasion/projected_gradient_descent/projected_gradient_descent_pytorch.py", line 284, in _generate_batch
    adv_x = self._compute_pytorch(
  File "adversarial-robustness-toolbox/art/attacks/evasion/projected_gradient_descent/projected_gradient_descent_pytorch.py", line 451, in _compute_pytorch
    perturbation = self._projection(x_adv - x_init, eps, self.norm)
  File "adversarial-robustness-toolbox/art/attacks/evasion/projected_gradient_descent/projected_gradient_descent_pytorch.py", line 500, in _projection
    values_norm == 0, torch.minimum(torch.ones(1), torch.tensor(eps).to(values_tmp.device) / values_norm)
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

Suggested solution

Just replace line 500, in _projection function in "adversarial-robustness-toolbox/art/attacks/evasion/projected_gradient_descent/projected_gradient_descent_pytorch.py":

From

values_norm == 0, torch.minimum(torch.ones(1), torch.tensor(eps).to(values_tmp.device) / values_norm)

To

values_norm == 0, torch.minimum(torch.ones(1).to(values_tmp.device), torch.tensor(eps).to(values_tmp.device) / values_norm)

System information (please complete the following information):

• OS: Linux-5.15.0-105-generic-x86_64-with-glibc2.31
• Python version: 3.9.18
• ART version: 1.18.2
• PyTorch version: 2.5.1+cu118

[beat-buesser]

Hi @windshadow233 Thank you very much for reporting this issue. We'll include a fix in ART 1.19.1.