Using Saved model for generating adversarial samples.

[chamanbanolia]

I am not able to create an attack instance when I am using loaded saved model (.pth file) for generating adversarial samples, can you please help find out, what can be the potential issue in this.

Below is the error I am getting

RuntimeError Traceback (most recent call last)
Input In [25], in
1 from torchattacks import FGSM, PGD
----> 3 atk = FGSM(saved_model)
4 print(atk)

File /opt/conda/lib/python3.8/site-packages/torchattacks/attacks/fgsm.py:30, in FGSM.init(self, model, eps)
29 def init(self, model, eps=8 / 255):
---> 30 super().init("FGSM", model)
31 self.eps = eps
32 self.supported_mode = ["default", "targeted"]

File /opt/conda/lib/python3.8/site-packages/torchattacks/attack.py:40, in Attack.init(self, name, model)
37 self.attack = name
38 self._attacks = OrderedDict()
---> 40 self.set_model(model)
41 try:
42 self.device = next(model.parameters()).device

File /opt/conda/lib/python3.8/site-packages/torchattacks/attack.py:10, in wrapper_method..wrapper_func(self, *args, **kwargs)
9 def wrapper_func(self, *args, **kwargs):
---> 10 result = func(self, *args, **kwargs)
11 for atk in self.dict.get("_attacks").values():
12 eval("atk." + func.name + "(*args, **kwargs)")

File /opt/conda/lib/python3.8/site-packages/torchattacks/attack.py:73, in Attack.set_model(self, model)
71 @wrapper_method
72 def set_model(self, model):
---> 73 self.model = model
74 self.model_name = model.class.name

File /opt/conda/lib/python3.8/site-packages/torchattacks/attack.py:563, in Attack.setattr(self, name, value)
560 if isinstance(items, Attack):
561 yield items
--> 563 for num, value in enumerate(get_all_values(value)):
564 attacks[name + "." + str(num)] = value
565 for subname, subvalue in value.dict.get("_attacks").items():

File /opt/conda/lib/python3.8/site-packages/torchattacks/attack.py:555, in Attack.setattr..get_all_values(items, stack)
553 items = list(items.keys()) + list(items.values())
554 for item in items:
--> 555 yield from get_all_values(item, stack)
556 else:
557 if isinstance(items, Attack):

File /opt/conda/lib/python3.8/site-packages/torchattacks/attack.py:549, in Attack.setattr..get_all_values(items, stack)
548 def get_all_values(items, stack=[]):
--> 549 if items not in stack:
550 stack.append(items)
551 if isinstance(items, list) or isinstance(items, dict):

RuntimeError: Boolean value of Tensor with more than one value is ambiguous

[rikonaka]

Hi @chamanbanolia , can you provide your full code?