def generateAdv(self, num_steps, epsilon, alpha):
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
if (torch.cuda.get_device_capability(0) < (5, 0)):
device = torch.device("cpu")
model = modelcnn.Net()
model.loadModel(pretrained_model="saved_model_state_CNN_final.pth")
adv = AdvGenerator()
model = model.to(device)
testloader = adv.loadData()
for target_label in adv.labels[:4]:
for data in testloader:
image, labels = data
image, labels = image.to(device), labels.to(device)
image.requires_grad = True
output = model.forward(image)
adv.createIterativeAdversarial(image, target_label.item(),
output, epsilon, alpha,
num_steps, model)
adv.moveUsedImage()
for data in testloader:
image, labels = data
image, labels = image.to(device), labels.to(device)
image.requires_grad = True
output = model.forward(image)
adv.createIterativeAdversarial(image, target_label.item(),
output, epsilon, alpha,
num_steps, model)
adv.moveUsedImage()
if __name__ == "__main__":
model = modelcnn.Net()
model.loadModel(pretrained_model="saved_model_state_CNN_final.pth")
adv = AdvGenerator()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = model.to(device)
testloader = adv.loadData()
for target_label in adv.labels[:4]:
for data in testloader:
image, labels = data
image, labels = image.to(device), labels.to(device)
image.requires_grad = True
output = model.forward(image)
adv.createIterativeAdversarial(image, target_label.item(), output,