Steganografiya quyidagi sohalarda qo'llaniladi, lekin ular bilan cheklanmaydi

sampled = torch.utils.data.Subset(
self.trainset, random.choices(range(len(self.trainset)), k=500))
trainloader = torch.utils.data.DataLoader(
sampled, batch_size=1, shuffle=True)

for data, target in trainloader:

data, target = data.to(self.device), target.to(self.device)
data.requires_grad = True
output = self.model(data)
init_pred = output.max(1, keepdim=True)[1]

# If the initial prediction is wrong, pass
if init_pred.item() != target.item():
continue

# Calculate the loss
loss = self.args.criterion(output, target)
self.model.zero_grad()
loss.backward()
data_grad = data.grad.data

perturbed_data = self.__fgsm_attack(data, epsilon, data_grad)
output = self.model(perturbed_data)

final_pred = output.max(1, keepdim=True)[1]
if final_pred.item() != target.item():
shape = perturbed_data.shape
triggerset.append((perturbed_data.reshape(shape[1:]),
final_pred.item()))

triggerloader = torch.utils.data.DataLoader(
triggerset, batch_size=self.args.batch_size, shuffle=True)

trainloader, valloader, testloader = self.loaders()
return trainloader, valloader, testloader, triggerloader

def __is_prediction_dawn(self, item):
""" Verify if prediction should be
correct as in the DAWN paper.

Args:
item (array): Single query

Returns:
(bool): Should the input be correctly classified ?

"""
bound = floor((2 ** self.args.precision_dawn)
* self.args.probability_dawn)
hashed = hmac.new(
self.args.key_dawn.encode("utf-8"),