def run():
torch.multiprocessing.freeze_support()
use_cuda=True
image_nc=1
epochs = 60
batch_size = 128
BOX_MIN = 0
BOX_MAX = 1
# Define what device we are using
print("CUDA Available: ",torch.cuda.is_available())
device = torch.device("cuda" if (use_cuda and torch.cuda.is_available()) else "cpu")
pretrained_model = "./MNIST_target_model.pth"
targeted_model = MNIST_target_net().to(device)
targeted_model.load_state_dict(torch.load(pretrained_model))
targeted_model.eval()
model_num_labels = 10
# MNIST train dataset and dataloader declaration
mnist_dataset = torchvision.datasets.MNIST('./dataset', train=True, transform=transforms.ToTensor(), download=True)
dataloader = DataLoader(mnist_dataset, batch_size=batch_size, shuffle=True, num_workers=1)
advGAN = AdvGAN_Attack(device,
targeted_model,
model_num_labels,
image_nc,
BOX_MIN,
BOX_MAX)
advGAN.train(dataloader, epochs)
if __name__ == "__main__":
use_cuda=True
image_nc=1
batch_size = 128
gen_input_nc = image_nc
# Define what device we are using
print("CUDA Available: ",torch.cuda.is_available())
device = torch.device("cuda" if (use_cuda and torch.cuda.is_available()) else "cpu")
# load the pretrained model
pretrained_model = "./MNIST_target_model.pth"
target_model = MNIST_target_net().to(device)
target_model.load_state_dict(torch.load(pretrained_model))
target_model.eval()
# load the generator of adversarial examples
pretrained_generator_path = './models/netG.pth.tar'
pretrained_G = models.Generator(gen_input_nc, image_nc).to(device)
pretrained_G.load_state_dict(torch.load(pretrained_generator_path))
pretrained_G.eval()
# test adversarial examples in MNIST training dataset
mnist_dataset = torchvision.datasets.MNIST('./dataset', train=True, transform=transforms.ToTensor(), download=True)
train_dataloader = DataLoader(mnist_dataset, batch_size=batch_size, shuffle=False, num_workers=1)
num_correct = 0
num_all = 0
for i, data in enumerate(train_dataloader, 0):
def run():
use_cuda = True
image_nc = 1
batch_size = 128
gen_input_nc = image_nc
# Define what device we are using
print("CUDA Available: ", torch.cuda.is_available())
device = torch.device("cuda" if (
use_cuda and torch.cuda.is_available()) else "cpu")
# load the pretrained model
pretrained_model = "./MNIST_target_model_2.pth"
target_model = MNIST_target_net().to(device)
target_model.load_state_dict(torch.load(pretrained_model))
target_model.eval()
# load the generator of adversarial examples
pretrained_generator_path = './models/netG_epoch_40.pth'
pretrained_G = models.Generator(gen_input_nc, image_nc).to(device)
pretrained_G.load_state_dict(torch.load(pretrained_generator_path))
pretrained_G.eval()
# test adversarial examples in MNIST training dataset
mnist_dataset = torchvision.datasets.MNIST('./dataset',
train=True,
transform=transforms.ToTensor(),
download=True)
train_dataloader = DataLoader(mnist_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=1)
num_correct = 0
for i, data in enumerate(train_dataloader, 0):
test_img, test_label = data
test_img, test_label = test_img.to(device), test_label.to(device)
perturbation = pretrained_G(test_img)
perturbation = torch.clamp(perturbation, -0.3, 0.3)
adv_img = perturbation + test_img
adv_img = torch.clamp(adv_img, 0, 1)
pred_lab = torch.argmax(target_model(adv_img), 1)
num_correct += torch.sum(pred_lab == test_label, 0)
print('MNIST training dataset:')
print('num_correct: ', num_correct.item())
print('accuracy of adv imgs in training set: %f\n' %
(num_correct.item() / len(mnist_dataset)))
# test adversarial examples in MNIST testing dataset
mnist_dataset_test = torchvision.datasets.MNIST(
'./dataset',
train=False,
transform=transforms.ToTensor(),
download=True)
test_dataloader = DataLoader(mnist_dataset_test,
batch_size=batch_size,
shuffle=False,
num_workers=1)
num_correct = 0
for i, data in enumerate(test_dataloader, 0):
test_img, test_label = data
test_img, test_label = test_img.to(device), test_label.to(device)
perturbation = pretrained_G(test_img)
perturbation = torch.clamp(perturbation, -0.3, 0.3)
adv_img = perturbation + test_img
adv_img = torch.clamp(adv_img, 0, 1)
pred_lab = torch.argmax(target_model(adv_img), 1)
num_correct += torch.sum(pred_lab == test_label, 0)
print('num_correct: ', num_correct.item())
print('accuracy of adv imgs in testing set: %f\n' %
(num_correct.item() / len(mnist_dataset_test)))
writer = SummaryWriter(log_base_dir)
else:
writer = None
if not os.path.exists(models_path):
os.makedirs(models_path)
if not os.path.exists(models_path + model_name):
os.makedirs(models_path + model_name)
# Define what device we are using print("CUDA Available: ",torch.cuda.is_available())
device = torch.device("cuda" if (
use_cuda and torch.cuda.is_available()) else "cpu")
pretrained_model = "./MNIST_target_model.pth"
targeted_model = MNIST_target_net().to(device)
targeted_model.load_state_dict(
torch.load(pretrained_model, map_location=device))
targeted_model.eval()
model_num_labels = 10
# MNIST train dataset and dataloader declaration
mnist_dataset = torchvision.datasets.MNIST('./dataset',
train=True,
transform=transforms.ToTensor(),
download=True)
dataloader = DataLoader(mnist_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=1)
advGAN = AdvGAN_Attack(device, targeted_model, model_num_labels, image_nc,
parser.add_argument('--data', default='./dataset/real_and_fake_face')
parser.add_argument('--seed', default=0, type=int)
parser.add_argument('--epsilon', default=0.3, type=float)
args = parser.parse_args()
for arg in vars(args):
print(arg, getattr(args, arg))
# Define what device we are using
print("CUDA Available: ", torch.cuda.is_available())
device = torch.device("cuda" if (use_cuda and torch.cuda.is_available()) else "cpu")
mnist_dataset = torchvision.datasets.MNIST('./dataset', train=True, transform=transforms.ToTensor(), download=True)
train_dataloader = DataLoader(mnist_dataset, batch_size=args.batch_size, shuffle=True, num_workers=1)
# training the target model
target_model = MNIST_target_net().to(device)
target_model.train()
opt_model = torch.optim.Adam(target_model.parameters(), lr=0.0001)
epochs = args.epochs
pgd = PGD(target_model, None, None, device, args.epsilon, 7, args.epsilon/4)
for epoch in range(epochs):
loss_epoch = 0
if epoch == 20:
opt_model = torch.optim.Adam(target_model.parameters(), lr=0.00001)
num_corrects = 0
adv_num_corrects = 0
total = 0
for i, data in enumerate(train_dataloader, 0):
train_imgs, train_labels = data
args = parser.parse_args()
for arg in vars(args):
print(arg, getattr(args, arg))
model_name = name_maker(args)
en_input_nc = image_nc
# Define what device we are using
print("CUDA Available: ", torch.cuda.is_available())
device = torch.device("cuda" if (use_cuda and torch.cuda.is_available()) else "cpu")
print()
# load the pretrained model
model_path = "./MNIST_target_model.pth"
target_model = MNIST_target_net().to(device)
target_model.load_state_dict(torch.load(model_path, map_location=device))
if args.parameters_count:
print('number of parameters(model):', parameters_count(target_model))
target_model.eval()
epoch = args.epoch
# load encoder & generators
E_path = models_path + model_name + 'E_epoch_{}.pth'.format(epoch)
E = models.Encoder(en_input_nc).to(device)
E.load_state_dict(torch.load(E_path, map_location=device))
if args.parameters_count:
print('number of parameters(E):', parameters_count(E))
E.eval()
