pickle.dump(fp_target, fp_target_pkl)
fp_target_pkl.close()
fp_target = util.np2var(fp_target, args.cuda)
fp = Fingerprints()
fp.dxs = fp_dx
fp.dys = fp_target
from model import CW2_Net as Net
#from res_model import ResNet as Net
from models import *
print("Train using model", Net)
model = Net()
if args.cuda:
model.cuda()
optimizer = optim.SGD(model.parameters(), lr=args.lr, weight_decay=1e-6, momentum=args.momentum)
print("Args:", args)
val_losses=[]
for epoch in range(1, args.epochs + 1):
if(epoch==1):
test_loss = fp_train.test(epoch, args, model, test_loader, fp.dxs, fp.dys, test_length=0.1*len(valid_dataset))
fp_train.train(epoch, args, model, optimizer, train_loader, fp.dxs, fp.dys)
test_loss = fp_train.test(epoch, args, model, test_loader, fp.dxs, fp.dys, test_length=0.1*len(valid_dataset))
val_losses.append(test_loss)
loss_flag = 1
pickle.dump(fp_target, fp_target_pkl)
fp_target_pkl.close()
fp_target = util.np2var(fp_target, args.cuda)
fp = Fingerprints()
fp.dxs = fp_dx
fp.dys = fp_target
from model import CW2_Net as Net
#from res_model import ResNet as Net
from models import *
print("Train using model", Net)
model = Net()
if args.cuda:
model.cuda()
optimizer = optim.SGD(model.parameters(),
lr=args.lr,
weight_decay=1e-6,
momentum=args.momentum)
print("Args:", args)
val_losses = []
for epoch in range(1, args.epochs + 1):
if (epoch == 1):
test_loss = fp_train.test(epoch,
args,
dataset = 'cifar'
list_adv_loader=[]
for advs in list_advs:
attack_file = os.path.join(args.adv_ex_dir, 'Adv_%s_%s.p' % (dataset, advs))
# FGSM attack is already shifted/normalized
adv_loader= torch.utils.data.DataLoader(
custom_datasets.Adv(filename=attack_file, transp=True),
batch_size=args.batch_size, shuffle=False, **kwargs)
list_adv_loader.append(adv_loader)
from model import CW2_Net as Net
#from small_model import Very_Small_Net as Net
print("Eval using model", Net)
model = Net()
print("Loading ckpt", args.ckpt)
model.load_state_dict(torch.load(args.ckpt))
model.train(False)
model.eval()
if args.cuda:
model.cuda()
print("Args:", args)
fixed_dxs = pickle.load(open(os.path.join(args.fingerprint_dir, "fp_inputs_dx.pkl"), "rb"))
fixed_dys = pickle.load(open(os.path.join(args.fingerprint_dir, "fp_outputs.pkl"), "rb"))
fp = Fingerprints()
fp.dxs = fixed_dxs
K.set_session(sess)
K.set_image_data_format('channels_first')
# Sample random test data
_, _, X_test, Y_test = get_data(dataset)
num_samples = np.shape(X_test)[0]
num_rand_samples = 1328
random_samples = np.random.randint(0, num_samples, num_rand_samples)
new_X_test = X_test[random_samples, :, :, :]
new_Y_test = Y_test[random_samples, :]
f = open(os.path.join(args.log_dir, 'Random_Test_%s_.p' % (dataset)), 'w')
pickle.dump({"adv_input": new_X_test, "adv_labels": new_Y_test}, f)
f.close()
if (args.attack == 'spsa' or args.attack == 'all'):
pytorch_network = Net()
pytorch_network.load_state_dict(torch.load(args_ckpt))
pytorch_network.eval()
model = Model(torch_model=pytorch_network)
keras_network = model.model
transfer.pytorch_to_keras(pytorch_network, model.model)
pytorch_network.eval()
model = model.model
model_logits = model
batch_size = 16
craft_one_type(sess,
model,
new_X_test,
new_Y_test,
dataset,
'spsa',