device = torch.device("cuda:0")
device2 = torch.device("cuda:0")
seed = 2018
np.random.seed(seed=seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
cudnn.benchmark = True
sub_x, test_data, test_label = get_data(train_size=args.train_size)
# oracle_model = Substitute(model=ResNet18(), save_path='stl10_checkpoint0/resnet_ckpt.t7', device=device)
# net = torch.load('stl10_checkpoint0/rdcnn32_10000.pkl')
# oracle_model = Oracle(model=Rh(num_layers=2, kernel_size=3, n=n_features),save_path='None',\
# svm_path='model/checkpoints_svm_3_2/svm.pkl', device=device)
oracle_model = Oracle(model=None, save_path='None',\
svm_path='checkpoints/%s.pkl' % args.target, device=device)
substitute_model = Substitute(model=LinearModel(in_node=32*32*3,num_classes=2), device=device2)
a, b = stl10_bbox_sub(param=param, oracle_model=oracle_model, substitute_model=substitute_model, \
x_sub=sub_x, test_data=test_data, test_label=test_label, aug_epoch=param['data_aug'],\
samples_max=6400, n_epoch=param['nb_epochs'], fixed_lambda=param['lambda'])
print('\n\nFinal results:')
print('Oracle model evaluation on clean data #%d:'%(test_data.size(0)))
oracle_model.eval(x=test_data, y=test_label, batch_size=oracle_size)
print('Substitute model evaluation on clean data: #%d:'%(test_data.size(0)))
substitute_model.eval(x=test_data, y=test_label, batch_size=512)
# test_adv = get_adv(model=substitute_model, x=test_data, y=test_label, epsilon=param['epsilon'])
# print('Substitute model FGSM attack itself\'s accuracy on adversarial samples #%d:'%(test_adv.size(0)))
# substitute_model.eval(x=test_adv, y=test_label, batch_size=512)
# print('Oracle model FGSM attack\'s accuracy on adversarial samples #%d:'%(test_adv.size(0)))
# oracle_model.eval(x=test_adv, y=test_label, batch_size=512)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
cudnn.benchmark = True
binarize = True if args.binarize else False
sub_x, test_data, test_label = get_data(train_size=args.train_size)
# nd_1 = noise_data(test_data, 0.3)
# nd_2 = noise_data(test_data, 0.5)
# oracle_model = Substitute(model=ResNet18(), save_path='stl10_checkpoint0/resnet_ckpt.t7', device=device)
# net = torch.load('stl10_checkpoint0/rdcnn32_10000.pkl')
# oracle_model = Oracle(model=Rh(num_layers=2, kernel_size=3, n=n_features),save_path='None',\
# svm_path='model/checkpoints_svm_3_2/svm.pkl', device=device)
oracle_model = Oracle(model=None, save_path='None', \
# svm_path='checkpoints/%s' % args.target,
svm_path=['checkpoints/%s_%d.h5' % (args.target, i) for i in range(100)],
device=device)
substitute_model = Substitute(model=LinearModel(in_node=image_shape, num_classes=args.n_classes), device=device)
a, b, c, d = stl10_bbox_sub(param=param, oracle_model=oracle_model, substitute_model=substitute_model, \
x_sub=sub_x, test_data=test_data, test_label=test_label, aug_epoch=param['data_aug'], \
samples_max=6400, n_epoch=param['nb_epochs'], fixed_lambda=param['lambda'])
print('\n\nFinal results:')
print('Oracle model evaluation on clean data #%d:' % (test_data.size(0)))
oracle_model.eval(x=test_data, y=test_label, batch_size=oracle_size)
print('Substitute model evaluation on clean data: #%d:' % (test_data.size(0)))
substitute_model.eval(x=test_data, y=test_label, batch_size=512)
if not os.path.exists('results/%s' % args.dataset):
os.makedirs('results/%s' % args.dataset)
filename = 'results/%s/%s_%s_%s_%d.csv' % (args.dataset, args.target, str(param['epsilon']),
device = torch.device("cuda:0")
device2 = torch.device("cuda:0")
seed = 2018
np.random.seed(seed=seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
cudnn.benchmark = True
sub_x, test_data, test_label = get_data(train_size=args.train_size)
# oracle_model = Substitute(model=ResNet18(), save_path='imagenet_checkpoint0/resnet_ckpt.t7', device=device)
# net = torch.load('imagenet_checkpoint0/rdcnn32_10000.pkl')
# oracle_model = Oracle(model=Rh(num_layers=2, kernel_size=3, n=n_features),save_path='None',\
# svm_path='model/checkpoints_svm_3_2/svm.pkl', device=device)
oracle_model = Oracle(model=None, save_path='None',\
svm_path='checkpoints/%s.pkl' % args.target, device=device)
substitute_model = Substitute(model=LinearModel(in_node=224 * 224 * 3,
num_classes=2),
device=device2)
a, b = imagenet_bbox_sub(param=param, oracle_model=oracle_model, substitute_model=substitute_model, \
x_sub=sub_x, test_data=test_data, test_label=test_label, aug_epoch=param['data_aug'],\
samples_max=6400, n_epoch=param['nb_epochs'], fixed_lambda=param['lambda'])
print('\n\nFinal results:')
print('Oracle model evaluation on clean data #%d:' % (test_data.size(0)))
oracle_model.eval(x=test_data, y=test_label, batch_size=oracle_size)
print('Substitute model evaluation on clean data: #%d:' %
(test_data.size(0)))
substitute_model.eval(x=test_data, y=test_label, batch_size=512)
# test_adv = get_adv(model=substitute_model, x=test_data, y=test_label, epsilon=param['epsilon'])
# print('Substitute model FGSM attack itself\'s accuracy on adversarial samples #%d:'%(test_adv.size(0)))
# substitute_model.eval(x=test_adv, y=test_label, batch_size=512)