def main():
args = parser.parse_args()
pp = pprint.PrettyPrinter()
pp.pprint(vars(args))
# SVHN
# 26032 examples
mat_content = sio.loadmat('./SVHN/test_32x32.mat')
test_label_npy = mat_content['y']
test_label_npy[test_label_npy == 10] = 0
test_image_npy = mat_content['X'].astype(np.float32) / 256.0
test_image_npy = np.transpose(test_image_npy, (3, 0, 1, 2))
config = vars(args)
discriminator1 = models.get_discriminator(args.d_architecture1,
scope='discriminator1',
output_size=args.output_size,
c_dim=args.c_dim,
f_dim=args.df_dim,
is_training=True)
discriminator2 = models.get_discriminator(args.d_architecture2,
scope='discriminator2',
output_size=args.output_size,
c_dim=args.c_dim,
f_dim=args.df_dim,
is_training=True)
black_box_attacks(discriminator1, discriminator2, args.model_file1,
args.model_file2, test_image_npy, test_label_npy, config)
def main():
args = parser.parse_args()
pp = pprint.PrettyPrinter()
pp.pprint(vars(args))
# test data
mnist = input_data.read_data_sets('MNIST_data', one_hot=False)
# mnist only
# https://gist.github.com/noahfl/0b244346d4ad2501718bbb226be16b1e
test_image, test_label = mnist.test.images, mnist.test.labels
test_image = tf.reshape(test_image, [10000, 28, 28, 1])
test_image_batch = tf.train.shuffle_batch([test_image, test_label],
batch_size=args.batch_size,
enqueue_many=True,
num_threads=16,
capacity=10000 +
3 * args.batch_size,
min_after_dequeue=10000)
config = vars(args)
discriminator1 = models.get_discriminator(args.d_architecture1,
scope='discriminator1',
output_size=args.output_size,
c_dim=args.c_dim,
f_dim=args.df_dim,
is_training=True)
discriminator2 = models.get_discriminator(args.d_architecture2,
scope='discriminator2',
output_size=args.output_size,
c_dim=args.c_dim,
f_dim=args.df_dim,
is_training=True)
black_box_attacks(discriminator1, discriminator2, args.model_file1,
args.model_file2, test_image_batch, config)
def main():
args = parser.parse_args()
pp = pprint.PrettyPrinter()
pp.pprint(vars(args))
# Test data
test_filename_queue = get_filename_queue(
split_file=os.path.join(args.data_dir, 'splits', args.dataset,
'test.lst'),
data_dir=os.path.join(args.data_dir, args.dataset))
test_image, test_label = get_input_cifar10(test_filename_queue)
test_image_batch = tf.train.shuffle_batch([test_image, test_label],
batch_size=args.batch_size,
num_threads=16,
capacity=10000 +
3 * args.batch_size,
min_after_dequeue=10000)
config = vars(args)
discriminator1 = models.get_discriminator(args.d_architecture1,
scope='discriminator1',
output_size=args.output_size,
c_dim=args.c_dim,
f_dim=args.df_dim,
is_training=True)
discriminator2 = models.get_discriminator(args.d_architecture2,
scope='discriminator2',
output_size=args.output_size,
c_dim=args.c_dim,
f_dim=args.df_dim,
is_training=True)
black_box_attacks(discriminator1, discriminator2, args.model_file1,
args.model_file2, test_image_batch, config)
def main():
args = parser.parse_args()
pp = pprint.PrettyPrinter()
pp.pprint(vars(args))
# Data
filename_queue = get_filename_queue(
split_file=os.path.join(args.data_dir, 'splits', args.dataset,
args.split + '.lst'),
data_dir=os.path.join(args.data_dir, args.dataset))
# test data
test_filename_queue = get_filename_queue(
split_file=os.path.join(args.data_dir, 'splits', args.dataset,
'test.lst'),
data_dir=os.path.join(args.data_dir, args.dataset))
image, label = get_input_cifar10(filename_queue)
output_size = 32
c_dim = 3
test_image, test_label = get_input_cifar10(test_filename_queue)
image_batch = create_batch([image, label],
batch_size=args.batch_size,
num_preprocess_threads=16,
min_queue_examples=10000)
test_image_batch = tf.train.shuffle_batch([test_image, test_label],
batch_size=args.batch_size,
num_threads=16,
capacity=10000 +
3 * args.batch_size,
min_after_dequeue=10000)
config = vars(args)
discriminator = models.get_discriminator(args.d_architecture,
scope='discriminator',
output_size=output_size,
c_dim=args.c_dim,
f_dim=args.df_dim,
is_training=True)
gen_adv_examples(discriminator, args.model_file, image_batch,
test_image_batch, config)
def main():
args = parser.parse_args()
pp = pprint.PrettyPrinter()
pp.pprint(vars(args))
# SVHN
mat_content = sio.loadmat('./SVHN/train_32x32.mat')
label_npy = mat_content['y']
label_npy[label_npy==10] = 0
image_npy = mat_content['X'].astype(np.float32)/256.0
# extra data
extra_content = sio.loadmat('./SVHN/extra_32x32.mat')
extra_label_npy = extra_content['y']
extra_label_npy[extra_label_npy==10] = 0
extra_image_npy = extra_content['X'].astype(np.float32)/256.0
idx = np.random.choice(531131, 80000)
extra_label_npy = extra_label_npy[idx]
extra_image_npy = extra_image_npy[:,:,:,idx]
image_combined = np.concatenate((image_npy, extra_image_npy), axis=3)
image_combined = np.transpose(image_combined, (3,0,1,2))
label_combined = np.concatenate((label_npy, extra_label_npy), axis=0)
# 26032 examples
mat_content = sio.loadmat('./SVHN/test_32x32.mat')
test_label_npy = mat_content['y']
test_label_npy[test_label_npy==10] = 0
test_image_npy = mat_content['X'].astype(np.float32)/256.0
test_image_npy = np.transpose(test_image_npy, (3,0,1,2))
config = vars(args)
generator = models.get_generator(args.g_architecture,
output_size=64, c_dim=args.c_dim, f_dim=args.gf_dim)
discriminator = models.get_discriminator(args.d_architecture,
output_size=64, c_dim=args.c_dim, f_dim=args.df_dim)
train(generator, discriminator, config, image_combined, label_combined, test_image_npy, test_label_npy)
def main():
args = parser.parse_args()
pp = pprint.PrettyPrinter()
pp.pprint(vars(args))
# Data
mnist = input_data.read_data_sets('MNIST_data', one_hot=False)
# mnist only
# https://gist.github.com/noahfl/0b244346d4ad2501718bbb226be16b1e
image, label = mnist.train.images, mnist.train.labels
image = tf.reshape(image, [55000, 28, 28, 1])
image_batch = tf.train.shuffle_batch([image, label],
batch_size=args.batch_size,
enqueue_many=True,
num_threads=16,
capacity=10000 + 3 * args.batch_size,
min_after_dequeue=10000)
test_image, test_label = mnist.test.images, mnist.test.labels
test_image = tf.reshape(test_image, [10000, 28, 28, 1])
test_image_batch = tf.train.shuffle_batch([test_image, test_label],
batch_size=args.batch_size,
enqueue_many=True,
num_threads=16,
capacity=10000 +
3 * args.batch_size,
min_after_dequeue=10000)
config = vars(args)
generator = models.get_generator(args.g_architecture,
output_size=64,
c_dim=args.c_dim,
f_dim=args.gf_dim)
discriminator = models.get_discriminator(args.d_architecture,
output_size=64,
c_dim=args.c_dim,
f_dim=args.df_dim)
train(generator, discriminator, image_batch, test_image_batch, config)
def main():
args = parser.parse_args()
pp = pprint.PrettyPrinter()
pp.pprint(vars(args))
config = vars(args)
# EAT Data
mat_content = np.load(config['eat_train_data'])
label_npy = mat_content['labels'].astype(np.int)
image_npy = mat_content['examples'].astype(np.float32)
# test data
mat_content = np.load(config['eat_test_data'])
test_label_npy = mat_content['labels'].astype(np.int)
test_image_npy = mat_content['examples'].astype(np.float32)
discriminator = models.get_discriminator(args.d_architecture,
output_size=32, c_dim=args.c_dim, f_dim=args.df_dim)
train(discriminator, config, image_npy, label_npy, test_image_npy, test_label_npy)