def download(model_path):
pth_path = os.path.join(model_path, 'cifar_resnet_2px.pth')
if not os.path.exists(pth_path):
if not os.path.exists(os.path.dirname(pth_path)):
os.makedirs(os.path.dirname(pth_path))
url = 'https://github.com/locuslab/convex_adversarial/blob/master/models_scaled/cifar_resnet_2px.pth?raw=true'
download_res(url, pth_path)
def download(model_path):
mat_name = 'cifar10-convnet-15742544.mat'
mat_path = os.path.abspath(os.path.join(model_path, mat_name))
tar_path = os.path.abspath(
os.path.join(model_path, 'mnist-cifar10-data-model.tar'))
if not os.path.exists(mat_path):
if not os.path.exists(os.path.dirname(tar_path)):
os.makedirs(os.path.dirname(tar_path), exist_ok=True)
if not os.path.exists(tar_path):
print(
'Please download "mnist-cifar10-data-model.tar" from ' +
'"https://drive.google.com/open?id=15xoZ-LUbc9GZpTlxmCJmvL_DR2qYEu2J", '
+ 'and save it to "{}".'.format(tar_path))
return
import tarfile
with tarfile.TarFile(tar_path) as f:
mat_file = f.extractfile('data/' + mat_name)
with open(mat_path, 'wb+') as t:
t.write(mat_file.read())
os.remove(tar_path)
stats_path = os.path.abspath(os.path.join(model_path, 'stats.mat'))
if not os.path.exists(stats_path):
download_res(
'http://ml.cs.tsinghua.edu.cn/~xiaoyang/downloads/realsafe/DeepDefense/stats.mat',
stats_path)
def download(model_path):
if not os.path.exists(model_path):
import tarfile
os.makedirs(model_path)
download_res('http://download.tensorflow.org/models/ens4_adv_inception_v3_2017_08_18.tar.gz',
os.path.join(model_path, 'ens4_adv_inception_v3_2017_08_18.tar.gz'))
tar = tarfile.open(os.path.join(model_path, 'ens4_adv_inception_v3_2017_08_18.tar.gz'))
file_names = tar.getnames()
for file_name in file_names:
tar.extract(file_name, model_path)
session = tf.Session(config=tf.ConfigProto(device_count={'GPU': 0}))
reader = tf.train.load_checkpoint(os.path.join(model_path, 'ens4_adv_inception_v3.ckpt'))
var_list = list(reader.get_variable_to_dtype_map().keys())
for var in var_list:
tf.Variable(reader.get_tensor(var), name=var.replace('InceptionV3', 'Ens4InceptionV3'))
session.run(tf.global_variables_initializer())
import shutil
shutil.rmtree(model_path)
os.makedirs(model_path)
saver = tf.train.Saver()
saver.save(session, os.path.join(model_path, 'ens4_adv_inception_v3.ckpt'))
def download(model_path):
if not os.path.exists(model_path):
if not os.path.exists(os.path.dirname(model_path)):
os.makedirs(os.path.dirname(model_path))
download_res(
'http://ml.cs.tsinghua.edu.cn/~xiaoyang/downloads/resnet56-cifar.ckpt',
model_path)
def download(model_path):
url = 'http://download.tensorflow.org/models/adversarial_logit_pairing/imagenet64_alp025_2018_06_26.ckpt.tar.gz'
if not os.path.exists(model_path + '.meta'):
if not os.path.exists(os.path.dirname(model_path)):
os.makedirs(os.path.dirname(model_path))
import tarfile
download_res(url, model_path + '.tar.gz')
t = tarfile.open(model_path + '.tar.gz')
t.extractall(os.path.dirname(model_path))
os.remove(model_path + '.tar.gz')
def download(model_path):
if not os.path.exists(os.path.join(model_path, 'models/adv_trained/checkpoint')):
zip_path = os.path.join(model_path, 'adv_trained.zip')
if not os.path.exists(zip_path):
if not os.path.exists(os.path.dirname(zip_path)):
os.makedirs(os.path.dirname(zip_path))
download_res('https://www.dropbox.com/s/g4b6ntrp8zrudbz/adv_trained.zip?dl=1', zip_path)
import zipfile
zipfile.ZipFile(zip_path).extractall(model_path)
os.remove(zip_path)
def download(model_path):
if not os.path.exists(model_path):
if not os.path.exists(os.path.dirname(model_path)):
os.makedirs(os.path.dirname(model_path), exist_ok=True)
gz_path = os.path.join(os.path.dirname(model_path), 'inception_v3_2016_08_28.tar.gz')
if not os.path.exists(gz_path):
download_res('http://download.tensorflow.org/models/inception_v3_2016_08_28.tar.gz', gz_path)
import tarfile
with tarfile.open(gz_path) as tar:
tar.extractall(os.path.dirname(model_path))
os.remove(gz_path)
def download(model_path):
if not os.path.exists(model_path):
os.makedirs(model_path)
h5_name = 'cifar10_ResNet110v2_model.200.h5'
h5_path = os.path.join(model_path, h5_name)
if not os.path.exists(h5_path):
url = 'http://ml.cs.tsinghua.edu.cn/~tianyu/ADP/pretrained_models/ADP_standard_3networks/' + h5_name
download_res(url, h5_path)
npy_path = os.path.join(model_path, 'mean.npy')
if not os.path.exists(npy_path):
download_res(
'http://ml.cs.tsinghua.edu.cn/~yinpeng/downloads/adp_datamean.npy',
npy_path)
r_filenames, r_labels, r_targets = [], [], []
if targets is None:
for filename, label in zip(filenames, labels):
if label == target_label:
r_filenames.append(filename)
r_labels.append(label)
return r_filenames, r_labels, None
else:
for filename, label, target in zip(filenames, labels, targets):
if label == target_label:
r_filenames.append(filename)
r_labels.append(label)
r_targets.append(target)
return r_filenames, r_labels, r_targets
if __name__ == '__main__':
if not os.path.exists(PATH_VAL_TXT):
os.makedirs(os.path.dirname(PATH_VAL_TXT), exist_ok=True)
download_res('http://ml.cs.tsinghua.edu.cn/~yinpeng/downloads/val.txt',
PATH_VAL_TXT)
if not os.path.exists(PATH_TARGET_TXT):
os.makedirs(os.path.dirname(PATH_TARGET_TXT), exist_ok=True)
download_res(
'http://ml.cs.tsinghua.edu.cn/~yinpeng/downloads/target.txt',
PATH_TARGET_TXT)
if not os.path.exists(PATH_IMGS):
print(
'Please download "ILSVRC2012_img_val.tar" from "http://www.image-net.org/download-images", '
+ 'and extract it to "{}".'.format(PATH_IMGS))
def download(model_path):
url = 'https://github.com/facebookresearch/ImageNet-Adversarial-Training/releases/download/v0.1/R152-Denoise.npz'
if not os.path.exists(model_path):
if not os.path.exists(os.path.dirname(model_path)):
os.makedirs(os.path.dirname(model_path), exist_ok=True)
download_res(url, model_path)
xs_test[offset:]), tf.data.Dataset.from_tensor_slices(ys_test[offset:])
ys = ys.map(lambda y: tf.cast(y, label_dtype))
ts = ts.map(lambda t: tf.cast(t, label_dtype))
ids = tf.data.Dataset.range(offset, len(ys_test) - offset)
if target_label is not None:
if load_target:
return tf.data.Dataset.zip(
(ids, xs, ys,
ts)).filter(lambda *ps: tf.math.equal(ps[2], target_label))
else:
return tf.data.Dataset.zip(
(ids, xs,
ys)).filter(lambda *ps: tf.math.equal(ps[2], target_label))
if load_target:
return tf.data.Dataset.zip((ids, xs, ys, ts))
else:
return tf.data.Dataset.zip((ids, xs, ys))
if __name__ == '__main__':
import os
from realsafe.utils import download_res
_ = load_data()
if not os.path.exists(PATH_TARGET):
os.makedirs(os.path.dirname(PATH_TARGET), exist_ok=True)
download_res('https://ml.cs.tsinghua.edu.cn/~qian/realsafe/target.npy',
PATH_TARGET)
