def load(self, **kwargs):
session = kwargs["session"]
assert isinstance(session, tf.Session)
x_input = tf.placeholder(tf.float32, shape=(None,) + self.x_shape)
logits = self.inference(
x_input,
reuse=False)
model_path = get_model_path('pgd_at')
url = 'https://www.dropbox.com/s/g4b6ntrp8zrudbz/adv_trained.zip?dl=1'
fname = os.path.join(model_path, url.split('/')[-1])
if not os.path.exists(fname):
if not os.path.exists(model_path):
os.makedirs(model_path)
from six.moves import urllib
urllib.request.urlretrieve(url, fname)
import zipfile
model_zip = zipfile.ZipFile(fname)
model_zip.extractall(model_path)
print('Extracted model in {}'.format(model_zip.namelist()[1]))
saver = tf.train.Saver()
checkpoint = tf.train.latest_checkpoint(model_path+'/adv_trained')
saver.restore(session, checkpoint)
def load(self, **kwargs):
session = kwargs["session"]
assert isinstance(session, tf.Session)
x_input = tf.placeholder(self.x_dtype, shape=(None, ) + self.x_shape)
with slim.arg_scope(resnet_v2.resnet_arg_scope()):
resnet_v2.resnet_v2_152(x_input,
num_classes=self.n_class,
is_training=False,
reuse=tf.AUTO_REUSE)
model_path = get_model_path('resnet_v2_152')
if not os.path.exists(model_path):
os.makedirs(model_path)
urllib.request.urlretrieve(
'http://download.tensorflow.org/models/resnet_v2_152_2017_04_14.tar.gz',
os.path.join(model_path, 'resnet_v2_152_2017_04_14.tar.gz'),
show_progress)
tar = tarfile.open(
os.path.join(model_path, 'resnet_v2_152_2017_04_14.tar.gz'))
file_names = tar.getnames()
for file_name in file_names:
tar.extract(file_name, model_path)
saver = tf.train.Saver(slim.get_model_variables(scope='resnet_v2'))
saver.restore(session, os.path.join(model_path, 'resnet_v2_152.ckpt'))
def load(self, **kwargs):
model_path = get_model_path('cifar10-convnet-15742544.mat')
mcn = sio.loadmat(model_path)
mcn_weights = dict()
mcn_weights['conv1.weights'] = mcn['net'][0][0][0][0][0][0][0][1][0][
0].transpose()
mcn_weights['conv1.bias'] = mcn['net'][0][0][0][0][0][0][0][1][0][
1].flatten()
mcn_weights['conv2.weights'] = mcn['net'][0][0][0][0][3][0][0][1][0][
0].transpose()
mcn_weights['conv2.bias'] = mcn['net'][0][0][0][0][3][0][0][1][0][
1].flatten()
mcn_weights['conv3.weights'] = mcn['net'][0][0][0][0][6][0][0][1][0][
0].transpose()
mcn_weights['conv3.bias'] = mcn['net'][0][0][0][0][6][0][0][1][0][
1].flatten()
mcn_weights['conv4.weights'] = mcn['net'][0][0][0][0][9][0][0][1][0][
0].transpose()
mcn_weights['conv4.bias'] = mcn['net'][0][0][0][0][9][0][0][1][0][
1].flatten()
mcn_weights['conv5.weights'] = mcn['net'][0][0][0][0][11][0][0][1][0][
0].transpose()
mcn_weights['conv5.bias'] = mcn['net'][0][0][0][0][11][0][0][1][0][
1].flatten()
for k in ['conv1', 'conv2', 'conv3', 'conv4', 'conv5']:
t = self.model.__getattr__(k)
assert t.weight.data.size() == mcn_weights['%s.weights' % k].shape
t.weight.data[:] = torch.from_numpy(mcn_weights['%s.weights' % k])
assert t.bias.data.size() == mcn_weights['%s.bias' % k].shape
t.bias.data[:] = torch.from_numpy(mcn_weights['%s.bias' % k])
def load(self, **kwargs):
session = kwargs["session"]
assert isinstance(session, tf.Session)
x_input = tf.placeholder(tf.float32, shape=(None, ) + self.x_shape)
with tf.contrib.framework.arg_scope(resnet_v2.resnet_arg_scope()):
logits, _ = resnet_v2.resnet_v2_50(x_input,
self.n_class,
is_training=False,
reuse=tf.AUTO_REUSE)
model_path = get_model_path('alp')
url = 'http://download.tensorflow.org/models/adversarial_logit_pairing/imagenet64_alp025_2018_06_26.ckpt.tar.gz'
fname = os.path.join(model_path, url.split('/')[-1])
if not os.path.exists(fname):
if not os.path.exists(model_path):
os.makedirs(model_path)
from six.moves import urllib
urllib.request.urlretrieve(url, fname, show_progress)
import tarfile
t = tarfile.open(fname)
t.extractall(model_path)
print('Extracted model')
saver = tf.train.Saver()
saver.restore(session, fname.split('.tar.gz')[0])
def load(self, **kwargs):
session = kwargs["session"]
assert isinstance(session, tf.Session)
keras.backend.set_session(session)
model_input = Input(shape=self.x_shape)
model_dic = {}
model_out = []
for i in range(3):
model_dic[str(i)] = resnet_v2(input=model_input,
depth=110,
num_classes=self.n_class)
model_out.append(model_dic[str(i)][2])
model_output = keras.layers.concatenate(model_out)
model = Model(inputs=model_input, outputs=model_output)
model_ensemble = keras.layers.Average()(model_out)
model_ensemble = Model(input=model_input, output=model_ensemble)
model_path = get_model_path('adp-ensemble')
fname = os.path.join(model_path, 'cifar10_ResNet110v2_model.200.h5')
if not os.path.exists(fname):
if not os.path.exists(model_path):
os.makedirs(model_path)
urllib.request.urlretrieve(
"http://ml.cs.tsinghua.edu.cn/~tianyu/ADP/pretrained_models/ADP_standard_3networks/cifar10_ResNet110v2_model.200.h5",
fname, show_progress)
datamean = os.path.join(model_path, 'mean.npy')
model.load_weights(fname)
self.model = model_ensemble
self.mean = np.load(datamean)
def load(self, **kwargs):
model_path = get_model_path('cifar_resnet_2px.pth')
try:
print(model_path)
checkpoint = torch.load(model_path)
self.model.load_state_dict(checkpoint['state_dict'][0])
self.eval()
except FileNotFoundError:
raise IOError
def load(self, session, **kwargs):
assert isinstance(session, tf.Session)
x_input = tf.placeholder(tf.float32, shape=(None,) + self.x_shape)
self.model.inference(x_input, self.num_residual_blocks, reuse=False)
model_path = get_model_path('resnet56-cifar.ckpt')
saver = tf.train.Saver(var_list=tf.global_variables())
saver.restore(session, model_path)
def load(self, **kwargs):
model_path = get_model_path('cifar10_vgg_rse.pth')
try:
checkpoint = torch.load(model_path)
from collections import OrderedDict
state_dict = OrderedDict()
for k, v in checkpoint.items():
name = k[7:]
state_dict[name] = v
self.model.load_state_dict(state_dict)
self.eval()
except FileNotFoundError:
raise IOError
def load(self, **kwargs):
model_path = get_model_path('model_cifar_wrn.pt')
if not os.path.exists(model_path):
if not os.path.exists(os.path.dirname(model_path)):
os.makedirs(os.path.dirname(model_path))
urllib.request.urlretrieve(
'http://people.virginia.edu/~yy8ms/TRADES/model_cifar_wrn.pt',
model_path, show_progress)
try:
checkpoint = torch.load(model_path)
self.model.load_state_dict(checkpoint)
self.eval()
except FileNotFoundError:
raise IOError
def load(self, **kwargs):
session = kwargs["session"]
assert isinstance(session, tf.Session)
x_input = tf.placeholder(tf.float32, shape=(None, ) + self.x_shape)
x_input = tf.transpose(x_input, [0, 3, 1, 2])
with TowerContext('', is_training=False):
with tf.variable_scope('', reuse=tf.AUTO_REUSE):
logits = self.model.get_logits(x_input)
model_path = get_model_path('R152-Denoise.npz')
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))
from six.moves import urllib
urllib.request.urlretrieve(url, model_path, show_progress)
get_model_loader(model_path).init(session)
def loadMat(self):
stats_path = get_model_path('stats.mat')
self.stats = sio.loadmat(stats_path)
self.mean = torch.from_numpy(self.stats['dataMean'][np.newaxis])
self.trans = torch.from_numpy(self.stats['Trans'].T)