compatible with py2 and py3

DataPoison/mnist_paddle.py

@@ -14,6 +14,7 @@
 
 from __future__ import print_function
 
+from builtins import range
 import os
 import argparse
 from PIL import Image

DataPoison/poison_mnist_paddle.py

@@ -14,6 +14,7 @@
 
 from __future__ import print_function
 
+from builtins import range
 import os
 import argparse
 from PIL import Image

DataPoison/poison_mnist_pytorch.py

@@ -1,3 +1,7 @@
+from __future__ import division
+from __future__ import print_function
+from builtins import range
+from past.utils import old_div
 import torch
 import torchvision
 import torch.nn as nn
@@ -90,7 +94,7 @@ def train(epoch):
     if batch_idx % log_interval == 0:
       print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
         epoch, batch_idx * len(data), len(train_loader.dataset),
-        100. * batch_idx / len(train_loader), loss.item()))
+        old_div(100. * batch_idx, len(train_loader)), loss.item()))
       train_losses.append(loss.item())
       train_counter.append(
         (batch_idx*64) + ((epoch-1)*len(train_loader.dataset)))
@@ -111,7 +115,7 @@ def test():
   test_losses.append(test_loss)
   print('\nTest set: Avg. loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
     test_loss, correct, len(test_loader.dataset),
-    100. * correct / len(test_loader.dataset)))
+    old_div(100. * correct, len(test_loader.dataset))))
 
 # Data poison
 poison = Net()
@@ -154,7 +158,7 @@ def p_test():
   test_losses.append(test_loss)
   print('\nTest set: Avg. loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
     test_loss, correct, len(test_loader.dataset),
-    100. * correct / len(test_loader.dataset)))
+    old_div(100. * correct, len(test_loader.dataset))))
 
 def poi_test():
   poison.eval()
@@ -176,7 +180,7 @@ def poi_test():
   test_losses.append(test_loss)
   print('\nPoisonTest set: Avg. loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
     test_loss, correct, len(test_loader.dataset),
-    100. * correct / len(test_loader.dataset)))
+    old_div(100. * correct, len(test_loader.dataset))))
 
 p_test()
 for epoch in range(1, n_epochs + 1):

advbox/adversary.py

@@ -17,6 +17,7 @@
 adversarial example.
 
 """
+from builtins import object
 import numpy as np
 import  logging
 logger=logging.getLogger(__name__)

advbox/attacks/base.py

@@ -14,14 +14,16 @@
 """
 The base model of the model.
 """
+from builtins import object
 import logging
 from abc import ABCMeta
 from abc import abstractmethod
 
 import numpy as np
+from future.utils import with_metaclass
 
 
-class Attack(object):
+class Attack(with_metaclass(ABCMeta, object)):
     """
     Abstract base class for adversarial attacks. `Attack` represent an
     adversarial attack which search an adversarial example. subclass should
@@ -31,7 +33,6 @@ class Attack(object):
         model(Model): an instance of the class advbox.base.Model.
 
     """
-    __metaclass__ = ABCMeta
 
     def __init__(self, model):
         self.model = model

advbox/attacks/cw.py

@@ -16,7 +16,9 @@
 L2 distance metrics especially
 """
 from __future__ import division
+from __future__ import print_function
 
+from builtins import range
 import logging
 import numpy as np
 
@@ -153,7 +155,7 @@ def _apply(self,
         if self.l2 == None:
             self.img_adv = img
         adv_label, adv_score = self._predict_adv(self.img_adv)
-        print ('predict label:', adv_label, 'softmax:', adv_score)
+        print(('predict label:', adv_label, 'softmax:', adv_score))
         # check adversary target if success
         self.img_adv = np.squeeze(self.img_adv)
         self.img_adv = self.img_adv.reshape(img.shape)

advbox/attacks/cw2_pytorch.py

@@ -18,7 +18,9 @@
 L2 distance metrics especially
 """
 from __future__ import division
+from __future__ import print_function
 
+from builtins import range
 import logging
 import numpy as np
 

advbox/attacks/deepfool.py

@@ -19,6 +19,8 @@
 """
 from __future__ import division
 
+from builtins import str
+from builtins import range
 import logging
 
 import numpy as np
@@ -75,7 +77,7 @@ def _apply(self, adversary, iterations=100, overshoot=0.02):
         gradient = self.model.gradient(adversary.original, pre_label)
         x = np.copy(adversary.original)
 
-        for iteration in xrange(iterations):
+        for iteration in range(iterations):
             w = np.inf
             w_norm = np.inf
             pert = np.inf

advbox/attacks/gradient_method.py

@@ -17,13 +17,15 @@
 """
 from __future__ import division
 
+from builtins import range
 import logging
 from collections import Iterable
 
 import numpy as np
 
 
 from .base import Attack
+from functools import reduce
 
 __all__ = [
     'GradientMethodAttack', 'FastGradientSignMethodAttack', 'FGSM',

advbox/attacks/lbfgs.py

@@ -16,6 +16,7 @@
 """
 from __future__ import division
 
+from builtins import range
 import logging
 
 import numpy as np

advbox/attacks/localsearch.py

@@ -17,6 +17,9 @@
 """
 from __future__ import division
 
+from builtins import zip
+from builtins import str
+from builtins import range
 import logging
 from collections import Iterable
 logger=logging.getLogger(__name__)

advbox/attacks/saliency.py

@@ -17,6 +17,7 @@
 """
 from __future__ import division
 
+from builtins import range
 import logging
 import random
 import numpy as np
@@ -74,7 +75,7 @@ def _apply(self,
             # count tracks how often each pixel was changed
             counts = np.zeros_like(original_image)
 
-            labels = range(self.model.num_classes())
+            labels = list(range(self.model.num_classes()))
             adv_img = original_image.copy()
             min_, max_ = self.model.bounds()
 
@@ -152,7 +153,7 @@ def _generate_random_target(self, original_label):
         num_classes = self.model.num_classes()
         assert num_random_target <= num_classes - 1
 
-        target_labels = random.sample(range(num_classes), num_random_target + 1)
+        target_labels = random.sample(list(range(num_classes)), num_random_target + 1)
         target_labels = [t for t in target_labels if t != original_label]
         target_labels = target_labels[:num_random_target]
 

advbox/attacks/tf/tools.py

@@ -20,6 +20,9 @@
 
 
 """
+from __future__ import division
+from builtins import range
+from past.utils import old_div
 import logging
 logger=logging.getLogger(__name__)
 
@@ -40,21 +43,21 @@ def fgsm(x,
     grad, = tf.gradients(loss, x)
 
     if  ord == 1:
-        red_ind = list(xrange(1, len(x.get_shape())))
+        red_ind = list(range(1, len(x.get_shape())))
         avoid_zero_div = 1e-8
         avoid_nan_norm = tf.maximum(avoid_zero_div,
                                     reduce_sum(tf.abs(grad),
                                                reduction_indices=red_ind,
                                                keepdims=True))
-        normalized_grad = grad / avoid_nan_norm
+        normalized_grad = old_div(grad, avoid_nan_norm)
     elif ord == 2:
-        red_ind = list(xrange(1, len(x.get_shape())))
+        red_ind = list(range(1, len(x.get_shape())))
         avoid_zero_div = 1e-8
         square = tf.maximum(avoid_zero_div,
                             reduce_sum(tf.square(grad),
                                        reduction_indices=red_ind,
                                        keepdims=True))
-        normalized_grad = grad / tf.sqrt(square)
+        normalized_grad = old_div(grad, tf.sqrt(square))
     else:
         normalized_grad = tf.sign(grad)
         normalized_grad = tf.stop_gradient(normalized_grad)
@@ -78,7 +81,7 @@ def deepfool(x,
 
     grad, = tf.gradients(loss, x)
 
-    r=grad*loss/tf.reduce_sum(tf.square(grad))
+    r=old_div(grad*loss,tf.reduce_sum(tf.square(grad)))
 
     #目标是让loss下降
     adv_x = x - r

advbox/defences/feature_squeezing.py

@@ -20,6 +20,8 @@
 
 
 """
+from __future__ import division
+from past.utils import old_div
 import logging
 logger=logging.getLogger(__name__)
 
@@ -40,14 +42,14 @@ def FeatureFqueezingDefence(x, y=None, bit_depth=None, clip_values=(0.0, 1.0)):
 
     #归一化到(0,1)
     x_ = x - LB
-    x_ = x_ / (UB-LB)
+    x_ = old_div(x_, (UB-LB))
 
     #转换到bit_depth整数
     max_value = np.rint(2 ** bit_depth - 1)
     res = np.rint(x_ * max_value)
 
     #还原到clip_values范围
-    res= res/max_value* (UB - LB) + LB
+    res= old_div(res,max_value)* (UB - LB) + LB
 
     #确保万无一失 clip生效
     assert (res <= UB).all() and (res >= LB).all()

advbox/defences/thermometer_encoding.py

@@ -20,6 +20,7 @@
 
 
 """
+from builtins import range
 import logging
 logger=logging.getLogger(__name__)
 
@@ -37,7 +38,7 @@ def _perchannel(x,num_space):
 
     onehot_rep = to_categorical(pos.reshape(-1), num_space)
 
-    for i in reversed(range(1, num_space)):
+    for i in reversed(list(range(1, num_space))):
         onehot_rep[:, i] += np.sum(onehot_rep[:, :i], axis=1)
 
 

advbox/models/base.py

@@ -14,13 +14,17 @@
 """
 The base model of the model.
 """
+from __future__ import division
+from builtins import object
+from past.utils import old_div
 from abc import ABCMeta
 from abc import abstractmethod
 
 import numpy as np
+from future.utils import with_metaclass
 
 
-class Model(object):
+class Model(with_metaclass(ABCMeta, object)):
     """
     Base class of model to provide attack.
 
@@ -31,7 +35,6 @@ class Model(object):
         preprocess(tuple): Two element tuple used to preprocess the input.
             First substract the first element, then divide the second element.
     """
-    __metaclass__ = ABCMeta
 
     def __init__(self, bounds, channel_axis, preprocess=None):
         assert len(bounds) == 2
@@ -72,7 +75,7 @@ def _process_input(self, input_):
             res = input_ - sub
         if not np.all(sub == 1):
             if res is None:  # "res = input_ - sub" is not executed!
-                res = input_ / div
+                res = old_div(input_, div)
             else:
                 res /= div
         if res is None:  # "res = (input_ - sub)/ div" is not executed!

advbox/models/keras.py

@@ -17,6 +17,7 @@
 """
 from __future__ import absolute_import
 
+from builtins import str
 import numpy as np
 import os
 

advbox/models/pytorch.py

@@ -16,6 +16,7 @@
 Pytorch model
 """
 from __future__ import absolute_import
+from __future__ import print_function
 
 import numpy as np
 import os

advbox/models/tensorflow.py

@@ -17,6 +17,8 @@
 """
 from __future__ import absolute_import
 
+from builtins import str
+from builtins import range
 import numpy as np
 import os