dalexanderch commited on
Commit
c0e9162
·
1 Parent(s): c7b4c51

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +12 -12
app.py CHANGED
@@ -32,8 +32,8 @@ def create_pattern_mnist(image, label, model):
32
  def fgsm_mnist(image, label, model, epsilon):
33
  pattern = create_pattern_mnist(image, label, model)
34
  adv_x = image + epsilon*pattern
35
- adv_x = tf.clip_by_value(adv_x, -1, 1)
36
- adv_x = adv_x * 0.5 + 0.5
37
  return adv_x.numpy()
38
 
39
 
@@ -46,8 +46,8 @@ def iterative_fgsm_mnist(image, label, model, epsilon, alpha, niter):
46
  adv_x = tf.clip_by_value(adv_x, image - epsilon, image+epsilon)
47
  adv_x = adv_x.numpy()
48
  adv_x = adv_x.reshape(adv_x.shape[1])
49
- adv_x = tf.clip_by_value(adv_x, -1, 1)
50
- adv_x = adv_x * 0.5 + 0.5
51
  return adv_x.numpy()
52
 
53
  def iterative_least_likely_fgsm_mnist(image, model, epsilon, alpha, niter, nb_classes):
@@ -61,8 +61,8 @@ def iterative_least_likely_fgsm_mnist(image, model, epsilon, alpha, niter, nb_cl
61
  adv_x = tf.clip_by_value(adv_x, image - epsilon, image+epsilon)
62
  adv_x = adv_x.numpy()
63
  adv_x = adv_x.reshape(adv_x.shape[1])
64
- adv_x = tf.clip_by_value(adv_x, -1, 1)
65
- adv_x = adv_x * 0.5 + 0.5
66
  return adv_x.numpy()
67
 
68
  # Attack functions cifar10
@@ -88,8 +88,8 @@ def create_pattern_cifar10(image, label, model):
88
  def fgsm_cifar10(image, label, model, epsilon):
89
  pattern = create_pattern_cifar10(image, label, model)
90
  adv_x = image + epsilon*pattern
91
- adv_x = tf.clip_by_value(adv_x, -1, 1)
92
- adv_x = adv_x * 0.5 + 0.5
93
  return adv_x.numpy()
94
 
95
 
@@ -102,8 +102,8 @@ def iterative_fgsm_cifar10(image, label, model, epsilon, alpha, niter):
102
  adv_x = tf.clip_by_value(adv_x, image - epsilon, image+epsilon)
103
  adv_x = adv_x.numpy()
104
  adv_x = adv_x.reshape((32,32,3))
105
- adv_x = tf.clip_by_value(adv_x, -1, 1)
106
- adv_x = adv_x * 0.5 + 0.5
107
  return adv_x.numpy()
108
 
109
  def iterative_least_likely_fgsm_cifar10(image, model, epsilon, alpha, niter, nb_classes):
@@ -117,8 +117,8 @@ def iterative_least_likely_fgsm_cifar10(image, model, epsilon, alpha, niter, nb_
117
  adv_x = tf.clip_by_value(adv_x, image - epsilon, image+epsilon)
118
  adv_x = adv_x.numpy()
119
  adv_x = adv_x.reshape((32,32,3))
120
- adv_x = tf.clip_by_value(adv_x, -1, 1)
121
- adv_x = adv_x * 0.5 + 0.5
122
  return adv_x.numpy()
123
 
124
  def fn(dataset, attack):
 
32
  def fgsm_mnist(image, label, model, epsilon):
33
  pattern = create_pattern_mnist(image, label, model)
34
  adv_x = image + epsilon*pattern
35
+ adv_x = tf.clip_by_value(adv_x, 0, 1)
36
+ # adv_x = adv_x * 0.5 + 0.5
37
  return adv_x.numpy()
38
 
39
 
 
46
  adv_x = tf.clip_by_value(adv_x, image - epsilon, image+epsilon)
47
  adv_x = adv_x.numpy()
48
  adv_x = adv_x.reshape(adv_x.shape[1])
49
+ adv_x = tf.clip_by_value(adv_x, 0, 1)
50
+ # adv_x = adv_x * 0.5 + 0.5
51
  return adv_x.numpy()
52
 
53
  def iterative_least_likely_fgsm_mnist(image, model, epsilon, alpha, niter, nb_classes):
 
61
  adv_x = tf.clip_by_value(adv_x, image - epsilon, image+epsilon)
62
  adv_x = adv_x.numpy()
63
  adv_x = adv_x.reshape(adv_x.shape[1])
64
+ adv_x = tf.clip_by_value(adv_x, 0, 1)
65
+ # adv_x = adv_x * 0.5 + 0.5
66
  return adv_x.numpy()
67
 
68
  # Attack functions cifar10
 
88
  def fgsm_cifar10(image, label, model, epsilon):
89
  pattern = create_pattern_cifar10(image, label, model)
90
  adv_x = image + epsilon*pattern
91
+ adv_x = tf.clip_by_value(adv_x, 0, 1)
92
+ # adv_x = adv_x * 0.5 + 0.5
93
  return adv_x.numpy()
94
 
95
 
 
102
  adv_x = tf.clip_by_value(adv_x, image - epsilon, image+epsilon)
103
  adv_x = adv_x.numpy()
104
  adv_x = adv_x.reshape((32,32,3))
105
+ adv_x = tf.clip_by_value(adv_x, 0, 1)
106
+ # adv_x = adv_x * 0.5 + 0.5
107
  return adv_x.numpy()
108
 
109
  def iterative_least_likely_fgsm_cifar10(image, model, epsilon, alpha, niter, nb_classes):
 
117
  adv_x = tf.clip_by_value(adv_x, image - epsilon, image+epsilon)
118
  adv_x = adv_x.numpy()
119
  adv_x = adv_x.reshape((32,32,3))
120
+ adv_x = tf.clip_by_value(adv_x, 0, 1)
121
+ # adv_x = adv_x * 0.5 + 0.5
122
  return adv_x.numpy()
123
 
124
  def fn(dataset, attack):