forked from TrustAI/DeepGame
-
Notifications
You must be signed in to change notification settings - Fork 3
/
CompetitiveAlphaBeta.py
223 lines (193 loc) · 8.95 KB
/
CompetitiveAlphaBeta.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
#!/usr/bin/python
# -*- coding: utf-8 -*-
"""
Construct a CompetitiveAlphaBeta class to compute
the lower bound of Player I’s maximum adversary distance
while Player II being competitive.
Author: Min Wu
Email: [email protected]
"""
from numpy import inf
from FeatureExtraction import *
from basics import *
class CompetitiveAlphaBeta:
def __init__(self, image, model, eta, tau, bounds=(0, 1)):
self.IMAGE = image
self.IMAGE_BOUNDS = bounds
self.MODEL = model
self.DIST_METRIC = eta[0]
self.DIST_VAL = eta[1]
self.TAU = tau
self.LABEL, _ = self.MODEL.predict(self.IMAGE)
feature_extraction = FeatureExtraction(pattern='grey-box')
self.PARTITIONS = feature_extraction.get_partitions(self.IMAGE, self.MODEL, num_partition=10)
self.ALPHA = {}
self.BETA = {}
self.MANI_BETA = {}
self.MANI_DIST = {}
self.CURRENT_MANI = ()
self.ROBUST_FEATURE_FOUND = False
self.ROBUST_FEATURE = []
self.FRAGILE_FEATURE_FOUND = False
self.LEAST_FRAGILE_FEATURE = []
print("Distance metric %s, with bound value %s." % (self.DIST_METRIC, self.DIST_VAL))
def target_pixels(self, image, pixels):
(row, col, chl) = image.shape
atomic_manipulations = []
manipulated_images = []
for (x, y) in pixels:
for z in range(chl):
atomic = (x, y, z, 1 * self.TAU)
valid, atomic_image = self.apply_atomic_manipulation(image, atomic)
if valid is True:
manipulated_images.append(atomic_image)
atomic_manipulations.append(atomic)
atomic = (x, y, z, -1 * self.TAU)
valid, atomic_image = self.apply_atomic_manipulation(image, atomic)
if valid is True:
manipulated_images.append(atomic_image)
atomic_manipulations.append(atomic)
manipulated_images = np.asarray(manipulated_images)
probabilities = self.MODEL.model.predict(manipulated_images)
labels = probabilities.argmax(axis=1)
for idx in range(len(manipulated_images)):
if not diffImage(manipulated_images[idx], self.IMAGE):
continue
dist = self.cal_distance(manipulated_images[idx], self.IMAGE)
if labels[idx] != self.LABEL:
self.MANI_BETA.update({self.CURRENT_MANI + atomic_manipulations[idx]: dist})
self.MANI_DIST.update({self.CURRENT_MANI + atomic_manipulations[idx]: dist})
else:
self.MANI_BETA.update({self.CURRENT_MANI + atomic_manipulations[idx]: inf})
self.MANI_DIST.update({self.CURRENT_MANI + atomic_manipulations[idx]: dist})
def apply_atomic_manipulation(self, image, atomic):
atomic_image = image.copy()
chl = atomic[0:3]
manipulate = atomic[3]
if (atomic_image[chl] >= max(self.IMAGE_BOUNDS) and manipulate >= 0) or (
atomic_image[chl] <= min(self.IMAGE_BOUNDS) and manipulate <= 0):
valid = False
return valid, atomic_image
else:
if atomic_image[chl] + manipulate > max(self.IMAGE_BOUNDS):
atomic_image[chl] = max(self.IMAGE_BOUNDS)
elif atomic_image[chl] + manipulate < min(self.IMAGE_BOUNDS):
atomic_image[chl] = min(self.IMAGE_BOUNDS)
else:
atomic_image[chl] += manipulate
valid = True
return valid, atomic_image
def cal_distance(self, image1, image2):
if self.DIST_METRIC == 'L0':
return l0Distance(image1, image2)
elif self.DIST_METRIC == 'L1':
return l1Distance(image1, image2)
elif self.DIST_METRIC == 'L2':
return l2Distance(image1, image2)
else:
print("Unrecognised distance metric. "
"Try 'L0', 'L1', or 'L2'.")
def play_game(self, image):
for partitionID, pixels in self.PARTITIONS.items():
self.MANI_BETA = {}
self.MANI_DIST = {}
self.CURRENT_MANI = ()
print("partition ID:", partitionID)
self.target_pixels(image, pixels)
while min(self.MANI_BETA.values()) is inf and min(self.MANI_DIST.values()) <= self.DIST_VAL:
min_dist = min(self.MANI_BETA.values())
print("Current min distance:", min_dist)
print("Adversary not found.")
mani_distance = copy.deepcopy(self.MANI_BETA)
for atom, _ in mani_distance.items():
self.MANI_BETA.pop(atom)
self.CURRENT_MANI = atom
self.MANI_DIST.pop(atom)
new_image = copy.deepcopy(self.IMAGE)
atomic_list = [atom[i:i + 4] for i in range(0, len(atom), 4)]
for atomic in atomic_list:
valid, new_image = self.apply_atomic_manipulation(new_image, atomic)
self.target_pixels(new_image, pixels)
if min(self.MANI_BETA.values()) > self.DIST_VAL or min(self.MANI_DIST.values()) > self.DIST_VAL:
print("Distance:", )
print("Adversarial distance exceeds distance bound.")
self.BETA.update({partitionID: None})
elif min(self.MANI_BETA.values()) is not inf:
print("Adversary found.")
adv_mani = min(self.MANI_BETA, key=self.MANI_BETA.get)
print("Manipulations:", adv_mani)
adv_dist = self.MANI_BETA[adv_mani]
print("Distance:", adv_dist)
self.BETA.update({partitionID: [adv_mani, adv_dist]})
for partitionID, beta in self.MANI_BETA:
print(partitionID, beta)
if beta is None:
print("Feature %s is robust." % partitionID)
self.MANI_BETA.pop(partitionID)
self.ROBUST_FEATURE_FOUND = True
self.ROBUST_FEATURE.append(partitionID)
if self.MANI_BETA:
self.FRAGILE_FEATURE_FOUND = True
self.ALPHA = max(self.BETA, key=self.BETA.get)
self.LEAST_FRAGILE_FEATURE = self.ALPHA
print("Among fragile features, the least fragile feature is:\n"
% self.LEAST_FRAGILE_FEATURE)
"""
def play_game(self, image):
self.player1(image)
for partitionID, beta in self.MANI_BETA:
print(partitionID, beta)
if beta is None:
print("Feature %s is robust." % partitionID)
self.MANI_BETA.pop(partitionID)
self.ROBUST_FEATURE_FOUND = True
self.ROBUST_FEATURE.append(partitionID)
if self.MANI_BETA:
self.FRAGILE_FEATURE_FOUND = True
self.ALPHA = max(self.BETA, key=self.BETA.get)
self.LEAST_FRAGILE_FEATURE = self.ALPHA
print("Among fragile features, the least fragile feature is:\n"
% self.LEAST_FRAGILE_FEATURE)
def player1(self, image, partition_idx=None):
# Alpha
if partition_idx is None:
for partitionID in self.PARTITIONS.keys():
self.MANI_BETA = {}
self.MANI_DIST = {}
self.CURRENT_MANI = ()
print("partition ID:", partitionID)
self.player2(image, partitionID)
else:
self.player2(image, partition_idx)
def player2(self, image, partition_idx):
# Beta
pixels = self.PARTITIONS[partition_idx]
if not self.MANI_DIST:
self.target_pixels(image, pixels)
self.player1(image, partition_idx=partition_idx)
else:
min_dist = min(self.MANI_BETA.values())
print("Current min distance:", min_dist)
if min_dist is not inf:
print("Adversary found.")
adv_mani = min(self.MANI_BETA, key=self.MANI_BETA.get)
print("Manipulations:", adv_mani)
adv_dist = self.MANI_BETA[adv_mani]
self.BETA.update({partition_idx: [adv_mani, adv_dist]})
elif min(self.MANI_DIST.values()) >= self.DIST_VAL:
print("Adversarial distance exceeds distance bound.")
self.BETA.update({partition_idx: None})
else:
print("Adversary not found.")
mani_distance = copy.deepcopy(self.MANI_BETA)
for atom, _ in mani_distance.items():
self.MANI_BETA.pop(atom)
self.CURRENT_MANI = atom
self.MANI_DIST.pop(atom)
new_image = copy.deepcopy(self.IMAGE)
atomic_list = [atom[i:i + 4] for i in range(0, len(atom), 4)]
for atomic in atomic_list:
valid, new_image = self.apply_atomic_manipulation(new_image, atomic)
self.target_pixels(new_image, pixels)
self.player1(image, partition_idx)
"""