-
Notifications
You must be signed in to change notification settings - Fork 3
/
inversion.py
executable file
·177 lines (145 loc) · 6.99 KB
/
inversion.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
import numpy as np
import sys
import torch
class Inversion:
def __init__(self, model, asr_bound=0.9, batch_size=32, shape=(32, 32),
channel=3, clip_max=1.0, num_classes=10, preprocess=None,
device=None):
self.model = model
self.asr_bound = asr_bound
self.batch_size = batch_size
self.height, self.width = shape
self.channel = channel
self.clip_max = clip_max
self.num_classes = num_classes
self.preprocess = preprocess
self.device = torch.device('cuda') if device is None else device
self.epsilon = 1e-7
self.patience = 10
self.rate_up = 1.5
self.rate_down = 1.5 ** 1.5
self.mask_size = [self.height, self.width]
self.pattern_size = [self.channel, self.height, self.width]
def generate(self, pair, x_set, y_set, attack_size=100, steps=1000,
init_cost=1e-3, lr=0.1, init_m=None, init_p=None,
grad_m=True, grad_p=True):
self.model.eval()
source, target = pair
cost = init_cost
cost_up = 0
cost_down = 0
mask_best = torch.zeros(self.pattern_size).to(self.device)
pattern_best = torch.zeros(self.pattern_size).to(self.device)
reg_best = float('inf')
init_mask = np.random.random(self.mask_size) \
if init_m is None else init_m
init_pattern = np.random.random(self.pattern_size) * self.clip_max \
if init_p is None else init_p
init_mask = np.clip(init_mask, 0.0, 1.0)
init_mask = np.arctanh((init_mask - 0.5) * (2 - self.epsilon))
init_pattern = np.clip(init_pattern, 0.0, self.clip_max)
init_pattern = np.arctanh((init_pattern / self.clip_max - 0.5) * \
(2 - self.epsilon))
self.mask_tensor = torch.Tensor(init_mask).to(self.device)
self.pattern_tensor = torch.Tensor(init_pattern).to(self.device)
self.mask_tensor.requires_grad = grad_m
self.pattern_tensor.requires_grad = grad_p
if source < self.num_classes:
indices = np.where(y_set == source)[0]
else:
indices = np.where(y_set != target)[0]
if indices.shape[0] > attack_size:
indices = np.random.choice(indices, attack_size, replace=False)
else:
attack_size = indices.shape[0]
if attack_size < self.batch_size:
self.batch_size = attack_size
x_set = x_set[indices]
y_set = torch.full((x_set.shape[0],), target)
x_set = x_set.to(self.device)
y_set = y_set.to(self.device)
criterion = torch.nn.CrossEntropyLoss(reduction='none')
optimizer = torch.optim.Adam([self.mask_tensor, self.pattern_tensor],
lr=lr, betas=(0.5, 0.9))
index_base = np.arange(x_set.shape[0])
for step in range(steps):
indices = np.arange(x_set.shape[0])
np.random.shuffle(indices)
index_base = index_base[indices]
x_set = x_set[indices]
y_set = y_set[indices]
loss_ce_list = []
loss_reg_list = []
loss_list = []
acc_list = []
for idx in range(x_set.shape[0] // self.batch_size):
x_batch = x_set[idx*self.batch_size : (idx+1)*self.batch_size]
y_batch = y_set[idx*self.batch_size : (idx+1)*self.batch_size]
self.mask = (torch.tanh(self.mask_tensor) / (2 - self.epsilon) \
+ 0.5).repeat(self.channel, 1, 1)
self.pattern = (torch.tanh(self.pattern_tensor) / (2 - \
self.epsilon) + 0.5)
x_adv = (1 - self.mask) * x_batch + self.mask * self.pattern
optimizer.zero_grad()
output = self.model(self.preprocess(x_adv))
pred = output.argmax(dim=1, keepdim=True)
acc = pred.eq(y_batch.view_as(pred)).sum().item() / x_batch.shape[0]
loss_ce = criterion(output, y_batch)
loss_reg = torch.sum(torch.abs(self.mask)) / self.channel
loss = loss_ce.mean() + loss_reg * cost
loss.backward()
optimizer.step()
loss_ce_list.extend(loss_ce.detach().cpu().numpy())
loss_reg_list.append(loss_reg.detach().cpu().numpy())
loss_list.append(loss.detach().cpu().numpy())
acc_list.append(acc)
avg_loss_ce = np.mean(loss_ce_list)
avg_loss_reg = np.mean(loss_reg_list)
avg_loss = np.mean(loss_list)
avg_acc = np.mean(acc_list)
if avg_acc >= self.asr_bound and avg_loss_reg < reg_best:
mask_best = self.mask
pattern_best = self.pattern
reg_best = avg_loss_reg
epsilon = 0.01
init_mask = mask_best[0, ...]
init_mask = init_mask + torch.distributions.Uniform(\
low=-epsilon, high=epsilon).sample(\
init_mask.shape).to(self.device)
init_mask = torch.clip(init_mask, 0.0, 1.0)
init_mask = torch.arctanh((init_mask - 0.5) * (2 - self.epsilon))
init_pattern = pattern_best + self.clip_max * \
torch.distributions.Uniform(low=-epsilon, \
high=epsilon).sample(init_pattern.shape)\
.to(self.device)
init_pattern = torch.clip(init_pattern, 0.0, self.clip_max)
init_pattern = torch.arctanh((init_pattern / self.clip_max - 0.5) \
* (2 - self.epsilon))
with torch.no_grad():
self.mask_tensor.copy_(init_mask)
self.pattern_tensor.copy_(init_pattern)
if avg_acc >= self.asr_bound:
cost_up += 1
cost_down = 0
else:
cost_up = 0
cost_down += 1
if cost_up >= self.patience:
cost_up = 0
if cost == 0:
cost = init_cost
else:
cost *= self.rate_up
elif cost_down >= self.patience:
cost_down = 0
cost /= self.rate_down
if step % 10 == 0:
sys.stdout.write('\rstep: {:3d}, asr: {:.2f}, loss: {:.4f}, '\
.format(step, avg_acc, avg_loss) + \
'ce: {:.4f}, reg: {:.4f}, reg_best: {:.4f}'\
.format(avg_loss_ce, avg_loss_reg, reg_best))
sys.stdout.flush()
sys.stdout.write('\nmask for {:d}-{:d}: {:f}\n'.format(source, target, \
mask_best.abs().sum()))
sys.stdout.flush()
return mask_best, pattern_best