-
Notifications
You must be signed in to change notification settings - Fork 16
/
attack.py
197 lines (164 loc) · 7.26 KB
/
attack.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
import argparse
import torchvision.models as models
import os
import json
import DataLoader
from utils import *
from FCN import *
from Normalize import Normalize, Permute
from imagenet_model.Resnet import resnet152_denoise, resnet101_denoise
def EmbedBA(function, encoder, decoder, image, label, config, latent=None):
device = image.device
if latent is None:
latent = encoder(image.unsqueeze(0)).squeeze().view(-1)
momentum = torch.zeros_like(latent)
dimension = len(latent)
noise = torch.empty((dimension, config['sample_size']), device=device)
origin_image = image.clone()
last_loss = []
lr = config['lr']
for iter in range(config['num_iters']+1):
perturbation = torch.clamp(decoder(latent.unsqueeze(0)).squeeze(0)*config['epsilon'], -config['epsilon'], config['epsilon'])
logit, loss = function(torch.clamp(image+perturbation, 0, 1), label)
if config['target']:
success = torch.argmax(logit, dim=1) == label
else:
success = torch.argmax(logit, dim=1) !=label
last_loss.append(loss.item())
if function.current_counts > 50000:
break
if bool(success.item()):
return True, torch.clamp(image+perturbation, 0, 1)
nn.init.normal_(noise)
noise[:, config['sample_size']//2:] = -noise[:, :config['sample_size']//2]
latents = latent.repeat(config['sample_size'], 1) + noise.transpose(0, 1)*config['sigma']
perturbations = torch.clamp(decoder(latents)*config['epsilon'], -config['epsilon'], config['epsilon'])
_, losses = function(torch.clamp(image.expand_as(perturbations) + perturbations, 0, 1), label)
grad = torch.mean(losses.expand_as(noise) * noise, dim=1)
if iter % config['log_interval'] == 0 and config['print_log']:
print("iteration: {} loss: {}, l2_deviation {}".format(iter, float(loss.item()), float(torch.norm(perturbation))))
momentum = config['momentum'] * momentum + (1-config['momentum'])*grad
latent = latent - lr * momentum
last_loss = last_loss[-config['plateau_length']:]
if (last_loss[-1] > last_loss[0]+config['plateau_overhead'] or last_loss[-1] > last_loss[0] and last_loss[-1]<0.6) and len(last_loss) == config['plateau_length']:
if lr > config['lr_min']:
lr = max(lr / config['lr_decay'], config['lr_min'])
last_loss = []
return False, origin_image
parser = argparse.ArgumentParser()
parser.add_argument('--config', default='config.json', help='config file')
parser.add_argument('--device', default='cuda:0', help='Device for Attack')
parser.add_argument('--save_prefix', default=None, help='override save_prefix in config file')
parser.add_argument('--model_name', default=None)
args = parser.parse_args()
with open(args.config) as config_file:
state = json.load(config_file)
if args.save_prefix is not None:
state['save_prefix'] = args.save_prefix
if args.model_name is not None:
state['model_name'] = args.model_name
new_state = state.copy()
new_state['batch_size'] = 1
new_state['test_bs'] = 1
device = torch.device(args.device if torch.cuda.is_available() else "cpu")
weight = torch.load(os.path.join("G_weight", state['generator_name']+".pytorch"), map_location=device)
encoder_weight = {}
decoder_weight = {}
for key, val in weight.items():
if key.startswith('0.'):
encoder_weight[key[2:]] = val
elif key.startswith('1.'):
decoder_weight[key[2:]] = val
_, dataloader, nlabels, mean, std = DataLoader.imagenet(new_state)
if 'OSP' in state:
if state['source_model_name'] == 'Adv_Denoise_Resnet152':
s_model = resnet152_denoise()
loaded_state_dict = torch.load(os.path.join('weight', state['source_model_name']+".pytorch"))
s_model.load_state_dict(loaded_state_dict)
if 'defense' in state and state['defense']:
source_model = nn.Sequential(
Normalize(mean, std),
Permute([2,1,0]),
s_model
)
else:
source_model = nn.Sequential(
Normalize(mean, std),
s_model
)
if state['model_name'] == 'Resnet34':
pretrained_model = models.resnet34(pretrained=True)
elif state['model_name'] == 'VGG19':
pretrained_model = models.vgg19_bn(pretrained=True)
elif state['model_name'] == 'Densenet121':
pretrained_model = models.densenet121(pretrained=True)
elif state['model_name'] == 'Mobilenet':
pretrained_model = models.mobilenet_v2(pretrained=True)
elif state['model_name'] == 'Adv_Denoise_Resnext101':
pretrained_model = resnet101_denoise()
loaded_state_dict = torch.load(os.path.join('weight', state['model_name']+".pytorch"))
pretrained_model.load_state_dict(loaded_state_dict, strict=True)
if 'defense' in state and state['defense']:
model = nn.Sequential(
Normalize(mean, std),
Permute([2,1,0]),
pretrained_model
)
else:
model = nn.Sequential(
Normalize(mean, std),
pretrained_model
)
encoder = Imagenet_Encoder()
decoder = Imagenet_Decoder()
encoder.load_state_dict(encoder_weight)
decoder.load_state_dict(decoder_weight)
model.to(device)
model.eval()
encoder.to(device)
encoder.eval()
decoder.to(device)
decoder.eval()
if 'OSP' in state:
source_model.to(device)
source_model.eval()
F = Function(model, state['batch_size'], state['margin'], nlabels, state['target'])
count_success = 0
count_total = 0
if not os.path.exists(state['save_path']):
os.mkdir(state['save_path'])
for i, (images, labels) in enumerate(dataloader):
images = images.to(device)
labels = int(labels)
logits = model(images)
correct = torch.argmax(logits, dim=1) == labels
if correct:
torch.cuda.empty_cache()
if state['target']:
labels = state['target_class']
if 'OSP' in state:
hinge_loss = MarginLoss_Single(state['white_box_margin'], state['target'])
images.requires_grad = True
latents = encoder(images)
for k in range(state['white_box_iters']):
perturbations = decoder(latents)*state['epsilon']
logits = source_model(torch.clamp(images+perturbations, 0, 1))
loss = hinge_loss(logits, labels)
grad = torch.autograd.grad(loss, latents)[0]
latents = latents - state['white_box_lr'] * grad
with torch.no_grad():
success, adv = EmbedBA(F, encoder, decoder, images[0], labels, state, latents.view(-1))
else:
with torch.no_grad():
success, adv = EmbedBA(F, encoder, decoder, images[0], labels, state)
count_success += int(success)
count_total += int(correct)
print("image: {} eval_count: {} success: {} average_count: {} success_rate: {}".format(i, F.current_counts, success, F.get_average(), float(count_success) / float(count_total)))
F.new_counter()
success_rate = float(count_success) / float(count_total)
if state['target']:
np.save(os.path.join(state['save_path'], '{}_class_{}.npy'.format(state['save_prefix'], state['target_class'])), np.array(F.counts))
else:
np.save(os.path.join(state['save_path'], '{}.npy'.format(state['save_prefix'])), np.array(F.counts))
print("success rate {}".format(success_rate))
print("average eval count {}".format(F.get_average()))