-
Notifications
You must be signed in to change notification settings - Fork 4
/
test_Tiny_imagenet.py
223 lines (183 loc) · 9.03 KB
/
test_Tiny_imagenet.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
from autoattack import AutoAttack
from models import *
from utils import *
import argparse
import sys
import os
sys.path.insert(0, '..')
from utils02 import (upper_limit, lower_limit, std, clamp, get_loaders, ImageNet_get_loaders,New_ImageNet_get_loaders_64,
evaluate_pgd, evaluate_standard)
m_sample_list = []
all_block_count = 0
block_count = 0
class PreActBlock(nn.Module):
'''Pre-activation version of the BasicBlock.'''
expansion = 1
def __init__(self,prob, in_planes, planes, stride=1):
super(PreActBlock, self).__init__()
self.bn1 = nn.BatchNorm2d(in_planes)
self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
self.bn2 = nn.BatchNorm2d(planes)
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=1, padding=1, bias=False)
self.prob = prob
self.m = torch.distributions.bernoulli.Bernoulli(torch.Tensor([self.prob]))
self.shortcut = nn.Sequential()
if stride != 1 or in_planes != self.expansion*planes:
self.shortcut = nn.Sequential(
nn.Conv2d(in_planes, self.expansion*planes, kernel_size=1, stride=stride, bias=False)
)
def forward(self, x):
identity = x.clone()
global m_sample_list
global all_block_count
if self.training:
if len(m_sample_list) != 8:
all_block_count = all_block_count + 1
m_sample = self.m.sample()
m_sample_list.append(m_sample)
if torch.equal(m_sample, torch.ones(1)):
print("******************")
self.conv1.weight.requires_grad = True
self.conv2.weight.requires_grad = True
out = F.relu(self.bn1(x))
shortcut = self.shortcut(out) if hasattr(self, 'shortcut') else x
out = self.conv1(out)
out = self.conv2(F.relu(self.bn2(out)))
out += shortcut
else:
print("!!!!!!!!!!!!!!!!!!")
self.conv1.weight.requires_grad = False
self.conv2.weight.requires_grad = False
out = F.relu(self.bn1(x))
out = self.shortcut(out) if hasattr(self, 'shortcut') else x
return out
else:
if torch.equal(m_sample_list[8 - all_block_count], torch.ones(1)):
all_block_count = all_block_count - 1
print("******************")
self.conv1.weight.requires_grad = True
self.conv2.weight.requires_grad = True
out = F.relu(self.bn1(x))
shortcut = self.shortcut(out) if hasattr(self, 'shortcut') else x
out = self.conv1(out)
out = self.conv2(F.relu(self.bn2(out)))
out += shortcut
else:
print("!!!!!!!!!!!!!!!!!!")
all_block_count = all_block_count - 1
self.conv1.weight.requires_grad = False
self.conv2.weight.requires_grad = False
out = F.relu(self.bn1(x))
out = self.shortcut(out) if hasattr(self, 'shortcut') else x
return out
else:
out = F.relu(self.bn1(x))
shortcut = self.shortcut(out) if hasattr(self, 'shortcut') else x
out = self.conv1(out)
out = self.conv2(F.relu(self.bn2(out)))
out += shortcut
return out
class PreActBottleneck(nn.Module):
'''Pre-activation version of the original Bottleneck module.'''
expansion = 4
def __init__(self, in_planes, planes, stride=1):
super(PreActBottleneck, self).__init__()
self.bn1 = nn.BatchNorm2d(in_planes)
self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=1, bias=False)
self.bn2 = nn.BatchNorm2d(planes)
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
self.bn3 = nn.BatchNorm2d(planes)
self.conv3 = nn.Conv2d(planes, self.expansion*planes, kernel_size=1, bias=False)
if stride != 1 or in_planes != self.expansion*planes:
self.shortcut = nn.Sequential(
nn.Conv2d(in_planes, self.expansion*planes, kernel_size=1, stride=stride, bias=False)
)
def forward(self, x):
out = F.relu(self.bn1(x))
shortcut = self.shortcut(out) if hasattr(self, 'shortcut') else x
out = self.conv1(out)
out = self.conv2(F.relu(self.bn2(out)))
out = self.conv3(F.relu(self.bn3(out)))
out += shortcut
return out
class PreActResNet_sto(nn.Module):
def __init__(self, block,prob_0_L, num_blocks, num_classes=200):
super(PreActResNet_sto, self).__init__()
self.in_planes = 64
self.prob_now = prob_0_L[0]
self.prob_delta = prob_0_L[0] - prob_0_L[1]
self.prob_step = self.prob_delta / (sum(num_blocks) - 1)
self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False)
self.layer1 = self._make_layer(block, 64, num_blocks[0], stride=1)
self.layer2 = self._make_layer(block, 128, num_blocks[1], stride=2)
self.layer3 = self._make_layer(block, 256, num_blocks[2], stride=2)
self.layer4 = self._make_layer(block, 512, num_blocks[3], stride=2)
self.linear = nn.Linear(512*block.expansion*4, num_classes)
def _make_layer(self, block, planes, num_blocks, stride):
strides = [stride] + [1]*(num_blocks-1)
layers = []
for stride in strides:
layers.append(block(self.prob_now,self.in_planes, planes, stride))
self.in_planes = planes * block.expansion
self.prob_now = self.prob_now - self.prob_step
return nn.Sequential(*layers)
def forward(self, x):
out = self.conv1(x)
out = self.layer1(out)
out = self.layer2(out)
out = self.layer3(out)
out = self.layer4(out)
out = F.avg_pool2d(out, 4)
out = out.view(out.size(0), -1)
out = self.linear(out)
return out
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument('--data_dir', type=str, default='/apdcephfs/share_1290939/jiaxiaojun/tiny-imagenet-200')
parser.add_argument('--norm', type=str, default='Linf')
parser.add_argument('--epsilon', type=int, default=8)
parser.add_argument('--model_path', type=str, default='/apdcephfs/share_1290939/jiaxiaojun/FGSM_SD_AT/Tiny_Imagenet/FGSM_DSD/epochs_110/lr_max_0.1/model_PreActResNest18/lr_schedule_multistep/alpha_10.0/prob_0_1.0/prob_1_0.5/factor_0.6/best_model.pth')
parser.add_argument('--model', default='PreActResNest18', type=str, help='model name')
parser.add_argument('--n_ex', type=int, default=10000)
parser.add_argument('--batch_size', type=int, default=500)
parser.add_argument('--out_dir', type=str, default='./data')
arguments = parser.parse_args()
return arguments
args = get_args()
if not os.path.exists(args.out_dir):
os.makedirs(args.out_dir)
logfile1 = os.path.join(args.out_dir, 'log_file1.txt')
if os.path.exists(logfile1):
os.remove(logfile1)
device = 'cuda' if torch.cuda.is_available() else 'cpu'
target_model=PreActResNet_sto(PreActBlock, prob_0_L=[1, 1], num_blocks = [2,2,2,2])
target_model = target_model.to(device)
checkpoint = torch.load(args.model_path)
target_model.load_state_dict(checkpoint)
target_model.eval()
train_loader, test_loader = New_ImageNet_get_loaders_64(args.data_dir, args.batch_size)
epsilon = args.epsilon
epsilon = float(epsilon) / 255.
print(epsilon)
AT_fgsm_loss,AT_fgsm_acc=evaluate_fgsm(test_loader, target_model, 1)
AT_pgd_loss_10, AT_pgd_acc_10 = evaluate_pgd(test_loader, target_model, 10, 1, epsilon / std)
AT_pgd_loss_20, AT_pgd_acc_20 = evaluate_pgd(test_loader, target_model, 20, 1, epsilon / std)
AT_pgd_loss_50, AT_pgd_acc_50 = evaluate_pgd(test_loader, target_model, 50, 1, epsilon / std)
AT_CW_loss_20, AT_pgd_cw_acc_20 = evaluate_pgd_cw(test_loader, target_model, 20, 1)
AT_models_test_loss, AT_models_test_acc = evaluate_standard(test_loader, target_model)
print('AT_models_test_acc:', AT_models_test_acc)
print('AT_fgsm_acc:', AT_fgsm_acc)
print('AT_pgd_acc_10:', AT_pgd_acc_10)
print('AT_pgd_acc_20:', AT_pgd_acc_20)
print('AT_pgd_acc_50:', AT_pgd_acc_50)
print('AT_pgd_cw_acc_20:', AT_pgd_cw_acc_20)
adversary1 = AutoAttack(target_model, norm=args.norm, eps=epsilon, version='standard', log_path=logfile1)
#adversary2 = AutoAttack(target_model, norm=args.norm, eps=epsilon, version='standard', log_path=logfile2)
l = [x for (x, y) in test_loader]
x_test = torch.cat(l, 0)
l = [y for (x, y) in test_loader]
y_test = torch.cat(l, 0)
adv_complete = adversary1.run_standard_evaluation(x_test[:args.n_ex], y_test[:args.n_ex],
bs=args.batch_size)
# adv_complete1 = adversary2.run_standard_evaluation_individual(x_test[:args.n_ex], y_test[:args.n_ex],
# bs=args.batch_size)