/
whitebox_test.py
170 lines (141 loc) · 6.4 KB
/
whitebox_test.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
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import numpy as np
import progress
from art.attacks import FastGradientMethod, BasicIterativeMethod, ProjectedGradientDescent
from art.attacks.saliency_map import SaliencyMapMethod
from art.classifiers import PyTorchClassifier
import cw
from config.dataset_config import getData
from networks.ensemble_resnet import avg_ensemble_3_resnet18, avg_ensemble_3_resnet18_fc
import argparse
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "3"
parser = argparse.ArgumentParser(description='PyTorch CNN Testing')
parser.add_argument('--method', type=str, default='pdd_deg', help='methods:[baseline adp deg pdd pdd_deg]')
parser.add_argument('--dataset', type=str, default='CIFAR100', help='datasets:[Tiny_Image FashionMNIST CIFAR100]')
parser.add_argument('--attack', type=str, default='PGD', help='attack methods:[FGSM PGD BIM JSMA CW]')
parser.add_argument('--norm', type=str, default='Linf', help='norm:[L2, Linf]')
parser.add_argument('--bs', type=int, default=20, help='batch size')
opt = parser.parse_args()
class LogNLLLoss(nn.Module):
def __init__(self, reduction='mean'):
super(LogNLLLoss, self).__init__()
assert reduction == 'mean'
self.reduction = reduction
def forward(self, x, targets):
log_x = torch.log(x)
log_nll_loss = nn.NLLLoss()(log_x, targets)
return log_nll_loss
if opt.dataset == 'FashionMNIST':
num_classes, train_data, test_data = getData('FashionMNIST')
if opt.dataset == 'CIFAR100':
num_classes, train_data, test_data = getData('CIFAR100')
if opt.dataset == 'Tiny_Image':
num_classes, train_data, test_data = getData('Tiny_Image')
testloader = torch.utils.data.DataLoader(
test_data,
batch_size=opt.bs,
shuffle=False,
num_workers=4,
pin_memory=True)
if opt.method == 'baseline':
model = avg_ensemble_3_resnet18(num_classes)
model_str = '/bs_ensemble_3_resnet18'
if opt.method == 'adp':
model = avg_ensemble_3_resnet18(num_classes)
model_str = '/adp_ensemble_3_resnet18'
if opt.method == 'deg':
model = avg_ensemble_3_resnet18(num_classes)
model_str = '/deg_ensemble_3_resnet18'
if opt.method == 'pdd':
model = avg_ensemble_3_resnet18_fc(num_classes)
model_str = '/pdd_ensemble_3_resnet18_fc'
if opt.method == 'pdd_deg':
model = avg_ensemble_3_resnet18_fc(num_classes)
model_str = '/pdd_deg_ensemble_3_resnet18_fc'
model_path = 'models/' + opt.dataset + model_str + '/best_model.pth'
model.load_state_dict(torch.load(model_path)['net'])
model.cuda()
model.eval()
criterion = LogNLLLoss()
classifier = PyTorchClassifier(model=model, loss=criterion, optimizer=None, clip_values=(0., 1.),input_shape=(1,3,32,32), nb_classes=num_classes)
def test_clean(testloader, model):
correct = 0
total = 0
for batch_idx, (inputs, targets) in enumerate(testloader):
if True:
inputs, targets = inputs.cuda(), targets.cuda()
outputs = model(inputs)
_, predicted = torch.max(outputs.data, 1)
correct += predicted.eq(targets.data).cpu().sum()
total += targets.size(0)
progress.progress_bar(
batch_idx,
len(testloader),
'clean_acc: %.3f%% (%d/%d)'
''%(100. *float(correct) /total,
correct,
total))
def test_robust(opt, model, classifier, attack_method, c, norm=None):
if opt.attack == 'FGSM':
adv_crafter = FastGradientMethod(classifier, norm=norm, eps=c, targeted=False, num_random_init=0, batch_size=opt.bs)
if opt.attack == 'PGD':
adv_crafter = ProjectedGradientDescent(classifier,norm=norm, eps=c, eps_step=c / 10., max_iter=10, targeted=False,num_random_init=1, batch_size=opt.bs)
if opt.attack == 'BIM':
adv_crafter = ProjectedGradientDescent(classifier, norm=norm, eps=c, eps_step=c / 10., max_iter=10, targeted=False,num_random_init=0, batch_size=bs)
if opt.attack == 'JSMA':
adv_crafter = SaliencyMapMethod(classifier, theta=0.1, gamma=c, batch_size=opt.bs)
if opt.attack == 'CW':
adv_crafter = cw.L2Adversary(targeted=False, confidence=0.01, c_range=(c, 1e10), max_steps=1000, abort_early=False,search_steps=5, box=(0.,1.0), optimizer_lr=0.01)
correct = 0
total = 0
total_sum = 0
common_id = []
for batch_idx, (inputs, targets) in enumerate(testloader):
inputs, targets = inputs.cuda(), targets.cuda()
output = classifier.predict(inputs.cpu().numpy(), batch_size=opt.bs)
output = torch.tensor(output)
output = output.cuda()
init_pred = output.max(1, keepdim=False)[1]
common_id = np.where(init_pred.cpu().numpy() == targets.cpu().numpy())[0]
if opt.attack == 'CW':
x_test_adv = adv_crafter(model, inputs, targets, to_numpy=True)
else:
x_test_adv = adv_crafter.generate(x=inputs.cpu().numpy())
perturbed_output = classifier.predict(x_test_adv)
perturbed_output = torch.tensor(perturbed_output)
perturbed_output = perturbed_output.cuda()
final_pred = perturbed_output.max(1, keepdim=False)[1]
total_sum += targets.size(0)
total += len(common_id)
correct += final_pred[common_id].eq(targets[common_id].data).cpu().sum()
attack_acc = 100. * float(correct) / total
progress.progress_bar(
batch_idx,
len(testloader),
'Attack Strength:%.3f, robust accuracy: %.3f%% (%d/%d)'
''%(c, attack_acc,
correct,
total))
if __name__ == '__main__':
print('Clean Accuracy:')
test_clean(testloader, model)
print('Attack: {}'.format(opt.attack))
if opt.attack == 'FGSM' or 'BIM' or 'PGD':
if opt.norm == 'Linf':
epsilons = [0.005, 0.01, 0.02, 0.04, 0.08, 0.16, 0.32, 0.64]
norm = np.inf
if opt.norm == 'L2':
epsilons = [0.1, 0.2, 0.4, 0.8, 1.6, 3.2, 6.4, 12.8]
norm = 2
if opt.attack == 'JSMA':
epsilons = [0.05, 0.1, 0.2, 0.4]
norm = None
if opt.attack == 'CW':
epsilons = [0.001, 0.01, 0.1]
norm = None
for eps in epsilons:
test_robust(opt, model, classifier, opt.attack, eps, norm=norm)