-
Notifications
You must be signed in to change notification settings - Fork 4
/
GenAdvExample.py
128 lines (101 loc) · 4.91 KB
/
GenAdvExample.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
import os
import time
import argparse
from PIL import Image
import torch
from torch.autograd import Variable
from HyperTools import *
from Models import *
DataName = {1:'PaviaU',2:'Salinas'}
def main(args):
if args.dataID==1:
num_classes = 9
num_features = 103
save_pre_dir = './Data/PaviaU/'
elif args.dataID==2:
num_classes = 16
num_features = 204
save_pre_dir = './Data/Salinas/'
X = np.load(save_pre_dir+'X.npy')
_,h,w = X.shape
Y = np.load(save_pre_dir+'Y.npy')
X_train = np.reshape(X,(1,num_features,h,w))
train_array = np.load(save_pre_dir+'train_array.npy')
Y_train = np.ones(Y.shape)*255
Y_train[train_array] = Y[train_array]
Y_train = np.reshape(Y_train,(1,h,w))
# define the targeted label in the attack
Y_tar = np.zeros(Y.shape)
Y_tar = np.reshape(Y_tar,(1,h,w))
save_path_prefix = args.save_path_prefix+'Exp_'+DataName[args.dataID]+'/'
if os.path.exists(save_path_prefix)==False:
os.makedirs(save_path_prefix)
num_epochs = 1000
if args.model=='SACNet':
Model = SACNet(num_features=num_features,num_classes=num_classes)
elif args.model=='DilatedFCN ':
Model = DilatedFCN (num_features=num_features,num_classes=num_classes)
elif args.model=='SpeFCN':
Model = SpeFCN(num_features=num_features,num_classes=num_classes)
num_epochs = 3000
elif args.model=='SpaFCN':
Model = SpaFCN(num_features=num_features,num_classes=num_classes)
elif args.model=='SSFCN':
Model = SSFCN(num_features=num_features,num_classes=num_classes)
Model = torch.nn.DataParallel(Model).cuda()
Model.train()
optimizer = torch.optim.Adam(Model.parameters(),lr=args.lr,weight_decay=args.decay)
images = torch.from_numpy(X_train).float().cuda()
label = torch.from_numpy(Y_train).long().cuda()
criterion = CrossEntropy2d().cuda()
# train the classification model
for epoch in range(num_epochs):
adjust_learning_rate(optimizer,args.lr,epoch,num_epochs)
tem_time = time.time()
optimizer.zero_grad()
output = Model(images)
seg_loss = criterion(output,label)
seg_loss.backward()
optimizer.step()
batch_time = time.time()-tem_time
if (epoch+1) % 1 == 0:
print('epoch %d/%d: time: %.2f cls_loss = %.3f'%(epoch+1, num_epochs,batch_time,seg_loss.item()))
Model.eval()
output = Model(images)
_, predict_labels = torch.max(output, 1)
predict_labels = np.squeeze(predict_labels.detach().cpu().numpy()).reshape(-1)
# adversarial attack
epsilon = [0.01,0.02,0.04,0.06,0.08,0.1,0.2,0.4,0.6,0.8,1,2,4,6,8,10]
for i in range(len(epsilon)):
print('Generate adversarial example with epsilon = %.2f'%(epsilon[i]))
processed_image = Variable(images)
processed_image = processed_image.requires_grad_()
label = torch.from_numpy(Y_tar).long().cuda()
output = Model(processed_image)
seg_loss = criterion(output,label)
seg_loss.backward()
adv_noise = epsilon[i] * processed_image.grad.data / torch.norm(processed_image.grad.data,float("inf"))
processed_image.data = processed_image.data - adv_noise
X_adv = torch.clamp(processed_image, 0, 1).cpu().data.numpy()[0]
noise_image = X_adv - images.cpu().data.numpy()[0]
noise_image[noise_image > 1] = 1
noise_image[noise_image < 0] = 0
if args.dataID == 1:
im = Image.fromarray(np.moveaxis((noise_image[[102,56,31],:,:]*25500).astype('uint8'),0,-1))
im.save(save_path_prefix+'perturbation'+str(epsilon[i])+'.png','png')
im = Image.fromarray(np.moveaxis((X_adv[[102,56,31],:,:]*255).astype('uint8'),0,-1))
im.save(save_path_prefix+'advimage'+str(epsilon[i])+'.png','png')
elif args.dataID == 2:
im = Image.fromarray(np.moveaxis((noise_image[[102,56,31],:,:]*25500).astype('uint8'),0,-1))
im.save(save_path_prefix+'perturbation'+str(epsilon[i])+'.png','png')
im = Image.fromarray(np.moveaxis((X_adv[[102,56,31],:,:]*255).astype('uint8'),0,-1))
im.save(save_path_prefix+'advimage'+str(epsilon[i])+'.png','png')
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--dataID', type=int, default=1)
parser.add_argument('--save_path_prefix', type=str, default='./')
parser.add_argument('--model', type=str, default='SACNet')
# train
parser.add_argument('--lr', type=float, default=5e-4)
parser.add_argument('--decay', type=float, default=5e-5)
main(parser.parse_args())