-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathcentral_adv_train.py
143 lines (117 loc) · 5.11 KB
/
central_adv_train.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
import os
os.environ["CUDA_VISIBLE_DEVICES"] = '0'
import torch
import torch.nn.functional as F
from torch.autograd import Variable
from torch.utils.data import DataLoader
from utils.data_provider import *
from utils.model_loader import *
from utils.hamming_matching import *
from model.backbone import *
from model.hashing_loss import *
def adv_loss(noisy_output, target_hash):
loss = torch.mean(noisy_output * target_hash)
return loss
def hash_adv(model, query, target_hash, epsilon, step=2, iteration=7, randomize=True):
delta = torch.zeros_like(query).cuda()
if randomize:
delta.uniform_(-epsilon, epsilon)
delta.data = (query.data + delta.data).clamp(0, 1) - query.data
delta.requires_grad = True
for i in range(iteration):
noisy_output = model(query + delta)
loss = adv_loss(noisy_output, target_hash.detach())
loss.backward()
delta.data = delta - step/255 * torch.sign(delta.grad.detach())
delta.data = delta.data.clamp(-epsilon, epsilon)
delta.data = (query.data + delta.data).clamp(0, 1) - query.data
delta.grad.zero_()
return query + delta.detach()
def hash_center_code(y, B, L, bit):
code = torch.zeros(y.size(0), bit).cuda()
for i in range(y.size(0)):
l = y[i].repeat(L.size(0), 1)
w = torch.sum(l*L, dim=1) / torch.sum(torch.sign(l + L), dim=1)
w1 = w.repeat(bit, 1).t()
w2 = 1 - torch.sign(w1)
c = w2.sum()/bit
w1 = 1 - w2
code[i] = torch.sign(torch.sum(c*w1*B-(L.size(0)-c)*w2*B, dim=0))
return code
classes_dic = {'FLICKR-25K': 38, 'NUS-WIDE':21, 'MS-COCO': 80, 'ImageNet': 100, 'CIFAR-10': 10}
dataset = 'NUS-WIDE'
DATA_DIR = '../data/{}'.format(dataset)
DATABASE_FILE = 'database_img.txt'
TRAIN_FILE = 'train_img.txt'
TEST_FILE = 'test_img.txt'
DATABASE_LABEL = 'database_label.txt'
TRAIN_LABEL = 'train_label.txt'
TEST_LABEL = 'test_label.txt'
num_classes = classes_dic[dataset]
model_name = 'DPSH'
backbone = 'VGG11'
batch_size = 32
bit = 32
epsilon = 8 / 255.0
epochs = 30
iteration = 7
lr = 1e-2
momentum=0.9
weight_decay = 5e-4
dset_database = HashingDataset(DATA_DIR, DATABASE_FILE, DATABASE_LABEL)
dset_train = HashingDataset(DATA_DIR, TRAIN_FILE, TRAIN_LABEL, training=True)
dset_test = HashingDataset(DATA_DIR, TEST_FILE, TEST_LABEL)
database_loader = DataLoader(dset_database, batch_size=batch_size, shuffle=False, num_workers=4)
train_loader = DataLoader(dset_train, batch_size=batch_size, shuffle=True, num_workers=4)
test_loader = DataLoader(dset_test, batch_size=batch_size, shuffle=False, num_workers=4)
num_database, num_train, num_test = len(dset_database), len(dset_train), len(dset_test)
database_labels = load_label(DATA_DIR, DATABASE_LABEL, True)
train_labels = load_label(DATA_DIR, TRAIN_LABEL, True).cuda()
test_labels = load_label(DATA_DIR, TEST_LABEL, True)
target_labels = database_labels.unique(dim=0)
model_path = 'checkpoint/Hash_{}_{}_{}_{}.pth'.format(dataset, model_name, backbone, bit)
robust_model_path = 'checkpoint/DHCAT_{}_{}_{}_{}.pth'.format(dataset, model_name, backbone, bit)
print(robust_model_path)
model = load_model(backbone, bit, model_path)
opt = torch.optim.SGD(model.parameters(), lr=lr, momentum=momentum, weight_decay=weight_decay)
lr_steps = epochs * len(train_loader)
scheduler = torch.optim.lr_scheduler.MultiStepLR(opt, milestones=[lr_steps / 2, lr_steps * 3 / 4], gamma=0.1)
train_B, train_L = generate_code_label(model, train_loader, num_train, bit, num_classes)
U_ben = torch.zeros(num_train, bit).cuda()
U_ben.data = train_B.data
U_adv = torch.zeros(num_train, bit).cuda()
B_ben = torch.zeros(num_train, bit).cuda()
B_ben.data = train_B.data
hash_criterion = get_hashing_loss(model_name, num_train, bit, num_classes, batch_size)
hash_criterion.preinstall(U_ben, train_L)
# adversarial training
for epoch in range(epochs):
if 'DSDH' in model_name:
hash_criterion.update_band_W()
epoch_loss = 0
for it, data in enumerate(train_loader):
x, y, index = data
x = x.cuda()
y = y.cuda()
batch_size_ = index.size(0)
center_codes_curr = hash_center_code(y, B_ben, train_L, bit)
x_adv = hash_adv(model, x, center_codes_curr, epsilon, step=2, iteration=iteration, randomize=True)
x_adv = x_adv.detach()
model.zero_grad()
output_adv = model(x_adv)
output_ben = model(x)
B_ben[index, :] = torch.sign(output_ben.data)
U_ben[index, :] = output_ben.data
loss_hash_ben = hash_criterion(output_ben, y, index)
center_codes = hash_center_code(y, B_ben, train_L, bit)
loss_hash_adv = -torch.mean((output_adv * center_codes))
loss = loss_hash_ben + loss_hash_adv
loss.backward()
opt.step()
scheduler.step()
epoch_loss += loss.item()
if it % 50 == 0:
print('epoch: {:2d}, step: {:3d}, lr: {:.5f}, loss: {:.5f}, loss_ben: {:.5f}, loss_adv: {:.5f}'.format(
epoch, it, scheduler.get_last_lr()[0], loss, loss_hash_ben, loss_hash_adv))
print('Epoch: %3d/%3d\tTrain_loss: %3.5f \n' % (epoch, epochs, epoch_loss / len(train_loader)))
save_model(model, robust_model_path)