-
Notifications
You must be signed in to change notification settings - Fork 25
/
Train_comdefend_Torch.py
330 lines (279 loc) · 11 KB
/
Train_comdefend_Torch.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
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
import os, time
import argparse
import itertools
import pickle
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
from torch.autograd import Variable
import shutil
import matplotlib
from PIL import Image
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import numpy as np
import math
import torch._utils
# import celeba_data
try:
torch._utils._rebuild_tensor_v2
except AttributeError:
def _rebuild_tensor_v2(storage, storage_offset, size, stride, requires_grad, backward_hooks):
tensor = torch._utils._rebuild_tensor(storage, storage_offset, size, stride)
tensor.requires_grad = requires_grad
tensor._backward_hooks = backward_hooks
return tensor
torch._utils._rebuild_tensor_v2 = _rebuild_tensor_v2
from pdb import set_trace as st
import models.cifar as models
def save_checkpoint(state_dict, is_best, filepath):
torch.save(state_dict, filepath)
if is_best:
shutil.copyfile(filepath, filepath.split('.pth')[0] + '_best.pth')
def save_image(save_path, tensor, ori_tensor):
img = tensor.data.cpu().numpy()
img = img.transpose(0, 2, 3, 1) * 255.0
img = np.array(img).astype(np.uint8)
img = np.concatenate(img, 1)
ori_img = ori_tensor.data.cpu().numpy()
ori_img = ori_img.transpose(0, 2, 3, 1) * 255.0
ori_img = np.array(ori_img).astype(np.uint8)
ori_img = np.concatenate(ori_img, 1)
vis = np.concatenate(np.array([ori_img, img]), 0)
img_pil = Image.fromarray(vis)
# img_pil = img_pil.resize((w // 16, h // 16))
img_pil.save(save_path)
class encoder(nn.Module):
# initializers
def __init__(self, d=128):
super(encoder, self).__init__()
self.features = nn.Sequential(
nn.Conv2d(3, 16, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(16, 32, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(32, 64, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(64, 128, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(128, 256, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(256, 128, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(128, 64, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(64, 32, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(32, 12, kernel_size=3, padding=1)
)
# Weight initialization
# for m in self.modules():
# if isinstance(m, nn.Conv2d):
# n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
# m.weight.data.normal_(0, math.sqrt(2. / n))
# elif isinstance(m, nn.BatchNorm2d):
# m.weight.data.fill_(1)
def forward(self, x):
return self.features(x)
class decoder(nn.Module):
# initializers
def __init__(self, d=128):
super(decoder, self).__init__()
self.features = nn.Sequential(
nn.Conv2d(12, 32, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(32, 64, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(64, 128, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(128, 256, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(256, 128, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(128, 64, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(64, 32, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(32, 16, kernel_size=3, padding=1),
nn.ELU(inplace=True),
nn.Conv2d(16, 3, kernel_size=3, padding=1),
nn.Sigmoid()
)
# # Weight initialization
# for m in self.modules():
# if isinstance(m, nn.Conv2d):
# n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
# m.weight.data.normal_(0, math.sqrt(2. / n))
# elif isinstance(m, nn.BatchNorm2d):
# m.weight.data.fill_(1)
def forward(self, x):
return self.features(x)
def train(args, trainloader, enc, dec, optimizer):
global mean, std
enc.train()
dec.train()
acc = 0
total = 0
total_loss = 0
count = 0
flag = False
for x, y in trainloader:
x, y = Variable(x.cuda()), Variable(y.cuda())
# x = torch.clamp(x, min=-1, max=1)
# st()
linear_code = enc(x)
noisy_code = linear_code - torch.randn(linear_code.size()).cuda() * args.std
binary_code = torch.sigmoid(noisy_code)
recons_x = dec(binary_code)
loss = ((recons_x - x) ** 2).mean() + (binary_code ** 2).mean() * 0.0001
# linear_code = enc(x)
# binary_code = torch.sigmoid(linear_code)
# recons_x = dec(binary_code)
# loss = ((recons_x - x)**2).mean()
if not flag:
flag = True
img_tensor = x
ori_img_tensor = recons_x
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_loss += loss.item()
if count % args.interval == 0:
print("loss: {}".format(loss.item()))
count += 1
print("training averag loss: ", total_loss / float(count))
return img_tensor, ori_img_tensor
def test(args, testloader, enc, dec, net):
global mean, std
enc.eval()
dec.eval()
acc = 0
total = 0
total_loss = 0
count = 0
flag = False
img_tensor = False
ori_img_tensor = False
for x, y in testloader:
x = Variable(x.cuda())
y = Variable(y.cuda())
linear_code = enc(x)
noisy_code = linear_code - torch.randn(linear_code.size()).cuda() * args.std
binary_code = torch.round(torch.sigmoid(noisy_code))
# binary_code
# z = x +x.sign().detach() - x.detach() differntiable version
recons_x = dec(binary_code)
if not flag:
flag = True
img_tensor = x
ori_img_tensor = recons_x
loss = ((recons_x - x) ** 2).mean()
total_loss += loss.item()
logits = net((recons_x - mean) / std)
_, pred = torch.max(logits, dim=1)
acc += (pred == y).sum().item()
total += y.size()[0]
count += 1
acc = acc / float(total)
avg_loss = total_loss / float(count)
print("test averag loss: ", avg_loss, " test acc: ", acc)
return acc, avg_loss, img_tensor, ori_img_tensor
def test_network_acc(testloader, net):
global mean, std
net.eval()
acc = 0.0
total = 0.0
for x, y in testloader:
x = x.cuda()
y = y.cuda()
logits = net((x - mean) / std)
_, pred = torch.max(logits, dim=1)
acc += (pred == y).sum().item()
total += y.size()[0]
print("accuracy: ", acc / float(total))
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='PyTorch comdefend Training')
parser.add_argument('--lr', default=1e-2, type=float)
parser.add_argument('--max_epochs', default=10, type=int)
parser.add_argument('--interval', default=100, type=int)
parser.add_argument('--std', default=20.0, type=float)
parser.add_argument('--gpu', default="0", type=str)
parser.add_argument('--batch_size', default=128, type=int)
parser.add_argument('--dataset', default='svhn', type=str)
parser.add_argument('--seed', default=100, type=int)
parser.add_argument('--sigma', default=2, type=float)
parser.add_argument('--workers', default=4, type=int)
args = parser.parse_args()
torch.manual_seed(args.seed)
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
transform = transforms.Compose([
# transforms.Scale(32),
transforms.ToTensor()
# transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5))
])
global mean, std
mean = torch.Tensor([0.5, 0.5, 0.5]).cuda()
std = torch.Tensor([0.5, 0.5, 0.5]).cuda()
mean = mean.view(1, 3, 1, 1)
std = std.view(1, 3, 1, 1)
# if args.dataset == 'cifar':
# trainloader = torch.utils.data.DataLoader(
# datasets.CIFAR10('./data/cifar/', train=True, transform=transform, download=True),
# batch_size=args.batch_size, shuffle=True)
# testloader = torch.utils.data.DataLoader(
# datasets.CIFAR10('./data/cifar/', train=False, transform=transform, download=True),
# batch_size=args.batch_size, shuffle=False)
# net = models.__dict__['resnet'](num_classes=10,depth=50)
# checkpoint_file = "models_chaowei/svhn_checkpoint.pth"
# checkpoint = torch.load( "models_chaowei/svhn_checkpoint.pth")
# net.load_state_dict(checkpoint['model'])
# print("svhn")
if args.dataset == 'svhn':
trainloader = torch.utils.data.DataLoader(
datasets.SVHN('./data/SVHN/', split='train', transform=transform, download=True),
batch_size=args.batch_size, shuffle=True)
testloader = torch.utils.data.DataLoader(
datasets.SVHN('./data/SVHN/', split='test', transform=transform, download=True),
batch_size=args.batch_size, shuffle=False)
net = models.__dict__['resnet'](num_classes=10, depth=50)
checkpoint_file = "save/svhn_checkpoint.pth"
checkpoint = torch.load(checkpoint_file)
net.load_state_dict(checkpoint['model'])
print("svhn")
else:
raise NotImplementedError("should implement this!")
enc = encoder()
dec = decoder()
# generator.cuda()
enc = enc.cuda()
dec = dec.cuda()
net = net.cuda()
test_network_acc(testloader, net)
params = list(enc.parameters()) + list(dec.parameters())
optimizer = optim.Adam(params, lr=args.lr, betas=(0.5, 0.999))
best_acc = 0
best_loss = 1e10
isbest = False
for epoch in range(args.max_epochs):
train_img, train_ori = train(args, trainloader, enc, dec, optimizer)
test_acc, test_loss, test_img, test_ori = test(args, testloader, enc, dec, net)
state_dict = {"enc": enc.state_dict(), "dec": dec.state_dict(), "optimizer": optimizer, "test_acc": test_acc,
"args": args}
filepath = "models_comdefend/{}_comdefend.pth".format(args.dataset)
img_path = "models_comdefend/{}/train_{}.jpg".format(args.dataset, epoch)
path = "/".join(img_path.split('/')[:-1])
if not os.path.exists(path):
os.makedirs(path)
save_image(img_path, train_img, train_ori)
img_path = "models_comdefend/{}/test_{}.jpg".format(args.dataset, epoch)
save_image(img_path, test_img, test_ori)
# if best_acc < test_acc:
if best_loss > test_loss:
best_acc = test_acc
best_loss = test_loss
isbest = True
save_checkpoint(state_dict, isbest, filepath)
print("best acc: {}, loss: {}".format(best_acc, best_loss))
isbest = False