-
Notifications
You must be signed in to change notification settings - Fork 2
/
main_std.py
310 lines (270 loc) · 14 KB
/
main_std.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
import os
import numpy as np
import copy
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import DataLoader, sampler
import torchvision as tv
from torchvision import models
from torch.autograd import Variable
from time import time
from attack import FastGradientSignUntargeted
from utils import makedirs, create_logger, tensor2cuda, numpy2cuda, evaluate, evaluate_, save_model
from argument import parser, print_args
import patch_dataset as patd
# os.environ["CUDA_VISIBLE_DEVICES"] = ','.join(map(str, [0,2]))
class AddGaussianNoise(object):
def __init__(self, mean=0., std=1.):
self.std = std
self.mean = mean
def __call__(self, tensor):
return tensor + torch.randn(tensor.size()) * self.std + self.mean
def __repr__(self):
return self.__class__.__name__ + '(mean={0}, std={1})'.format(self.mean, self.std)
class Trainer():
def __init__(self, args, logger, attack):
self.args = args
self.logger = logger
self.attack = attack
def train(self, model, tr_loader, va_loader=None, adv_train=False):
args = self.args
logger = self.logger
opt = torch.optim.Adam(model.parameters(), args.learning_rate, weight_decay=args.weight_decay)
scheduler = torch.optim.lr_scheduler.MultiStepLR(opt, milestones=[100, 200], gamma=0.1)
_iter = 0
begin_time = time()
best_loss = 999
for epoch in range(1, args.max_epoch+1):
for data, label in tr_loader:
# for sample in tr_loader:
# data, label = sample['buffers'], sample['labels']
# print('data shape is ', data.shape)
# print('label shape is ', label.shape)
data, label = tensor2cuda(data), tensor2cuda(label)
if adv_train:
# When training, the adversarial example is created from a random
# close point to the original data point. If in evaluation mode,
# just start from the original data point.
adv_data = self.attack.perturb(data, label, 'mean', True)
model.train()
output = model(adv_data)
else:
model.train()
output = model(data)
loss = F.binary_cross_entropy(torch.sigmoid(output), label)
opt.zero_grad()
loss.backward()
opt.step()
t = Variable(torch.Tensor([0.5]).cuda()) # threshold to compute accuracy
if _iter % args.n_eval_step == 0:
t1 = time()
if adv_train:
with torch.no_grad():
model.eval()
stand_output = model(data)
# pred = torch.max(stand_output, dim=1)[1] # this give us the indices tensor
pred = torch.sigmoid(stand_output)
out = (pred > t).float()
# print(pred.shape)
# print(out.shape)
# std_acc = evaluate(pred.cpu().numpy(), label.cpu().numpy()) * 100
stdacc_list = evaluate_(out.cpu().numpy(), label.cpu().numpy())
# print('std accuracy list shape: ', np.array(stdacc_list).shape)
# pred = torch.max(output, dim=1)[1]
pred = torch.sigmoid(output)
out = (pred > t).float()
# print(pred)
# adv_acc = evaluate(pred.cpu().numpy(), label.cpu().numpy()) * 100
advacc_list = evaluate_(out.cpu().numpy(), label.cpu().numpy())
# print('adv accuracy list shape: ', np.array(stdacc_list).shape)
else:
adv_data = self.attack.perturb(data, label, 'mean', False)
with torch.no_grad():
model.eval()
adv_output = model(adv_data)
# adv_output = model(adv_data, _eval=True)
# pred = torch.max(adv_output, dim=1)[1]
pred = torch.sigmoid(adv_output)
out = (pred > t).float()
# print(label)
# print(pred)
# adv_acc = evaluate(pred.cpu().numpy(), label.cpu().numpy()) * 100
advacc_list = evaluate_(out.cpu().numpy(), label.cpu().numpy())
# pred = torch.max(output, dim=1)[1]
pred = torch.sigmoid(output)
out = (pred > t).float()
# print(pred)
# std_acc = evaluate(pred.cpu().numpy(), label.cpu().numpy()) * 100
stdacc_list = evaluate_(out.cpu().numpy(), label.cpu().numpy())
t2 = time()
print('%.3f' % (t2 - t1))
logger.info('epoch: %d, iter: %d, spent %.2f s, tr_loss: %.3f' % (
epoch, _iter, time()-begin_time, loss.item()))
begin_time = time()
if _iter % args.n_checkpoint_step == 0:
file_name = os.path.join(args.model_folder, 'checkpoint_%d.pth' % _iter)
save_model(model, file_name)
_iter += 1
scheduler.step()
if va_loader is not None:
t1 = time()
va_acc, va_adv_acc, va_stdloss, va_advloss = self.test(model, va_loader, True, False)
va_acc, va_adv_acc = va_acc * 100.0, va_adv_acc * 100.0
if va_stdloss < best_loss:
best_loss = va_stdloss
file_name = os.path.join(args.model_folder, 'checkpoint_best.pth')
save_model(model, file_name)
t2 = time()
logger.info('\n'+'='*20 +' evaluation at epoch: %d iteration: %d '%(epoch, _iter) \
+'='*20)
logger.info('test acc: %.3f %%, test adv acc: %.3f %%, spent: %.3f' % (
va_acc, va_adv_acc, t2-t1))
logger.info('test loss: %.3f , test adv loss: %.3f , spent: %.3f' % (
va_stdloss, va_advloss, t2-t1))
logger.info('='*28+' end of evaluation '+'='*28+'\n')
def test(self, model, loader, adv_test=False,
use_pseudo_label=False, if_AUC=False):
# adv_test is False, return adv_acc as -1
total_acc = 0.0
num = 0
total_adv_acc = 0.0
total_stdloss = 0.0
total_advloss = 0.0
t = Variable(torch.Tensor([0.5]).cuda()) # threshold to compute accuracy
label_list = []
pred_list = []
predadv_list = []
with torch.no_grad():
for data, label in loader:
# for sample in loader:
# data, label = sample['buffers'], sample['labels']
data, label = tensor2cuda(data), tensor2cuda(label)
model.eval()
output = model(data)
std_loss = F.binary_cross_entropy(torch.sigmoid(output), label)
pred = torch.sigmoid(output)
out = (pred > t).float()
te_acc = np.mean(evaluate_(out.cpu().numpy(), label.cpu().numpy()))
total_acc += te_acc
total_stdloss += std_loss
if if_AUC:
label_list.append(label.cpu().numpy())
pred_list.append(pred.cpu().numpy())
# num += output.shape[0]
num += 1
if adv_test:
# use predicted label as target label
with torch.enable_grad():
adv_data = self.attack.perturb(data,
pred if use_pseudo_label else label,
'mean', False)
model.eval()
adv_output = model(adv_data)
adv_loss = F.binary_cross_entropy(torch.sigmoid(adv_output), label)
adv_pred = torch.sigmoid(adv_output)
if if_AUC:
predadv_list.append(adv_pred.cpu().numpy())
adv_out = (adv_pred > t).float()
adv_acc = np.mean(evaluate_(adv_out.cpu().numpy(), label.cpu().numpy()))
total_adv_acc += adv_acc
total_advloss += adv_loss
else:
total_adv_acc = -num
if if_AUC:
pred = np.squeeze(np.array(pred_list))
predadv = np.squeeze(np.array(predadv_list))
label = np.squeeze(np.array(label_list))
np.save(os.path.join(self.args.log_folder, 'y_pred.npy'), pred)
# np.save(os.path.join(self.args.log_folder, 'y_predadv_'+str(args.epsilon)+'.npy'), predadv)
np.save(os.path.join(self.args.log_folder, 'y_true.npy'), label)
# PRED_label = ['No Finding', 'Cardiomegaly', 'Edema',
# 'Consolidation', 'Pneumonia', 'Atelectasis',
# 'Pneumothorax', 'Pleural Effusion']
# PRED_label = ['healthy', 'partially injured', 'completely ruptured']
# PRED_label = ['malignancy']
# plot_AUC(pred, label, self.args.log_folder, 'auc.png', PRED_label)
# plot_AUC(predadv, label, self.args.log_folder, 'auc_'+str(args.epsilon)+'.png', PRED_label)
# np.save('predstd_'+str(args.epsilon)+'.npy', pred)
# np.save('predstdadv_'+str(args.epsilon)+'.npy', predadv)
# np.save('labelstd_'+str(args.epsilon)+'.npy', label)
else:
return total_acc / num, total_adv_acc / num, total_stdloss / num, total_advloss / num
def main(args):
save_folder = '%s_%s' % (args.dataset, args.affix)
log_folder = os.path.join(args.log_root, save_folder)
model_folder = os.path.join(args.model_root, save_folder)
makedirs(log_folder)
makedirs(model_folder)
setattr(args, 'log_folder', log_folder)
setattr(args, 'model_folder', model_folder)
logger = create_logger(log_folder, args.todo, 'info')
print_args(args, logger)
# model = WideResNet(depth=34, num_classes=10, widen_factor=10, dropRate=0.0)
model = models.resnet50(pretrained=args.pretrain)
num_classes=8
# model.classifier = nn.Linear(model.classifier.in_features, num_classes)
model.fc = nn.Linear(model.fc.in_features, num_classes)
attack = FastGradientSignUntargeted(model,
args.epsilon,
args.alpha,
min_val=0,
max_val=1,
max_iters=args.k,
_type=args.perturbation_type)
if torch.cuda.is_available():
model.cuda()
# model = nn.DataParallel(model).cuda()
trainer = Trainer(args, logger, attack)
if args.todo == 'train':
transform_train = tv.transforms.Compose([
tv.transforms.Resize(256),
tv.transforms.ToTensor(),
tv.transforms.Lambda(lambda x: F.pad(x.unsqueeze(0),
(4*6,4*6,4*6,4*6), mode='constant', value=0).squeeze()),
tv.transforms.ToPILImage(),
tv.transforms.RandomHorizontalFlip(),
tv.transforms.ColorJitter(brightness=0.3, contrast=0.3,
saturation=0.3, hue=0.3),
# tv.transforms.RandomRotation(25),
tv.transforms.RandomAffine(25, translate=(0.2, 0.2),
scale=(0.8,1.2), shear=10),
tv.transforms.RandomCrop(256),
tv.transforms.ToTensor(),
AddGaussianNoise(0.5, args.epsilon)
])
tr_dataset = patd.PatchDataset(path_to_images=args.data_root,
fold='train',
sample=args.subsample,
transform=transform_train)
tr_loader = DataLoader(tr_dataset, batch_size=args.batch_size, shuffle=True, num_workers=24)
# evaluation during training
transform_test = tv.transforms.Compose([
tv.transforms.Resize(256),
# tv.transforms.CenterCrop(224),
tv.transforms.ToTensor(),
# tv.transforms.Normalize(mean, std)
])
te_dataset = patd.PatchDataset(path_to_images=args.data_root,
fold='valid',
transform=transform_test)
te_loader = DataLoader(te_dataset, batch_size=args.batch_size, shuffle=False, num_workers=24)
trainer.train(model, tr_loader, te_loader, args.adv_train)
elif args.todo == 'test':
te_dataset = patd.PatchDataset(path_to_images=args.data_root,
fold='test',
transform=tv.transforms.Compose([
tv.transforms.Resize(256),
tv.transforms.ToTensor(),
]))
te_loader = DataLoader(te_dataset, batch_size=1, shuffle=False, num_workers=1)
checkpoint = torch.load(args.load_checkpoint)
model.load_state_dict(checkpoint)
std_acc, adv_acc = trainer.test(model, te_loader, adv_test=True, use_pseudo_label=False, if_AUC=True)
print("std acc: %.4f, adv_acc: %.4f" % (std_acc * 100, adv_acc * 100))
else:
raise NotImplementedError
if __name__ == '__main__':
args = parser()
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
main(args)