-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathnotebook.py
277 lines (229 loc) · 12.4 KB
/
notebook.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
"""
Implementation in PyTorch of the method and multiclass classification experiments
in CIFAR10, CIFAR100 and SVHN datasets extending the VGG16 and the Wide Residual Network
architecture, as described in the paper:
N. Gkalelis, V. Mezaris, "Subclass deep neural networks: re-enabling neglected classes
in deep network training for multimedia classification", Proc. 26th Int. Conf. on
Multimedia Modeling (MMM2020), Daejeon, Korea, Jan. 2020.
History
-------
DATE | DESCRIPTION | NAME | ORGANIZATION |
21/07/2019 | first creation | Nikolaos Gkalelis | CERTH-ITI |
"""
import argparse
import numpy as np
import json
from datetime import datetime
import pandas as pd
import torch
import torch.backends.cudnn as cudnn
from torch.optim.lr_scheduler import MultiStepLR
from torchvision import datasets, transforms
from util.cutout import Cutout
from util.subclassutils import partition_data
from util.trainutils import train_subclass_one_epoch, compute_neglected_classes
from util.testutils import test
from model.wide_resnet import WideResNet
from model.vgg import vgg16_bn
model_options = ['wideresnet', 'vgg16']
dataset_options = ['cifar10', 'cifar100', 'svhn']
parser = argparse.ArgumentParser(description='SCNN')
parser.add_argument('--dataset', default='cifar10',
choices=dataset_options)
parser.add_argument('--model', default='vgg16',
choices=model_options)
parser.add_argument('--batch_size', type=int, default=128,
help='input batch size for training')
parser.add_argument('--epochs_class', type=int, default=10,
help='number of epochs to train for identifying the neglected classes')
parser.add_argument('--epochs', type=int, default=300,
help='number of epochs to train the SDNN')
parser.add_argument('--learning_rate', type=float, default=0.1, # cifar10/100: 0.1; svhn: 0.01
help='learning rate')
parser.add_argument('--data_augmentation', action='store_true', default=True,
help='augment data by flipping and cropping') # cifar10/100: True; svhn: False
parser.add_argument('--cutout', action='store_true', default=False, help='apply cutout')
parser.add_argument('--subclass', action='store_true', default=False, help='apply subclasses')
# 1. means ignore misclassification cost of assigning the observation to another subclass of same class
# 0. means place full misclassification cost of assigning the observation to another subclass of same class,
# i.e., the other subclasses are treated as completely different classes
parser.add_argument('--subclass_label_weight', type=float, default=0.9, help='Weight in [0,1] to weight the subclass label'
'of the subclasses belonging to different classes in the subclass CE criterion')
parser.add_argument('--numClasses2Repartition', default=2, type=int,
help='classes to repartition')
parser.add_argument('--numSubclassesPerClass', default=2, type=int,
help='number of subclasses for each class we partition')
parser.add_argument('--n_holes', type=int, default=1,
help='number of holes to cut out from image')
parser.add_argument('--length', type=int, default=16,
help='length of the holes') # cifar10: 16; cifar100: 8; svhn: 20
parser.add_argument('--no-cuda', action='store_true', default=False,
help='enables CUDA training')
parser.add_argument('--seed', type=int, default=0,
help='random seed (default: 1)')
if __name__ == '__main__':
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
cudnn.benchmark = True # Should make training should go faster for large models
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
print(args)
# Normalization
if args.dataset == 'svhn':
normalize = transforms.Normalize(mean=[x / 255.0 for x in[109.9, 109.7, 113.8]],
std=[x / 255.0 for x in [50.1, 50.6, 50.8]])
else:
normalize = transforms.Normalize(mean=[x / 255.0 for x in [125.3, 123.0, 113.9]],
std=[x / 255.0 for x in [63.0, 62.1, 66.7]])
# transforms
train_transform = transforms.Compose([])
if args.data_augmentation:
train_transform.transforms.append(transforms.RandomCrop(32, padding=4))
train_transform.transforms.append(transforms.RandomHorizontalFlip())
train_transform.transforms.append(transforms.ToTensor())
train_transform.transforms.append(normalize)
# cutout
if args.cutout:
train_transform.transforms.append(Cutout(n_holes=args.n_holes, length=args.length))
test_transform = transforms.Compose([transforms.ToTensor(), normalize])
# dataset
if args.dataset == 'cifar10':
num_classes = 10
train_dataset = datasets.CIFAR10(root='data/',
train=True,
transform=train_transform,
download=True)
test_dataset = datasets.CIFAR10(root='data/',
train=False,
transform=test_transform,
download=True)
elif args.dataset == 'cifar100':
num_classes = 100
train_dataset = datasets.CIFAR100(root='data/',
train=True,
transform=train_transform,
download=True)
test_dataset = datasets.CIFAR100(root='data/',
train=False,
transform=test_transform,
download=True)
elif args.dataset == 'svhn':
num_classes = 10
train_dataset = datasets.SVHN(root='data/',
split='train',
transform=train_transform,
download=True)
extra_dataset = datasets.SVHN(root='data/',
split='extra',
transform=train_transform,
download=True)
# Combine training and extra datasets
data = np.concatenate([train_dataset.data, extra_dataset.data], axis=0)
labels = np.concatenate([train_dataset.labels, extra_dataset.labels], axis=0)
train_dataset.data = data
train_dataset.labels = labels
test_dataset = datasets.SVHN(root='data/',
split='test',
transform=test_transform,
download=True)
else:
raise NameError("Unexpected dataset name: " + args.dataset)
balancingSampler = None
trnLdrDoShuffling = True
############################
# identify neglected classes
############################
train_loader_class = torch.utils.data.DataLoader(dataset=train_dataset,
batch_size=args.batch_size,
shuffle=trnLdrDoShuffling,
pin_memory=True,
num_workers=2,
sampler=balancingSampler)
if args.model == 'wideresnet':
if args.dataset == 'svhn':
cnn_class = WideResNet(depth=16, num_classes=num_classes, widen_factor=8,
dropRate=0.4)
else:
cnn_class = WideResNet(depth=28, num_classes=num_classes, widen_factor=10,
dropRate=0.3)
elif args.model == 'vgg16':
cnn_class = vgg16_bn(numClass=num_classes)
else:
raise NameError("Unexpected model name: " + args.model)
cnn_class = cnn_class.cuda()
cnn_class_optimizer = torch.optim.SGD(cnn_class.parameters(), lr=args.learning_rate, momentum=0.9, nesterov=True,
weight_decay=5e-4)
scheduler_class = MultiStepLR(cnn_class_optimizer, milestones=[10, 20, 30], gamma=0.2)
cnn_optimizer_class = torch.optim.SGD(cnn_class.parameters(), lr=args.learning_rate, momentum=0.9, nesterov=True,
weight_decay=5e-4)
tti = compute_neglected_classes(train_loader_class, cnn_class, cnn_optimizer_class,
scheduler_class, args.epochs_class, num_classes)
args.class2repartition = torch.argsort(input=tti, descending=True)[0:args.numClasses2Repartition].tolist()
del cnn_class_optimizer, scheduler_class, cnn_class, train_loader_class
############
# train scnn
############
# augment neglected classes and subclass them
numSubclassesPerClass = [args.numSubclassesPerClass] * args.numClasses2Repartition
train_dataset, subclass2classIdx, classSubclasses = partition_data(args.dataset,
train_dataset,
args.class2repartition,
numSubclassesPerClass)
subclass2classIdx = subclass2classIdx.cuda()
num_subclasses = len(subclass2classIdx)
# data loaders
train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
batch_size=args.batch_size,
shuffle=trnLdrDoShuffling,
pin_memory=True,
num_workers=2,
sampler=balancingSampler)
test_loader = torch.utils.data.DataLoader(dataset=test_dataset,
batch_size=args.batch_size,
shuffle=False,
pin_memory=True,
num_workers=2)
# cnn architectures
if args.model == 'wideresnet':
if args.dataset == 'svhn':
cnn = WideResNet(depth=16, num_classes=num_subclasses, widen_factor=8,
dropRate=0.4)
else:
cnn = WideResNet(depth=28, num_classes=num_subclasses, widen_factor=10,
dropRate=0.3)
elif args.model == 'vgg16':
cnn = vgg16_bn(numClass=num_subclasses)
else:
raise NameError("Unexpected model name: " + args.model)
cnn = cnn.cuda()
cnn_optimizer = torch.optim.SGD(cnn.parameters(), lr=args.learning_rate, momentum=0.9, nesterov=True, weight_decay=5e-4)
if args.dataset == 'svhn':
scheduler = MultiStepLR(cnn_optimizer, milestones=[80, 120, 160, 220], gamma=0.1)
else:
scheduler = MultiStepLR(cnn_optimizer, milestones=[60, 120, 160, 220, 260], gamma=0.2)
test_id = args.dataset + '_' + args.model + '_' + datetime.now().strftime("%Y%m%d_%H%M%S")
log_filename = 'logs/' + test_id + '.csv'
best_model_filename = 'checkpoints/' + test_id + '.pt'
log = pd.DataFrame(index=[], columns=[
'epoch', 'lr', 'trn_acc', 'tst_acc'
])
# train the scnn
best_acc = 0.
for epoch in range(args.epochs):
scheduler.step(epoch)
train_acc = train_subclass_one_epoch(train_loader, cnn, cnn_optimizer, epoch, subclass2classIdx, classSubclasses, args.subclass_label_weight)
test_acc, test_time = test(test_loader, cnn, subclass2classIdx)
tmp = pd.Series([
epoch,
scheduler.get_lr()[0],
train_acc,
test_acc
], index=['epoch', 'lr', 'trn_acc', 'tst_acc'])
log = log.append(tmp, ignore_index=True)
log.to_csv(log_filename, index=False)
print('test_acc: {}, test_time: {}'.format(test_acc, test_time))
if test_acc > best_acc:
torch.save(cnn.state_dict(), best_model_filename)
best_acc = test_acc
print("=> saved best model")