-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
346 lines (292 loc) · 10.6 KB
/
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
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
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
import torch
import os
import argparse
import shutil
import utils
import numpy as np
import random
import torchvision.transforms.functional as transforms_F
from torch import optim
from torch.utils.data import DataLoader
from PIL import Image
from torchvision import transforms, models
from datetime import datetime
from torchvision.datasets import ImageFolder
from losses import TripletLossHuman
from model import FTModel
from tqdm import tqdm
current_time = datetime.now().strftime("%d_%m_%Y-%H_%M")
parser = argparse.ArgumentParser(description='Material Similarity Training')
parser.add_argument('--train-dir',
metavar='DIR', help='path to dataset',
default='data/split_dataset')
parser.add_argument('--test-dir',
metavar='DIR', help='path to dataset',
default='data/havran1_ennis_298x298_LDR')
parser.add_argument('-j', '--workers',
default=12, type=int, metavar='N',
help='number of data loading workers (default: 12)')
parser.add_argument('--epochs',
default=25, type=int, metavar='N',
help='number of total epochs to run')
parser.add_argument('--start-epoch',
default=0, type=int, metavar='N',
help='manual epoch number (useful on restarts)')
parser.add_argument('--adjust-epoch',
nargs='+', default=[10, 15, 20],
type=int, help='milestones to adjust the learning rate')
parser.add_argument('--num-classes', default=100, type=int,
help='number of classes in the problem')
parser.add_argument('--emb-size',
default=128, type=int, help='size of the embedding')
parser.add_argument('-b', '--batch-size',
default=20, type=int,
metavar='N', help='mini-batch size (default: 20)')
parser.add_argument('--lr', '--learning-rate',
default=1e-3, type=float,
metavar='LR', help='initial learning rate', dest='lr')
parser.add_argument('--wd', '--weight-decay',
default=5e-4, type=float,
metavar='W', help='weight decay (default: 5e-4)',
dest='weight_decay')
parser.add_argument('--betas',
nargs='+', default=[0.9, 0.999], type=float,
help='beta values for ADAM')
parser.add_argument('--momentum',
default=0.9, type=float,
help='momentum in the SGD')
parser.add_argument('--margin',
default=0.3, type=float,
help='triplet loss margin')
parser.add_argument('--checkpoint-folder',
default='./checkpoints',
type=str, help='folder to store the trained models')
parser.add_argument('--model-name',
default='resnet_similarity', type=str,
help='name given to the model')
parser.add_argument('--resume',
default='', type=str, metavar='PATH',
help='path to latest checkpoint (default: none)')
parser.add_argument('-e', '--evaluate',
dest='evaluate', action='store_true',
help='evaluate model on validation set')
parser.add_argument('--seed',
default=2851, type=int,
help='seed for initializing training. ')
class AverageMeter(object):
"""
https://github.com/pytorch/examples/blob/master/imagenet/single.py
Computes and stores the average and current value
"""
def __init__(self):
self.reset()
def reset(self):
self.val = 0
self.avg = 0
self.sum = 0
self.count = 0
def update(self, val, n=1):
self.val = val
self.sum += val * n
self.count += n
self.avg = self.sum / self.count
class RandomResize(object):
def __init__(self, low, high, interpolation=Image.BILINEAR):
self.low = low
self.high = high
self.interpolation = interpolation
def __call__(self, img):
size = np.random.randint(self.low, self.high)
return transforms_F.resize(img, size, self.interpolation)
def train_model(loader, epoch):
def update_progress_bar(progress_bar, losses):
description = '[' + str(epoch) + '-train]'
description += ' Triplet loss: '
description += '%.4f/ %.4f (AVG)' % (losses.val, losses.avg)
progress_bar.set_description(description)
global model
global criterion
global optimizer
# keep track of the loss value
losses = AverageMeter()
progress_bar = tqdm(loader, total=len(loader))
for imgs, targets in progress_bar:
with torch.set_grad_enabled(True):
imgs = imgs.to(device, dtype)
targets = targets.to(device, dtype)
# forward through the model and compute error
_, embeddings = model(imgs)
loss = criterion(embeddings, targets)
losses.update(loss.item(), imgs.size(0))
# compute gradient and update parameters
optimizer.zero_grad()
loss.backward()
optimizer.step()
update_progress_bar(progress_bar, losses)
return losses.avg
def evaluate_model(mturk_images):
global model
global criterion
global optimizer
with torch.set_grad_enabled(False):
# get current agreement with users answers
current_agreement_train = criterion.get_majority_accuracy(
mturk_images=mturk_images,
model=model,
train=True,
unit_norm=True
)
current_agreement = criterion.get_majority_accuracy(
mturk_images=mturk_images,
model=model,
train=False,
unit_norm=True
)
tqdm.write('[Train]Current agreement %.4f (Best agreement %.4f)' %
(current_agreement_train, best_agreement))
tqdm.write('[Test]Current agreement %.4f (Best agreement %.4f)' %
(current_agreement, best_agreement))
return current_agreement
def get_transforms():
# set image transforms
trf_train = transforms.Compose([
transforms.RandomRotation(degrees=(-5, 5)),
transforms.CenterCrop(size=384),
RandomResize(low=256, high=384),
transforms.RandomCrop(size=224),
transforms.RandomVerticalFlip(),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
])
trf_test = transforms.Compose([
transforms.Resize(size=256),
transforms.CenterCrop(size=224),
transforms.ToTensor(),
])
return trf_train, trf_test
def get_dataloaders(trf_train):
global args
def _init_loader_(worker_id):
np.random.seed(args.seed + worker_id)
loader_args = {
'batch_size': args.batch_size,
'num_workers': args.workers,
'pin_memory': True,
'worker_init_fn': _init_loader_,
}
loader_train = DataLoader(
dataset=ImageFolder(
root=os.path.join(args.train_dir, 'train'),
transform=trf_train,
),
shuffle=True,
drop_last=True,
**loader_args
)
loader_val = DataLoader(
dataset=ImageFolder(
root=os.path.join(args.train_dir, 'val'),
transform=trf_train,
),
shuffle=True,
**loader_args
)
return loader_train, loader_val
def save_checkpoint(state, is_best, folder, model_name='checkpoint', ):
"""
if the current state is the best it saves the pytorch model
in folder with name filename
"""
path = os.path.join(folder, model_name)
os.makedirs(path, exist_ok=True)
path = os.path.join(path, 'model')
torch.save(state, path + '.pth.tar')
if is_best:
shutil.copyfile(path + '.pth.tar', path + '_best.pth.tar')
if __name__ == '__main__':
# get input arguments
args = parser.parse_args()
# set seeds
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
np.random.seed(args.seed)
random.seed(args.seed)
# set device and dtype
dtype = torch.float
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
if device.type == 'cuda':
# # comment this if we want reproducibility
# torch.backends.cudnn.benchmark = True
# torch.backends.cudnn.enabled = True
# # this might affect performance but allows reproducibility
torch.backends.cudnn.enabled = False
torch.backends.cudnn.deterministic = True
# define dataset
trf_train, trf_test = get_transforms()
loader_train, loader_val = get_dataloaders(trf_train)
mturk_images, _ = utils.load_imgs(args.test_dir, trf_test)
# create model
model = FTModel(
models.resnet34(pretrained=True),
layers_to_remove=1,
num_features=args.emb_size,
num_classes=args.num_classes,
)
model = model.to(device, dtype)
# define loss function
criterion = TripletLossHuman(
margin=args.margin,
unit_norm=True,
device=device,
seed=args.seed
)
# define optimizer
# optimizer = optim.Adam(
# model.parameters(),
# betas=args.betas,
# weight_decay=args.weight_decay,
# lr=args.lr,
# amsgrad=True
# )
optimizer = optim.SGD(
model.parameters(),
weight_decay=args.weight_decay,
lr=args.lr,
momentum=args.momentum,
nesterov=True
)
# define LR scheduler
lr_scheduler = optim.lr_scheduler.MultiStepLR(
optimizer,
milestones=args.adjust_epoch,
gamma=0.1,
)
# set a high value for the error
best_agreement = 0
if args.evaluate:
# evaluation step
model = model.eval()
evaluate_model()
else:
# start training and evaluation loop
for epoch in range(args.start_epoch + 1, args.epochs + 1):
# train step
model = model.train()
train_model(loader_train, epoch)
lr_scheduler.step()
# evaluation step
model = model.eval()
current_agreement = evaluate_model(mturk_images)
# checkpoint model if it is the best
is_best = current_agreement > best_agreement
best_agreement = max(current_agreement, best_agreement)
save_checkpoint(
{
'epoch': epoch,
'state_dict': model.state_dict(),
'best_agreement': best_agreement,
'optimizer': optimizer.state_dict(),
},
is_best, folder=args.checkpoint_folder,
model_name=args.model_name + '-' + current_time
)