-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
310 lines (258 loc) · 11.3 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
import warnings
from MobileNet import MobileNetV2
import pretrainedmodels
import torchvision
warnings.filterwarnings("ignore")
from include import *
from datasets import Dataset_
from models import *
from utils import *
import argparse
from Nadam import Nadam
from losses import CrossEntropyLabelSmooth
import model
def softmax_loss(results, labels):
# print(results.shape)
# labels = labels.view(-1)
# print(labels.shape)
loss = F.cross_entropy(results, labels, reduce=True)
return loss
CE_smooth = CrossEntropyLabelSmooth(num_classes=50)
def do_valid(net, valid_loader):
valid_num = 0
truths = []
losses = []
corrects = []
probs = []
labels = []
with torch.no_grad():
for input, truth_ in valid_loader:
input = input.cuda()
truth_ = truth_.cuda().long()
# input = to_var(input)
# truth_ = to_var(truth_)
logit = net(input)
loss = softmax_loss(logit, truth_)
probs.append(logit)
labels.append(truth_)
valid_num += len(input)
loss_tmp = loss.data.cpu().numpy().reshape([1])
losses.append(loss_tmp)
truths.append(truth_.data.cpu().numpy())
assert (valid_num == len(valid_loader.sampler))
# ------------------------------------------------------
loss = np.concatenate(losses,axis=0)
loss = loss.mean()
prob = torch.cat(probs)
label = torch.cat(labels)
_, precision = metric(prob, label)
return loss, precision
def run_train(config, fold=0):
if config.model == 'res50':
net = res50()
elif config.model == 'se50':
net = se50()
elif config.model == 'res34':
net = res34()
elif config.model == 'se101':
net = se101()
elif config.model == 'res34_atten':
net = res34_attention_pool()
elif config.model == 'se154':
net = se154()
elif config.model =='densenet201':
net = torchvision.models.densenet201(pretrained=True)
net.classifier = nn.Linear(net.classifier.in_features, 5)
elif config.model =='resnet152':
net = torchvision.models.resnet152(num_classes=1000, pretrained=True)
net.fc = nn.Linear(net.fc.in_features, 5)
elif config.model =='densenet161':
net = torchvision.models.densenet161(pretrained=True)
net.classifier = nn.Linear(net.classifier.in_features, 5)
elif config.model == 'inceptionv4':
net = pretrainedmodels.__dict__['inceptionv4'](num_classes=1000, pretrained='imagenet')
net.last_linear = nn.Linear(net.last_linear.in_features, 5)
# IMAGE_SIZE = 299
elif config.model == 'inceptionresnetv2':
net = model.inceptionresnetv2_finetune(5)
elif config.model == 'mobilenet':
net = MobileNetV2(n_class=1000)
net.load_state_dict(torch.load("../mobilenet_v2.pth.tar"))
net.classifier = nn.Linear(net.last_channel, 5)
#net.load_state_dict(torch.load('../ckpt/res34/model_6'))
net = nn.DataParallel(net)
net.cuda()
train_dataset = Dataset_('train', image_size=(config.image_h, config.image_w), is_pseudo=True, fold=fold)
train_loader = DataLoader(train_dataset, shuffle=True, batch_size=config.batch_size, num_workers=14, pin_memory=True)
valid_dataset = Dataset_('val', image_size=(config.image_h, config.image_w), is_pseudo=False, fold=fold)
valid_loader = DataLoader(valid_dataset, shuffle=False, batch_size=config.batch_size, num_workers=14, pin_memory=True)
if not os.path.isdir('../logs_extra/{}'.format(config.model)):
os.mkdir('../logs_extra/{}'.format(config.model))
log = open('../logs_extra/{}'.format(config.model)+'/log.train.txt', mode='a')
log.write('\t__file__ = %s\n')
log.write('\tout_dir = %s\n')
log.write('\n')
log.write('\t<additional comments>\n')
log.write('\t ... xxx baseline ... \n')
log.write('\n')
## dataset ----------------------------------------
log.write('** dataset setting **\n')
assert(len(train_dataset)>=config.batch_size)
log.write('batch_size = %d\n'%(config.batch_size))
log.write('train_dataset : \n%s\n'%(train_dataset))
log.write('valid_dataset : \n%s\n'%(valid_dataset))
log.write('\n')
## net ----------------------------------------
log.write('** net setting **\n')
# if initial_checkpoint is not None:
# log.write('\tinitial_checkpoint = %s\n' % initial_checkpoint)
# net.load_state_dict(torch.load(initial_checkpoint, map_location=lambda storage, loc: storage))
# print('\tinitial_checkpoint = %s\n' % initial_checkpoint)
log.write('%s\n'%(type(net)))
log.write('\n')
# optimizer = torch.optim.Adam(filter(lambda p: p.requires_grad, net.parameters()), lr=config.lr)
optimizer = optim.Adam(filter(lambda p: p.requires_grad, net.parameters()), lr=config.lr)
#optimizer = Nadam(filter(lambda p: p.requires_grad, net.parameters()), lr=config.lr)
iter_smooth = 20
start_iter = 0
log.write('\n')
## start training here! ##############################################
log.write('** top_n step 100,60,60,60 **\n')
log.write('** start training here! **\n')
log.write(' |---- VALID ----|- TRAIN/BATCH -| \n')
log.write('rate iter epoch | loss acc-1 | loss acc-1 | time \n')
log.write('--------------------------------------------------------------\n')
print('** start training here! **\n')
print(' |----- VALID ----|--TRAIN/BATCH -| \n')
print('rate iter epoch | loss acc-1 | loss acc-1 | time \n')
print('---------------------------------------------------------------\n')
def adjust_lr(optimizer, ep):
if ep < 12:
lr = 3e-4
elif ep < 16:
lr = 1e-4
elif ep < 19:
lr = 1e-5
else:
lr = 1e-6
for p in optimizer.param_groups:
p['lr'] = lr
return lr
# def get_lr(ep):
# if ep < 12:
# lr = 3e-4
# elif ep < 16:
# lr = 1e-4
# elif ep < 19:
# lr = 1e-5
# else:
# lr = 1e-6
# return lr
# def adjust_lr_ep(optimizer, ep):
# lr = config.lr * ep / 10
# def adjust_lr(optimizer, lr):
# for p in optimizer.param_groups:
# p['lr'] = lr
i = 0
start = timer()
max_valid = 0.
patience = 0
max_lr_change = 3
lrs = [3e-4, 1e-4, 1e-5, 1e-6]
k = 0
for epoch in range(config.train_epoch):
train_loss = []
train_acc = []
valid_loss = []
valid_acc = []
rate = adjust_lr(optimizer, epoch)
# rate = get_lr(epoch)
optimizer.zero_grad()
# rate, hard_ratio = adjust_lr_and_hard_ratio(optimizer, epoch + 1)
# rate = lrs[k]
for input, truth_ in train_loader:
iter = i + start_iter
# one iteration update -------------
net.train()
input = input.cuda()
truth_ = truth_.cuda().long()
# print(truth_)
# input = to_var(input)
# truth_ = to_var(truth_)
logit = net(input)
# loss_focal = focal_OHEM(logit, truth_,truth, hard_ratio)* config.focal_w
loss_softmax = softmax_loss(logit, truth_) * config.softmax_w
# loss_triplet = TripletLoss(margin=0.3)(feas, truth_, normalize_feature=True) * config.triplet_w
loss = loss_softmax
_, precision = metric(logit, truth_)
loss.backward()
optimizer.step()
optimizer.zero_grad()
train_loss.append(loss.item())
train_acc.append(precision.item())
train_loss = np.mean(train_loss)
train_acc = np.mean(train_acc)
net.eval()
valid_loss, val_acc = do_valid(net, valid_loader)
net.train()
print('%0.6f %5.1f %6.1f | %0.3f %0.3f%s | %0.3f %0.3f | %s' % (\
rate, iter, epoch,
valid_loss, val_acc,' ',
train_loss, train_acc,
time_to_str((timer() - start),'min')))
log.write('%0.6f %5.1f %6.1f | %0.3f %0.3f%s | %0.3f %0.3f | %s' % (\
rate, iter, epoch,
valid_loss, val_acc,' ',
train_loss, train_acc,
time_to_str((timer() - start),'min')))
log.write('\n')
if max_valid < val_acc:
patience = 0
max_valid = val_acc
# print('save max valid!!!!!! : ' + str(max_valid))
# log.write('save max valid!!!!!! : ' + str(max_valid))
# log.write('\n')
if not os.path.isdir('../ckpt_extra/{}_4/'.format(config.model)):
os.mkdir('../ckpt_extra/{}_4/'.format(config.model))
torch.save(net.state_dict(), '../ckpt_extra/{}_4/{}_{}'.format(config.model, config.model_name, fold))
# else:
# patience += 1
# if patience == 4:
# k += 1
# if k == 4:
# break
# # adjust_lr(optimizer, ep)
# adjust_lr(optimizer, lrs[k])
# net.load_state_dict(torch.load('../ckpt/{}_4/{}_{}'.format(config.model, config.model_name, fold)))
# patience = 0
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--train_fold_index', type=int, default = 0)
parser.add_argument('--model', type=str, default='res34')
parser.add_argument('--model_name', type=str, default='model')
parser.add_argument('--log_name', type=str, default='log1')
parser.add_argument('--batch_size', type=int, default=64)
parser.add_argument('--lr', type=float, default=3e-4)
parser.add_argument('--image_h', type=int, default=512)
parser.add_argument('--image_w', type=int, default=512)
parser.add_argument('--s1', type=float, default=64.0)
parser.add_argument('--m1', type=float, default=0.0)
parser.add_argument('--s2', type=float, default=16.0)
parser.add_argument('--focal_w', type=float, default=1.0)
parser.add_argument('--softmax_w', type=float, default=1.0)
parser.add_argument('--triplet_w', type=float, default=1.0)
# parser.add_argument('--mode', type=str, default='train', choices=['train', 'val','val_fold','test_classifier','test','test_fold'])
# parser.add_argument('--pretrained_model', type=str, default=None)
#
parser.add_argument('--mode', type=str, default='test_classifier', choices=['train', 'val','val_fold','test_classifier','test','test_fold'])
parser.add_argument('--pretrained_model', type=str, default='max_valid_model.pth')
# parser.add_argument('--fold', type=int, default=4, required=True)
parser.add_argument('--iter_save_interval', type=int, default=5)
parser.add_argument('--train_epoch', type=int, default=22)
config = parser.parse_args()
for fold in range(5):
# fold = config.fold
print('fold_{}'.format(fold))
# if fold > 0:
# break
run_train(config, fold)