-
Notifications
You must be signed in to change notification settings - Fork 2
/
train_single_network.py
445 lines (390 loc) · 20.8 KB
/
train_single_network.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
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
import logging
import math
import os
import time
import warnings
import numpy as np
import torch.optim
from tensorboardX import SummaryWriter
from torch import nn
import argparse
from losses import Losses_mcau, Losses_u2net, Losses_unet, Loss_unetpp, Loss_unet3p
from DataSet import load_train_val_data, load_ddr_train_val
from torch.optim.lr_scheduler import CosineAnnealingWarmRestarts
from nets import U2Net, UNet, unetpp, ENet
from nets.CAUNet import CAUNet
from nets.UCTransNet import UCTransNet, get_CTranS_config
from nets.UDTransNet import UDTransNet, get_model_config
from utils import auc_on_batch, aupr_on_batch, plot
from nets.attention_unet import AttU_Net
from nets.vit_seg_modeling import VisionTransformer as ViT_seg
from nets.vit_seg_modeling import CONFIGS as CONFIGS_ViT_seg
from nets.UNet3Plus import UNet3Plus_DeepSup
from nets.attention_unet import AttU_Net
from nets.res_unet_plus import ResUnetPlusPlus
os.environ["CUDA_VISIBLE_DEVICES"] = '0'
warnings.filterwarnings("ignore")
os.environ['QTQPAPLATFORM'] = 'offscreen'
def logger_config(log_path):
loggerr = logging.getLogger()
loggerr.setLevel(level=logging.INFO)
handler = logging.FileHandler(log_path, encoding='UTF-8')
handler.setLevel(logging.INFO)
formatter = logging.Formatter('%(message)s')
handler.setFormatter(formatter)
console = logging.StreamHandler()
console.setLevel(logging.INFO)
loggerr.addHandler(handler)
loggerr.addHandler(console)
return loggerr
def print_summary_whole(epoch, i, nb_batch, loss, loss_name, batch_time,
average_loss, average_time, mode, lr, ex, ex_mean, he, he_mean, ma, ma_mean, se, se_mean):
summary = ' [' + str(mode) + '] Epoch: [{0}][{1}/{2}] '.format(
epoch, i, nb_batch)
string = ''
string += '{}: {:.3f} '.format(loss_name, loss)
string += '(Avg {:.4f}) '.format(average_loss)
if not math.isnan(ex):
string += 'EX_AUPR {:.3f} '.format(ex)
string += '(Avg {:.4f}) '.format(ex_mean)
if not math.isnan(he):
string += 'HE_AUPR {:.3f} '.format(he)
string += '(Avg {:.4f}) '.format(he_mean)
if not math.isnan(ma):
string += 'MA_AUPR {:.3f} '.format(ma)
string += '(Avg {:.4f}) '.format(ma_mean)
if not math.isnan(se):
string += 'SE_AUPR {:.3f} '.format(se)
string += '(Avg {:.4f}) '.format(se_mean)
if mode == 'Train':
string += 'LR {:.2e} '.format(lr)
string += 'Time {:.1f} '.format(batch_time)
string += '(Avg {:.1f}) '.format(average_time)
summary += string
logger.info(summary)
def save_checkpoint(state, save_path):
logger.info('\t Saving to {}'.format(save_path))
if not os.path.isdir(save_path):
os.makedirs(save_path)
epoch = state['epoch'] # epoch no
best_model = state['best_model'] # bool
model = state['model'] # model type
loss = state['loss'] # loss name
if best_model:
filename = save_path + '/' + \
'best_model.{}--{}.pth.tar'.format(loss, model)
else:
filename = save_path + '/' + \
'model.{}--{}--{:02d}.pth.tar'.format(loss, model, epoch)
torch.save(state, filename)
# Train One Epoch
def train_one_epoch(loader, model, criterion, optimizer, writer, epoch, lr_scheduler, model_type, lbtw_algorithm,
batch_size):
logging_mode = 'Train' if model.training else 'Val'
end = time.time()
time_sum, loss_sum = 0, 0
EX_auc, EX_aupr = [], []
HE_auc, HE_aupr = [], []
MA_auc, MA_aupr = [], []
SE_auc, SE_aupr = [], []
for (i, sample) in enumerate(loader, 1):
images, masks = sample['image'], sample['masks'].permute(0, 2, 3, 1)
try:
loss_name = criterion._get_name()
except AttributeError:
loss_name = criterion.__name__
# Take variable and put them to GPU
images = images.to(device, non_blocking=True).float()
masks = masks.to(device, non_blocking=True).float()
if model_type == 'u2net':
if model.training:
optimizer.zero_grad()
output, d1, d2, d3, d4, d5, d6 = model(images)
out_loss, loss = criterion(output, d1, d2, d3, d4, d5, d6, masks)
loss.backward()
optimizer.step()
else:
output, d1, d2, d3, d4, d5, d6 = model(images)
out_loss, loss = criterion(output, d1, d2, d3, d4, d5, d6, masks)
elif model_type == 'mcaunet':
output, pred_comb2, pred_comb3, pred_comb1 = model(images)
losses, out_loss, c3_loss, c2_loss, c1_loss = criterion(output, pred_comb3, pred_comb2, pred_comb1, masks)
if model.training:
loss, w0, w3, w2, w1 = lbtw_algorithm(i, out_loss, c3_loss, c2_loss, c1_loss)
optimizer.zero_grad()
loss.backward()
optimizer.step()
elif model_type == 'unet3p':
outputs = model(images)
output, pred_comb3, pred_comb2, pred_comb1, pred_comb0 = outputs[0], outputs[1], outputs[2], outputs[3], \
outputs[4]
losses, out_loss, c3_loss, c2_loss, c1_loss, c0_loss = criterion(output, pred_comb3, pred_comb2, pred_comb1,
pred_comb0, masks)
if model.training:
loss, w0, w3, w2, w1, w0 = lbtw_algorithm(i, out_loss, c3_loss, c2_loss, c1_loss, c0_loss)
optimizer.zero_grad()
loss.backward()
optimizer.step()
elif model_type == 'unet++':
outputs = model(images)
output, pred_comb3, pred_comb2, pred_comb1 = outputs[-1], outputs[-2], outputs[-3], outputs[-4]
losses, out_loss, c3_loss, c2_loss, c1_loss = criterion(output, pred_comb3, pred_comb2, pred_comb1, masks)
if model.training:
loss, w0, w3, w2, w1 = lbtw_algorithm(i, out_loss, c3_loss, c2_loss, c1_loss)
optimizer.zero_grad()
loss.backward()
optimizer.step()
elif model_type == 'unet' or 'uctransnet' or 'udtransnet' or 'attenunet' or 'resunetpp' or 'transunet' or 'enet':
output = model(images)
loss = criterion(output, masks)
out_loss = loss
if model.training:
optimizer.zero_grad()
loss.backward()
optimizer.step()
# auc aupr for single lesion
EX_masks, EX_output = masks[..., 1], output[..., 1]
HE_masks, HE_output = masks[..., 2], output[..., 2]
MA_masks, MA_output = masks[..., 3], output[..., 3]
SE_masks, SE_output = masks[..., 4], output[..., 4]
EX_cur_auc, EX_cur_aupr = auc_on_batch(EX_masks, EX_output), aupr_on_batch(EX_masks, EX_output)
HE_cur_auc, HE_cur_aupr = auc_on_batch(HE_masks, HE_output), aupr_on_batch(HE_masks, HE_output)
MA_cur_auc, MA_cur_aupr = auc_on_batch(MA_masks, MA_output), aupr_on_batch(MA_masks, MA_output)
SE_cur_auc, SE_cur_aupr = auc_on_batch(SE_masks, SE_output), aupr_on_batch(SE_masks, SE_output)
if not math.isnan(EX_cur_auc):
EX_auc.append(EX_cur_auc), EX_aupr.append(EX_cur_aupr)
if not math.isnan(HE_cur_auc):
HE_auc.append(HE_cur_auc), HE_aupr.append(HE_cur_aupr)
if not math.isnan(MA_cur_auc):
MA_auc.append(MA_cur_auc), MA_aupr.append(MA_cur_aupr)
if not math.isnan(SE_cur_auc):
SE_auc.append(SE_cur_auc), SE_aupr.append(SE_cur_aupr)
# measure elapsed time
batch_time = time.time() - end
time_sum += len(images) * batch_time
loss_sum += len(images) * out_loss
if i == len(loader):
average_loss = loss_sum / (batch_size * (i - 1) + len(images))
average_time = time_sum / (batch_size * (i - 1) + len(images))
else:
average_loss = loss_sum / (i * batch_size)
average_time = time_sum / (i * batch_size)
end = time.time()
torch.cuda.empty_cache()
if i % print_frequency == 0:
print_summary_whole(epoch + 1, i, len(loader), out_loss, loss_name, batch_time,
average_loss, average_time, logging_mode,
min(g["lr"] for g in optimizer.param_groups),
EX_cur_aupr, np.mean(EX_aupr), HE_cur_aupr, np.mean(HE_aupr),
MA_cur_aupr, np.mean(MA_aupr), SE_cur_aupr, np.mean(SE_aupr))
if tensorboard:
step = epoch * len(loader) + i
writer.add_scalar(logging_mode + '_' + loss_name, out_loss.item(), step)
# plot metrics in tensorboard
if not math.isnan(EX_cur_auc):
writer.add_scalar(logging_mode + '_EX_auc', EX_cur_auc, step)
writer.add_scalar(logging_mode + '_EX_aupr', EX_cur_aupr, step)
if not math.isnan(HE_cur_auc):
writer.add_scalar(logging_mode + '_HE_auc', HE_cur_auc, step)
writer.add_scalar(logging_mode + '_HE_aupr', HE_cur_aupr, step)
if not math.isnan(MA_cur_auc):
writer.add_scalar(logging_mode + '_MA_auc', MA_cur_auc, step)
writer.add_scalar(logging_mode + '_MA_aupr', MA_cur_aupr, step)
if not math.isnan(SE_cur_auc):
writer.add_scalar(logging_mode + '_SE_auc', SE_cur_auc, step)
writer.add_scalar(logging_mode + '_SE_aupr', SE_cur_aupr, step)
torch.cuda.empty_cache()
if lr_scheduler is not None:
lr_scheduler.step()
return np.mean(EX_aupr), np.mean(HE_aupr), np.mean(MA_aupr), np.mean(SE_aupr)
def main_loop(batch_size, model_type, tensorboard=True, task_name=None, dataset=None):
# Load train and val data
global model, criterion
tasks = task_name
n_labels = len(tasks) + 1
lr = learning_rate
n_channels = 3
if dataset == 'idrid':
train_loader, val_loader = load_train_val_data(batch_size=batch_size)
elif dataset == 'ddr':
train_loader, val_loader = load_ddr_train_val(batch_size=batch_size)
if model_type == 'u2net':
model = U2Net.U2NET(n_channels, n_labels)
criterion = Losses_u2net.LBTW_Loss(Losses_u2net.WeightedDiceBCE()).to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9, weight_decay=5e-4)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=50, gamma=0.5)
lbtw_algorithm = Losses_u2net.LBTW_algorithm()
elif model_type == 'mcaunet':
model = CAUNet(n_channels, n_labels)
criterion = Losses_mcau.LBTW_Loss(Losses_mcau.WeightedDiceBCE()).to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=lr) # Choose optimize
lr_scheduler = CosineAnnealingWarmRestarts(optimizer, T_0=15, T_mult=1, eta_min=5e-5)
lbtw_algorithm = Losses_mcau.LBTW_algorithm()
elif model_type == 'unet':
model = UNet.UNet(n_channels, n_labels)
criterion = Losses_unet.LBTW_Loss(Losses_unet.WeightedDiceBCE()).to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9, weight_decay=5e-4)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=50, gamma=0.5)
lbtw_algorithm = Losses_u2net.LBTW_algorithm()
elif model_type == 'enet':
model = ENet.ENet(n_labels)
criterion = Losses_unet.LBTW_Loss(Losses_unet.WeightedDiceBCE()).to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9, weight_decay=5e-4)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=50, gamma=0.5)
lbtw_algorithm = Losses_u2net.LBTW_algorithm()
elif model_type == 'unet++':
model = unetpp.NestedUNet(n_channels, n_labels, deepsupervision=True)
criterion = Loss_unetpp.LBTW_Loss(Loss_unetpp.WeightedDiceBCE()).to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9, weight_decay=5e-4)
lr_scheduler = CosineAnnealingWarmRestarts(optimizer, T_0=15, T_mult=1, eta_min=5e-5)
lbtw_algorithm = Loss_unetpp.LBTW_algorithm()
elif model_type == 'uctransnet':
config = get_CTranS_config()
model = UCTransNet(config, n_channels, n_labels, img_size=640)
criterion = Losses_unet.LBTW_Loss(Losses_unet.WeightedDiceBCE()).to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9, weight_decay=5e-4)
lr_scheduler = CosineAnnealingWarmRestarts(optimizer, T_0=15, T_mult=1, eta_min=5e-5)
lbtw_algorithm = Losses_u2net.LBTW_algorithm()
elif model_type == 'udtransnet':
config = get_model_config()
model = UDTransNet(config, n_channels, n_labels, img_size=640)
criterion = Losses_unet.LBTW_Loss(Losses_unet.WeightedDiceBCE()).to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9, weight_decay=5e-4)
lr_scheduler = CosineAnnealingWarmRestarts(optimizer, T_0=15, T_mult=1, eta_min=5e-5)
lbtw_algorithm = Losses_u2net.LBTW_algorithm()
elif model_type == 'attenunet':
model = AttU_Net(n_channels, n_labels)
criterion = Losses_unet.LBTW_Loss(Losses_unet.WeightedDiceBCE()).to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9, weight_decay=5e-4)
lr_scheduler = CosineAnnealingWarmRestarts(optimizer, T_0=15, T_mult=1, eta_min=5e-5)
lbtw_algorithm = Losses_u2net.LBTW_algorithm()
elif model_type == 'resunetpp':
model = ResUnetPlusPlus(n_channels)
criterion = Losses_unet.LBTW_Loss(Losses_unet.WeightedDiceBCE()).to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9, weight_decay=5e-4)
lr_scheduler = CosineAnnealingWarmRestarts(optimizer, T_0=15, T_mult=1, eta_min=5e-5)
lbtw_algorithm = Losses_u2net.LBTW_algorithm()
elif model_type == 'unet3p':
model = UNet3Plus_DeepSup(n_channels, n_labels)
criterion = Loss_unet3p.LBTW_Loss(Loss_unet3p.WeightedDiceBCE()).to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9, weight_decay=5e-4)
lr_scheduler = CosineAnnealingWarmRestarts(optimizer, T_0=15, T_mult=1, eta_min=5e-5)
lbtw_algorithm = Loss_unet3p.LBTW_algorithm()
elif model_type == 'transunet':
parser = argparse.ArgumentParser()
parser.add_argument('--img_size', type=int,
default=640, help='input patch size of network input')
parser.add_argument('--n_skip', type=int,
default=3, help='using number of skip-connect, default is num')
parser.add_argument('--vit_name', type=str,
default='R50-ViT-B_16', help='select one vit model')
parser.add_argument('--vit_patches_size', type=int,
default=16, help='vit_patches_size, default is 16')
args = parser.parse_args()
config_vit = CONFIGS_ViT_seg[args.vit_name]
config_vit.n_classes = 5
config_vit.n_skip = args.n_skip
if args.vit_name.find('R50') != -1:
config_vit.patches.grid = (
int(args.img_size / args.vit_patches_size), int(args.img_size / args.vit_patches_size))
model = ViT_seg(config_vit, img_size=args.img_size, num_classes=config_vit.n_classes)
model.load_from(weights=np.load(config_vit.pretrained_path))
criterion = Losses_unet.LBTW_Loss(Losses_unet.WeightedDiceBCE()).to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=0.9, weight_decay=5e-4)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=50, gamma=0.5)
lbtw_algorithm = Losses_u2net.LBTW_algorithm()
model = nn.DataParallel(model)
model = model.to(device)
if checkpoint:
checkpoints = torch.load(checkpoint, map_location=device)
model.load_state_dict(checkpoints['state_dict'])
print("Model Loaded!")
print("Let's use {0} GPUs!".format(torch.cuda.device_count()))
loss_name = criterion._get_name()
if tensorboard:
log_dir = tensorboard_folder + session_name + '/'
logger.info('log dir: '.format(log_dir))
if not os.path.isdir(log_dir):
os.makedirs(log_dir)
writer = SummaryWriter(log_dir)
EX_best, HE_best, MA_best, SE_best, mean_best = 0, 0, 0, 0, 0
EX_best_epoch, HE_best_epoch, MA_best_epoch, SE_best_epoch, mean_best_epoch = 0, 0, 0, 0, 0
logger.info('Train Start')
for epoch in range(epochs): # loop over the dataset multiple times
logger.info('\n========= Epoch [{}/{}] ========='.format(epoch + 1, epochs))
logger.info(session_name)
# train for one epoch
model.train(True)
logger.info('Training with batch size : {}'.format(batch_size))
train_one_epoch(train_loader, model, criterion, optimizer, writer, epoch, lr_scheduler, model_type,
lbtw_algorithm,
batch_size=batch_size)
if epoch >= 100:
with torch.no_grad():
model.eval()
ex_aupr, he_aupr, ma_aupr, se_aupr = train_one_epoch(val_loader, model, criterion, optimizer,
writer,
epoch, None,
model_type, lbtw_algorithm,
batch_size=batch_size)
save_flag = False
if ex_aupr > EX_best:
EX_best = ex_aupr
EX_best_epoch = epoch
save_flag = True
if he_aupr > HE_best:
HE_best = he_aupr
HE_best_epoch = epoch
save_flag = True
if ma_aupr > MA_best:
MA_best = ma_aupr
MA_best_epoch = epoch
save_flag = True
if se_aupr > SE_best:
SE_best = se_aupr
SE_best_epoch = epoch
save_flag = True
current_mean = (ex_aupr + he_aupr + ma_aupr + se_aupr) / 4
if current_mean > mean_best:
mean_best = current_mean
mean_best_epoch = epoch
save_flag = True
logger.info('Best model: EX:{} \t HE:{} \t MA:{} \t SE:{} \n mAUPR:{}'.format(EX_best_epoch, HE_best_epoch,
MA_best_epoch, SE_best_epoch,
mean_best_epoch))
logger.info(
'Best AUPR: EX:{} \t HE:{} \t MA:{} \t SE:{} \n mAUPR:{}'.format(EX_best, HE_best, MA_best, SE_best,
mean_best))
if (save_flag and epoch >= 250) or (save_flag and checkpoint):
save_checkpoint({'epoch': epoch,
'best_model': False,
'model': model_type,
'state_dict': model.state_dict(),
'val_loss': '123',
'loss': loss_name,
'optimizer': optimizer.state_dict()}, model_path)
if __name__ == '__main__':
task_names = ['EX', 'MA', 'HE', 'SE']
# opitons: ['mcaunet', 'u2net', 'unet', 'unet++', 'uctransnet', 'udtransnet', 'attenunet', 'resunetpp', 'unet3p', 'transunet', 'enet']
model_name = 'unet'
dataset = 'idrid' # idrid or ddr
device = 'cuda' if torch.cuda.is_available() else 'cpu'
checkpoint = None
learning_rate = 5e-3
epochs = 400
batch_size = 2
print_frequency = 1
save_frequency = 10
save_model = True
tensorboard = True
save_path = os.path.join('./single_network_log', model_name)
session_name = 'Test_session' + '_' + time.strftime('%m.%d %Hh%M')
model_path = save_path + 'models/' + session_name + '/'
tensorboard_folder = save_path + 'tensorboard_logs/'
logger_path = save_path + 'log_file/'
if not os.path.isdir(logger_path):
os.makedirs(logger_path)
logger_path = save_path + 'log_file/' + session_name + ".log"
logger = logger_config(log_path=logger_path)
main_loop(model_type=model_name, tensorboard=True,
task_name=task_names, batch_size=batch_size, dataset=dataset)