-
Notifications
You must be signed in to change notification settings - Fork 16
/
train_synapse.py
221 lines (162 loc) · 7.43 KB
/
train_synapse.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
import torch
from torch import nn
from torch.cuda.amp import autocast, GradScaler
from torch.utils.data import DataLoader
import torch.distributed as dist
from torch.nn.parallel import DistributedDataParallel as DDP
from torch.utils.data.distributed import DistributedSampler
from torchvision import transforms
from datasets.dataset import RandomGenerator
from engine_synapse import *
from models.vmunet.vmunet import VMUNet
import os
import sys
os.environ["CUDA_VISIBLE_DEVICES"] = "0" # "0, 1, 2, 3"
from utils import *
from configs.config_setting_synapse import setting_config
import warnings
warnings.filterwarnings("ignore")
def main(config):
print('#----------Creating logger----------#')
sys.path.append(config.work_dir + '/')
log_dir = os.path.join(config.work_dir, 'log')
checkpoint_dir = os.path.join(config.work_dir, 'checkpoints')
resume_model = os.path.join(checkpoint_dir, 'latest.pth')
outputs = os.path.join(config.work_dir, 'outputs')
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
if not os.path.exists(outputs):
os.makedirs(outputs)
global logger
logger = get_logger('train', log_dir)
log_config_info(config, logger)
print('#----------GPU init----------#')
set_seed(config.seed)
gpu_ids = [0]# [0, 1, 2, 3]
torch.cuda.empty_cache()
gpus_type, gpus_num = torch.cuda.get_device_name(), torch.cuda.device_count()
if config.distributed:
print('#----------Start DDP----------#')
dist.init_process_group(backend='nccl', init_method='env://')
torch.cuda.manual_seed_all(config.seed)
config.local_rank = torch.distributed.get_rank()
print('#----------Preparing dataset----------#')
train_dataset = config.datasets(base_dir=config.data_path, list_dir=config.list_dir, split="train",
transform=transforms.Compose(
[RandomGenerator(output_size=[config.input_size_h, config.input_size_w])]))
train_sampler = DistributedSampler(train_dataset, shuffle=True) if config.distributed else None
train_loader = DataLoader(train_dataset,
batch_size=config.batch_size//gpus_num if config.distributed else config.batch_size,
shuffle=(train_sampler is None),
pin_memory=True,
num_workers=config.num_workers,
sampler=train_sampler)
val_dataset = config.datasets(base_dir=config.volume_path, split="test_vol", list_dir=config.list_dir)
val_sampler = DistributedSampler(val_dataset, shuffle=False) if config.distributed else None
val_loader = DataLoader(val_dataset,
batch_size=1, # if config.distributed else config.batch_size,
shuffle=False,
pin_memory=True,
num_workers=config.num_workers,
sampler=val_sampler,
drop_last=True)
print('#----------Prepareing Models----------#')
model_cfg = config.model_config
if config.network == 'vmunet':
model = VMUNet(
num_classes=model_cfg['num_classes'],
input_channels=model_cfg['input_channels'],
depths=model_cfg['depths'],
depths_decoder=model_cfg['depths_decoder'],
drop_path_rate=model_cfg['drop_path_rate'],
load_ckpt_path=model_cfg['load_ckpt_path'],
)
model.load_from()
else: raise('Please prepare a right net!')
if config.distributed:
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).cuda()
model = DDP(model, device_ids=[config.local_rank], output_device=config.local_rank)
else:
model = torch.nn.DataParallel(model.cuda(), device_ids=gpu_ids, output_device=gpu_ids[0])
print('#----------Prepareing loss, opt, sch and amp----------#')
criterion = config.criterion
optimizer = get_optimizer(config, model)
scheduler = get_scheduler(config, optimizer)
scaler = GradScaler()
print('#----------Set other params----------#')
min_loss = 999
start_epoch = 1
min_epoch = 1
if os.path.exists(resume_model):
print('#----------Resume Model and Other params----------#')
checkpoint = torch.load(resume_model, map_location=torch.device('cpu'))
model.module.load_state_dict(checkpoint['model_state_dict'])
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
scheduler.load_state_dict(checkpoint['scheduler_state_dict'])
saved_epoch = checkpoint['epoch']
start_epoch += saved_epoch
min_loss, min_epoch, loss = checkpoint['min_loss'], checkpoint['min_epoch'], checkpoint['loss']
log_info = f'resuming model from {resume_model}. resume_epoch: {saved_epoch}, min_loss: {min_loss:.4f}, min_epoch: {min_epoch}, loss: {loss:.4f}'
logger.info(log_info)
print('#----------Training----------#')
for epoch in range(start_epoch, config.epochs + 1):
torch.cuda.empty_cache()
train_sampler.set_epoch(epoch) if config.distributed else None
loss = train_one_epoch(
train_loader,
model,
criterion,
optimizer,
scheduler,
epoch,
logger,
config,
scaler=scaler
)
if loss < min_loss:
torch.save(model.module.state_dict(), os.path.join(checkpoint_dir, 'best.pth'))
min_loss = loss
min_epoch = epoch
if epoch % config.val_interval == 0:
mean_dice, mean_hd95 = val_one_epoch(
val_dataset,
val_loader,
model,
epoch,
logger,
config,
test_save_path=outputs,
val_or_test=False
)
torch.save(
{
'epoch': epoch,
'min_loss': min_loss,
'min_epoch': min_epoch,
'loss': loss,
'model_state_dict': model.module.state_dict(),
'optimizer_state_dict': optimizer.state_dict(),
'scheduler_state_dict': scheduler.state_dict(),
}, os.path.join(checkpoint_dir, 'latest.pth'))
if os.path.exists(os.path.join(checkpoint_dir, 'best.pth')):
print('#----------Testing----------#')
best_weight = torch.load(config.work_dir + 'checkpoints/best.pth', map_location=torch.device('cpu'))
model.module.load_state_dict(best_weight)
mean_dice, mean_hd95 = val_one_epoch(
val_dataset,
val_loader,
model,
epoch,
logger,
config,
test_save_path=outputs,
val_or_test=True
)
os.rename(
os.path.join(checkpoint_dir, 'best.pth'),
os.path.join(checkpoint_dir,
f'best-epoch{min_epoch}-mean_dice{mean_dice:.4f}-mean_hd95{mean_hd95:.4f}.pth')
)
if __name__ == '__main__':
config = setting_config
main(config)