-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_new.py
272 lines (210 loc) · 7.46 KB
/
train_new.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
import torch
import albumentations as A
from albumentations.pytorch import ToTensorV2
from tqdm import tqdm
import time
import torch.nn as nn
import numpy as np
import torch.optim as optim
from PottsMGNet_model import POTTSNET
from utils_new_testL import (
load_checkpoint,
save_checkpoint,
get_loaders,
check_accuracy,
save_predictions_as_imgs,
save_compare_as_imgs,
)
fname='taiv4varyingNoise'
taskNo='2654'
learning_rate=1e-4
device="cuda" if torch.cuda.is_available() else "cpu"
# hyper parameters
batch_size=8
num_epochs=500
num_workers=0
image_height=128
image_width=192
trainL=800 # training data size, use -1 for all data
testL=200 # training data size, use -1 for all data
pin_memory=False
load_model=False
train_dir="../../data/train" # dictionary of training image
train_maskdir="../../data/train_mask" # dictionary of training mask
val_dir="../../data/test" # dictionary of testing image
val_maskdir="../../data/test_mask" # dictionary of testing mask
#
class pp:
in_channels=4
mid_channels=[32, 32, 64, 128, 256]
times_list=[3, 3, 3, 5, 5]
tau=0.5 # times step size
cnsts=[1., 40.] # [epsilon* tau, lambda/epsilon]. The result can be sensitive to epsilon* tau. The larger it is, the smoother the result is. Set it between 0.7 and 1 gives good results.
num_blocks=4 # number of blocks
epsilon=cnsts[0]/tau
lambdaa=epsilon*cnsts[1] # weight of the length penalty term
sigma=0.5 # sd of the Gaussian kernel
iter_num=1 # number of fixed iteration
kernel_size_bound=5 # largest kernel size allowed
connect=True # True if use skip-connections between encoder and decoder
BNLearn=True # True if learn paramters in batch normalization
device="cuda" if torch.cuda.is_available() else "cpu"
alpha=1 # relaxation ration in fixed point iteration. Changing this parameter may lead unstability of the method.
lambdaLearn=False #True if we allow lambda to be learnable
timevarying=True # True if we allow wegiths are different among blocks
tau_explicit=True # True if we explicitly use time evolution. False if absorb linear terms into convolution
args=pp
def train_fn(loader,model,optimizer,loss_fn):
# loop=tqdm(loader)
start=time.time()
losses=[]
nums=[]
for batch_idx, (data,targets) in enumerate(loader):
data=data.to(device=device)
targets=targets.float().unsqueeze(1).to(device=device)
predictions=model(data)
loss=loss_fn(predictions,targets)
losses.append(loss.item())
nums.append(len(data))
# backward
optimizer.zero_grad()
loss.backward()
optimizer.step()
end=time.time()
tt=end-start
total=np.sum(nums)
avg_loss=np.sum(np.multiply(losses,nums))/total
return tt, avg_loss
def main():
print('filename= ', fname)
print('mid_channels= ',args.mid_channels)
print('times_list= ',args.times_list)
print('cnsts=', args.cnsts)
print('lambdaa= ',args.cnsts[0]*args.cnsts[1]/args.tau)
print('tau= ',args.tau)
print('epsilon= ',args.cnsts[0]/args.tau)
print('sigma= ',args.sigma)
print('num_blocks= ',args.num_blocks)
print('M= ',args.M)
print('iter_num= ',args.iter_num)
print('skip_connect= ',args.connect)
print('BNLearn= ',args.BNLearn)
print('lambdaLearn= ',args.lambdaLearn)
print('timevarying= ',args.timevarying)
print('tau*eps= ', args.cnsts[0])
print('lambda/eps= ', args.cnsts[1])
print('tau_explicit= ', args.tau_explicit)
lossAll=[]
accAll=[]
diceAll=[]
traintime=[]
model=POTTSNET(args).to(device)
loss_fn=nn.BCELoss()
optimizer=optim.Adam(model.parameters(), lr=learning_rate)
if load_model:
checkpoint_old=torch.load(loadname)
model.load_state_dict(checkpoint_old["state_dict"])
optimizer.load_state_dict(checkpoint_old['optimizer'])
accAll=checkpoint_old['accAll']
lossAll=checkpoint_old['lossAll']
diceAll=checkpoint_old['diceAll']
traintime=checkpoint_old['traintime']
acc_old=checkpoint_old['acc_old']
dice_old=checkpoint_old['dice_old']
else:
acc_old=0
dice_old=0
print('acc_old= ',acc_old)
print('dice_old= ',dice_old)
loss=10.
epoch=0
# progressive training
for noise_sd in (0.,0.3,0.5,0.8,1.):
savename=fname+'_sd0'+str(int(noise_sd*10))+".pth.tar"
train_transform =A.Compose(
[
A.Resize(height=image_height, width=image_width),
A.HorizontalFlip(p=0.5),
A.VerticalFlip(p=0.5),
A.Normalize(
mean=[0.0, 0.0, 0.0],
std=[1.0, 1.0, 1.0],
max_pixel_value=1.0,
),
A.GaussNoise(var_limit=(0,(noise_sd)**2),always_apply=True,p=1),
ToTensorV2(),
])
val_transform=A.Compose(
[
A.Resize(height=image_height, width=image_width),
A.Normalize(
mean=[0.0, 0.0, 0.0],
std=[1.0, 1.0, 1.0],
max_pixel_value=1.0,
),
ToTensorV2(),
])
train_loader, val_loader=get_loaders(
train_dir,
train_maskdir,
val_dir,
val_maskdir,
batch_size,
train_transform,
val_transform,
trainL,
testL,
num_workers,
pin_memory,
)
for epoch in range(num_epochs):
tt,loss=train_fn(train_loader, model, optimizer, loss_fn)
lossAll.append(loss)
# check accuracy
acc,dice=check_accuracy(val_loader, model,device=device)
accAll.append(acc)
diceAll.append(dice)
traintime.append(tt)
flag=(acc>acc_old)
if flag:
acc_old=acc
dice_old=dice
print(
f"noise: {noise_sd}, Epoch {epoch}/{num_epochs}, loss: {loss:.4f}, dice: {dice:.4f}, acc: {acc:.2f}, **acc: {acc_old:.2f}, time used: {tt:.2f}s")
# save model
checkpoint={
"parameters":args,
"times_list":args.times_list,
"mid_channels":args.mid_channels,
"num_blocks": args.num_blocks,
"lambdaa": args.lambdaa,
"tau": args.tau,
"epsilon": args.epsilon,
# "M":args.M,
"iter_num":args.iter_num,
# "iter_num_len":args.iter_num_len,
"sigma":args.sigma,
"kernel_size_bound":args.kernel_size_bound,
"connect":args.connect,
"BNLearn":args.BNLearn,
"alpha":args.alpha,
"lambdaLearn":args.lambdaLearn,
"timevarying":args.timevarying,
"tau_explicit":args.tau_explicit,
"state_dict": model.state_dict(),
"optimizer": optimizer.state_dict(),
"lossAll": lossAll,
"accAll": accAll,
"diceAll": diceAll,
"acc_old": acc_old,
"dice_old": dice_old,
"traintime": traintime,
}
torch.save(checkpoint,savename)
###########print some examples
# if epoch%200==0:
# save_compare_as_imgs(
# loader=val_loader,model=model,batch_size=batch_size,folder="saved_images/",device=device,
# )
if __name__ == "__main__":
main()