-
Notifications
You must be signed in to change notification settings - Fork 11
/
Copy pathutils.py
257 lines (242 loc) · 11.9 KB
/
utils.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
import json
import re
from datetime import datetime
from pathlib import Path
import random
import numpy as np
import torch
import torchvision
def save_weights(model, model_path, ep, step, train_metrics, valid_metrics):
torch.save({'model': model.state_dict(), 'epoch': ep, 'step': step, 'valid_loss': valid_metrics['loss1'], 'train_loss': train_metrics['loss1']},
str(model_path)
)
def write_tensorboard(writer, model, epoch, train_metrics, valid_metrics):
## epoch starts from 1, not 0
## save the results on epoch 1
if (epoch-1) % 5 == 0:
train_image, train_mask, train_prob = train_metrics['image'], train_metrics['mask'], train_metrics['prob']
valid_image, valid_mask, valid_prob = valid_metrics['image'], valid_metrics['mask'], valid_metrics['prob']
saved_images = torchvision.utils.make_grid(train_image, nrow=train_image.shape[0], padding=10, pad_value=1)
writer.add_image('train/Image', saved_images, epoch)
######
for n in range(train_mask.shape[1]):
saved_images = torch.cat((train_mask.narrow(1,n,1), train_prob.narrow(1, n, 1)), 0)
saved_images = torchvision.utils.make_grid(saved_images, nrow=train_image.shape[0], padding=10, pad_value=1)
writer.add_image('train/Mask%s' % n, saved_images, epoch)
##### test
saved_images = torchvision.utils.make_grid(valid_image, nrow=valid_image.shape[0], padding=10, pad_value=1)
writer.add_image('test/Image', saved_images, epoch)
######
######
for n in range(valid_mask.shape[1]):
saved_images = torch.cat((valid_mask.narrow(1, n, 1), valid_prob.narrow(1, n, 1)), 0)
saved_images = torchvision.utils.make_grid(saved_images, nrow=valid_image.shape[0], padding=10, pad_value=1)
writer.add_image('test/Mask%s' % n, saved_images, epoch)
######
for name, param in model.named_parameters():
writer.add_histogram(name, param.clone().cpu().data.numpy(), epoch)
if (epoch-1) % 1 == 0:
#####
valid_loss, valid_loss1, valid_loss2, valid_loss3 = valid_metrics['loss'],valid_metrics['loss1'],valid_metrics['loss2'],valid_metrics['loss3']
train_loss, train_loss1, train_loss2, train_loss3 = train_metrics['loss'],train_metrics['loss1'],train_metrics['loss2'],train_metrics['loss3']
train_jaccard, valid_jaccard = train_metrics['jaccard'],valid_metrics['jaccard']
writer.add_scalars('loss', {'train': train_loss, 'test': valid_loss}, epoch)
writer.add_scalars('loss1',{'train': train_loss1, 'test': valid_loss1},epoch)
writer.add_scalars('loss2', {'train': train_loss2, 'test': valid_loss2}, epoch)
writer.add_scalars('loss3', {'train': train_loss3, 'test': valid_loss3}, epoch)
for out in range(1,3):
for auc in range(1,6):
key = 'out{}auc{}'.format(str(out), str(auc))
writer.add_scalars(key, {'train': train_metrics[key],'test': valid_metrics[key]}, epoch)
writer.add_scalars('jaccard', {'train': train_jaccard,'test': valid_jaccard}, epoch)
#####
def write_event(log, step, epoch, train_metrics, valid_metrics):
# train_metrics['step'] = step
# train_metrics['epoch'] = epoch
# train_metrics['valid_metrics'] = step
# train_metrics['valid_metrics'] = epoch
# train_metrics['dt'] = datetime.now().isoformat()
# valid_metrics['dt'] = datetime.now().isoformat()
# log.write(json.dumps(train_metrics, sort_keys=True))
# log.write(json.dumps(valid_metrics, sort_keys=True))
# log.write('\n')
# log.flush()
#print(data['loss'])
CMD = 'epoch:{} step:{} time:{:.2f} train_loss:{:.3f} {:.3f} {:.3f} {:.3f} train_auc1:{} {} {} {} {} train_auc2:{} {} {} {} {} train_jaccard:{:.3f} {:.3f} {:.3f} {:.3f} {:.3f} {:.3f} \n valid_loss:{:.3f} {:.3f} {:.3f} {:.3f} valid_auc1:{} {} {} {} {} valid_auc2:{} {} {} {} {} valid_jaccard:{:.3f} {:.3f} {:.3f} {:.3f} {:.3f} {:.3f}'.format(
epoch, step, train_metrics['epoch_time'],
train_metrics['loss'],train_metrics['loss1'],train_metrics['loss2'],train_metrics['loss3'],
train_metrics['out1auc1'],train_metrics['out1auc2'],train_metrics['out1auc3'],train_metrics['out1auc4'],train_metrics['out1auc5'],
train_metrics['out2auc1'],train_metrics['out2auc2'],train_metrics['out2auc3'],train_metrics['out2auc4'],train_metrics['out2auc5'],
train_metrics['jaccard'],train_metrics['jaccard1'],train_metrics['jaccard2'],train_metrics['jaccard3'],train_metrics['jaccard4'],train_metrics['jaccard5'],
valid_metrics['loss'], valid_metrics['loss1'], valid_metrics['loss2'], valid_metrics['loss3'],
valid_metrics['out1auc1'], valid_metrics['out1auc2'], valid_metrics['out1auc3'], valid_metrics['out1auc4'],
valid_metrics['out1auc5'],
valid_metrics['out2auc1'], valid_metrics['out2auc2'], valid_metrics['out2auc3'], valid_metrics['out2auc4'],
valid_metrics['out2auc5'],
valid_metrics['jaccard'],valid_metrics['jaccard1'],valid_metrics['jaccard2'],valid_metrics['jaccard3'],valid_metrics['jaccard4'],valid_metrics['jaccard5'],
)
print(CMD)
log.write(json.dumps(CMD))
# log.write(json.dumps(valid_metrics, sort_keys=True))
log.write('\n')
log.flush()
#for keys in sorted(train_metrics):
# print('{}:{}'.format(keys,train_metrics[keys]))
def print_model_summay(model):
for name, para in model.named_parameters():
print(name, para.numel(), para.requires_grad)
return None
def set_freeze_layers(model, freeze_layer_names=None):
for name, para in model.named_parameters():
#print(name, para.numel(), para.requires_grad)
if freeze_layer_names is None:
para.requires_grad = True
else:
if name in freeze_layer_names:
print('Freeze layer->', name)
para.requires_grad = False
else:
para.requires_grad = True
## no need to return the model
## model has been modified within the function
## pass by reference
#return model
def set_train_layers(model, train_layer_names=None):
for name, para in model.named_parameters():
#print(name, para.numel(), para.requires_grad)
if train_layer_names is None:
para.requires_grad = False
else:
if name in train_layer_names:
print('Train layer ->', name)
para.requires_grad = True
else:
para.requires_grad = False
## no need to return the model
## model has been modified within the function
## pass by reference
#return model
def get_freeze_layer_names(model,part):
#
# if part == 'encoder':
# layers = {
# 'module.encoder.0.weight':False,
# 'module.encoder.0.bias':False,
# 'module.encoder.2.weight':False,
# 'module.encoder.2.bias':False,
# 'module.encoder.5.weight':False,
# 'module.encoder.5.bias':False,
# 'module.encoder.7.weight':False,
# 'module.encoder.7.bias':False,
# 'module.encoder.10.weight':False,
# 'module.encoder.10.bias':False,
# 'module.encoder.12.weight':False,
# 'module.encoder.12.bias':False,
# 'module.encoder.14.weight':False,
# 'module.encoder.14.bias':False,
# 'module.encoder.17.weight':False,
# 'module.encoder.17.bias':False,
# 'module.encoder.19.weight':False,
# 'module.encoder.19.bias':False,
# 'module.encoder.21.weight':False,
# 'module.encoder.21.bias':False,
# 'module.encoder.24.weight':False,
# 'module.encoder.24.bias':False,
# 'module.encoder.26.weight':False,
# 'module.encoder.26.bias':False,
# 'module.encoder.28.weight':False,
# 'module.encoder.28.bias':False,
# 'module.center.block.0.conv.weight':True,
# 'module.center.block.0.conv.bias':True,
# 'module.center.block.1.weight':True,
# 'module.center.block.1.bias':True,
# 'module.center_Conv2d.weight':True,
# 'module.center_Conv2d.bias':True,
# 'module.dec5.block.0.conv.weight':True,
# 'module.dec5.block.0.conv.bias':True,
# 'module.dec5.block.1.weight':True,
# 'module.dec5.block.1.bias':True,
# 'module.dec4.block.0.conv.weight':True,
# 'module.dec4.block.0.conv.bias':True,
# 'module.dec4.block.1.weight':True,
# 'module.dec4.block.1.bias':True,
# 'module.dec3.block.0.conv.weight':True,
# 'module.dec3.block.0.conv.bias':True,
# 'module.dec3.block.1.weight':True,
# 'module.dec3.block.1.bias':True,
# 'module.dec2.block.0.conv.weight':True,
# 'module.dec2.block.0.conv.bias':True,
# 'module.dec2.block.1.weight':True,
# 'module.dec2.block.1.bias':True,
# 'module.dec1.conv.weight':True,
# 'module.dec1.conv.bias':True,
# 'module.final.weight':True,
# 'module.final.bias':True
# }
# if part == 'decoder':
# layers = {
# 'module.encoder.0.weight': True,
# 'module.encoder.0.bias': True,
# 'module.encoder.2.weight': True,
# 'module.encoder.2.bias': True,
# 'module.encoder.5.weight': True,
# 'module.encoder.5.bias': True,
# 'module.encoder.7.weight': True,
# 'module.encoder.7.bias': True,
# 'module.encoder.10.weight': True,
# 'module.encoder.10.bias': True,
# 'module.encoder.12.weight': True,
# 'module.encoder.12.bias': True,
# 'module.encoder.14.weight': True,
# 'module.encoder.14.bias': True,
# 'module.encoder.17.weight': True,
# 'module.encoder.17.bias': True,
# 'module.encoder.19.weight': True,
# 'module.encoder.19.bias': True,
# 'module.encoder.21.weight': True,
# 'module.encoder.21.bias': True,
# 'module.encoder.24.weight': True,
# 'module.encoder.24.bias': True,
# 'module.encoder.26.weight': True,
# 'module.encoder.26.bias': True,
# 'module.encoder.28.weight': True,
# 'module.encoder.28.bias': True,
# 'module.center.block.0.conv.weight': True,
# 'module.center.block.0.conv.bias': True,
# 'module.center.block.1.weight': True,
# 'module.center.block.1.bias': True,
# 'module.center_Conv2d.weight': True,
# 'module.center_Conv2d.bias': True,
# 'module.dec5.block.0.conv.weight': False,
# 'module.dec5.block.0.conv.bias': False,
# 'module.dec5.block.1.weight': False,
# 'module.dec5.block.1.bias': False,
# 'module.dec4.block.0.conv.weight': False,
# 'module.dec4.block.0.conv.bias': False,
# 'module.dec4.block.1.weight': False,
# 'module.dec4.block.1.bias': False,
# 'module.dec3.block.0.conv.weight': False,
# 'module.dec3.block.0.conv.bias': False,
# 'module.dec3.block.1.weight': False,
# 'module.dec3.block.1.bias': False,
# 'module.dec2.block.0.conv.weight': False,
# 'module.dec2.block.0.conv.bias': False,
# 'module.dec2.block.1.weight': False,
# 'module.dec2.block.1.bias': False,
# 'module.dec1.conv.weight': False,
# 'module.dec1.conv.bias': False,
# 'module.final.weight': False,
# 'module.final.bias': False
# }
freeze_layers = []
for ind, (name, para) in enumerate(model.named_parameters()):
if re.search(part, name):
#print(ind, name, para.numel(), para.requires_grad)
freeze_layers.append(name)
if part == 'encoder':
if re.search('center', name):
freeze_layers.append(name)
# for ly in layers:
# if not layers[ly]:
# freeze_layers.append(ly)
return freeze_layers