/
solver.py
284 lines (242 loc) · 11 KB
/
solver.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
import os
import numpy as np
import time
import datetime
import torch
import torchvision
from torch import optim
from torch.autograd import Variable
import torch.nn.functional as F
from evaluation import *
from network import R2U_Net,AttU_Net,R2AttU_Net,Structure_U_Net,Analyzer_U_Net
import csv
from tqdm import tqdm as tqdm
from PIL import Image
import cv2
from progress.bar import Bar
import math
class AverageMeter(object):
"""Computes and stores the average and current value"""
def __init__(self):
self.reset()
def reset(self):
self.val = 0
self.avg = 0
self.sum = 0
self.count = 0
def update(self, val, n=1):
self.val = val
self.sum += val * n
self.count += n
self.avg = self.sum / self.count
class Solver(object):
def __init__(self, config, train_loader, test_loader):
# Data loader
self.train_loader = train_loader
self.test_loader = test_loader
# Models
self.unet = None
self.optimizer = None
self.img_ch = config.img_ch
self.output_ch = config.output_ch
# Hyper-parameters
self.lr = config.lr
# Training settings
self.num_epochs = config.num_epochs
self.num_epochs_decay = config.num_epochs_decay
self.batch_size = config.batch_size
# Step size
self.log_step = config.log_step
# Path
self.model_path = config.model_path
self.result_path = config.result_path
self.mode = config.mode
self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
self.model_type = config.model_type
self.t = config.t
self.build_model()
def build_model(self):
"""Build generator and discriminator."""
if self.model_type =='U_Net':
self.unet = U_Net(img_ch=3,output_ch=1)
elif self.model_type =='R2U_Net':
self.unet = R2U_Net(img_ch=3,output_ch=1,t=self.t)
elif self.model_type =='AttU_Net':
self.unet = AttU_Net(img_ch=3,output_ch=1)
elif self.model_type == 'R2AttU_Net':
self.unet = R2AttU_Net(img_ch=3,output_ch=1,t=self.t)
elif self.model_type =='ASM_U_Net':
self.unet =Structure_U_Net(img_ch=3,output_ch=1)
self.analyzer =Analyzer_U_Net(img_ch=1,output_ch=1)
self.optimizer_unet = optim.Adam(params=list(self.unet.parameters()),
lr= self.lr,weight_decay=1e-5)
self.optimizer_analyzer = optim.Adam(params=list(self.analyzer.parameters()),
lr=self.lr*0.2,weight_decay=1e-5)
self.unet.to(self.device)
self.unet = torch.nn.DataParallel(self.unet)
self.analyzer.to(self.device)
self.analyzer = torch.nn.DataParallel(self.analyzer)
# self.print_network(self.unet, self.model_type)
def print_network(self, model, name):
"""Print out the network information."""
num_params = 0
for p in model.parameters():
num_params += p.numel()
print(model)
print(name)
print("The number of parameters: {}".format(num_params))
def to_data(self, x):
"""Convert variable to tensor."""
if torch.cuda.is_available():
x = x.cpu()
return x.data
def update_lr(self, g_lr, d_lr):
for param_group in self.optimizer_unet.param_groups:
param_group['lr'] = lr
for param_group in self.optimizer_analyzer.param_groups:
param_group['lr'] = lr
def reset_unet_grad(self):
"""Zero the gradient buffers."""
self.unet.zero_grad()
def reset_analyzer_grad(self):
"""Zero the gradient buffers."""
self.analyzer.zero_grad()
def tensor2img(self,x):
img = (x[:,0,:,:]>x[:,1,:,:]).float()
img = img*255
return img
def regular_loss(self,pred,gt):
b,c,h,w=pred.size()
row_std=torch.std(pred,dim=3)
row_mean=torch.mean(pred,dim=3)
thres_upper=(row_mean+row_std*2).view(b,c,h,1).expand_as(gt)
thres_lower=(row_mean-row_std*2).view(b,c,h,1).expand_as(gt)
idx_upper=torch.gt(pred,thres_upper)
idx_lower=torch.gt(thres_lower,pred)
loss=torch.zeros_like(pred)
loss[idx_upper]=torch.pow((pred-thres_upper),2)[idx_upper]
loss[idx_lower]=torch.pow((thres_lower-pred),2)[idx_lower]
return loss.mean()
def train(self):
"""Train encoder, generator and discriminator."""
#====================================== Training ===========================================#
#===========================================================================================#
unet_path = os.path.join(self.model_path, 'S_%s-%d.pth' %(self.model_type,self.num_epochs))
analyzer_path = os.path.join(self.model_path, 'A_%s-%d.pth' %(self.model_type,self.num_epochs))
# U-Net Train
# Train for Encoder
lr = self.lr
best_mse=1000.0
start_epoch=0
if os.path.isfile(unet_path):
# Load the pretrained Encoder
self.unet.load_state_dict(torch.load(unet_path))
self.analyzer.load_state_dict(torch.load(analyzer_path))
print("Load from pretrained model!")
start_epoch=42
for epoch in range(start_epoch,self.num_epochs):
w=math.pow(10,-epoch/self.num_epochs)
self.unet.train(True)
self.analyzer.train(True)
epoch_loss = 0
epoch_asm_loss = 0
length = 0
length_asm=0
structure_losses = AverageMeter()
bar = Bar('Training', max=len(self.train_loader))
for batch_idx, (data) in enumerate(self.train_loader):
# GT : Ground Truth
image = data['image'].to(self.device)
pmap = data['densitymap'].to(self.device)
for step_idx in range(4):
self.reset_unet_grad()
structure_loss=0.
# SR : Segmentation Result
pmap_pred = self.unet(image)
pmap_feat,pmap_rec=self.analyzer(pmap)
pmap_pred_feat,pmap_pred_rec=self.analyzer(pmap_pred)
for feat_idx in range(len(pmap_pred_feat)):
structure_loss +=F.mse_loss(pmap_pred_feat[feat_idx],pmap_feat[feat_idx])
structure_loss/=len(pmap_pred_feat)
#structure_loss+=F.mse_loss(pmap_pred,pmap)
epoch_loss += structure_loss.item()
structure_losses.update(structure_loss.item(), image.size(0))
# Backprop + optimize
structure_loss.backward()
length += image.size(0)
self.optimizer_unet.step()
bar.suffix = '({batch}/{size}) | Total: {total:} | ETA: {eta:} | S Loss: {loss:.6f} |'.format(
batch=batch_idx + 1,
size=len(self.train_loader),
total=bar.elapsed_td,
eta=bar.eta_td,
loss=structure_losses.avg
)
bar.next()
self.reset_analyzer_grad()
analyzer_loss=0.
pmap_pred = self.unet(image)
pmap_feat,pmap_rec=self.analyzer(pmap)
pmap_pred_feat,pmap_pred_rec=self.analyzer(pmap_pred)
for feat_idx in range(len(pmap_pred_feat)):
analyzer_loss +=((-0.2)*F.mse_loss(pmap_pred_feat[feat_idx],pmap_feat[feat_idx]))
analyzer_loss/=len(pmap_pred_feat)
analyzer_loss+=10*(F.mse_loss(pmap_rec,pmap))
epoch_asm_loss += analyzer_loss.item()
analyzer_loss.backward()
self.optimizer_analyzer.step()
length_asm += image.size(0)
epoch_loss/=length
epoch_asm_loss/=length_asm
bar.finish()
# Print the log info
print('Epoch [%d/%d], Loss: %.6f, ASM Loss: %.6f' % (epoch+1, self.num_epochs, epoch_loss,epoch_asm_loss))
# Decay learning rate
if (epoch+1) in self.num_epochs_decay:
for param_group in self.optimizer_analyzer.param_groups:
param_group['lr'] = self.lr*0.02
for param_group in self.optimizer_unet.param_groups:
param_group['lr'] = self.lr*0.1
self.lr=self.lr*0.1
print ('Decay learning rate to lr: {}.'.format(self.lr))
#===================================== Validation ====================================#
self.unet.train(False)
self.unet.eval()
length=0
epoch_loss = 0.
for i, (data) in tqdm(enumerate(self.test_loader)):
image = data['image'].to(self.device)
pmap = data['densitymap'].to(self.device)
# SR : Segmentation Result
pmap_pred = self.unet(image)
loss =F.mse_loss(pmap_pred,pmap)
epoch_loss+=loss.item()
pmap_pred=pmap_pred.detach().cpu().numpy()[0][0]
# pmap_pred[pmap_pred>1]=1
# pmap_pred[pmap_pred<0]=0
pmap_pred=np.asarray(pmap_pred*255, dtype=np.uint8)
feature_img = cv2.applyColorMap(pmap_pred, cv2.COLORMAP_JET)
cv2.imwrite("result/%s/%d.png"%(self.model_type,i), feature_img)
length += image.size(0)
epoch_loss/=length
print('[Validation] Loss: %.6f'%(epoch_loss))
'''
torchvision.utils.save_image(images.data.cpu(),
os.path.join(self.result_path,
'%s_valid_%d_image.png'%(self.model_type,epoch+1)))
torchvision.utils.save_image(SR.data.cpu(),
os.path.join(self.result_path,
'%s_valid_%d_SR.png'%(self.model_type,epoch+1)))
torchvision.utils.save_image(GT.data.cpu(),
os.path.join(self.result_path,
'%s_valid_%d_GT.png'%(self.model_type,epoch+1)))
'''
# Save Best U-Net model
if epoch_loss < best_mse:
best_mse = epoch_loss
best_epoch = epoch
best_unet = self.unet.state_dict()
best_analyzer=self.analyzer.state_dict()
print('Best %s model score : %.4f'%(self.model_type,best_mse))
torch.save(best_unet,unet_path)
torch.save(best_analyzer,analyzer_path)