-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathtrain.py
More file actions
179 lines (155 loc) · 8.1 KB
/
train.py
File metadata and controls
179 lines (155 loc) · 8.1 KB
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
import torch
import torch.nn as nn
import torch.nn.functional as F
import matplotlib.pyplot as plt
import numpy as np
import scipy.io as sio
import load_data
import network
import os
import time
from skimage.measure import compare_psnr, compare_ssim, compare_nrmse
save_path = 'snapshot'
if not os.path.exists(save_path):
os.mkdir(save_path)
######### hyper-parameters
batch_size = 2
learning_rate = 1e-4
weight_decay = 1e-5
num_epoches = 90
test_interval = 10
num_chanel_Res = 16
num_blocks_Res = 9
num_chanel_Dense = 16
depth_Dense = 4
gps=3
blocks=19
Mask = sio.loadmat('Mask_8.mat')['Mask']
Mask = torch.Tensor(Mask)
train_path = "sliceT1.txt"
test_path = "T1_mat/Test.txt"
######### data_loader and network
train_loader = torch.utils.data.DataLoader(load_data.Train_List(train_path),batch_size=4,shuffle=True, num_workers=2)
test_loader = torch.utils.data.DataLoader(load_data.Test_List(test_path),batch_size=1,shuffle=True, num_workers=1,drop_last=True)
modelFFA_res_0 = network.FFA_res().cuda()
# modelFFA_res_0 = torch.load(save_path+'/modelFFA_res32_30_5.pk', map_location='cuda:0')
# modelFFA_res_1 = torch.load(save_path+'/modelFFA_res32_31_5.pk', map_location='cuda:0')
# modelFFA_res_2 = torch.load(save_path+'/modelFFA_res32_32_5.pk', map_location='cuda:0')
# modelFFA_res_3 = torch.load(save_path+'/modelFFA_res32_33_5.pk', map_location='cuda:0')
modelupdata_x2 = network.updata_x2(Mask).cuda()
######optimizer and loss
optimizerFFA_res_0 = torch.optim.Adam(modelFFA_res_0.parameters(),lr=learning_rate,weight_decay=weight_decay)
# optimizerFFA_res_1 = torch.optim.Adam(modelFFA_res_1.parameters(),lr=learning_rate,weight_decay=weight_decay)
# optimizerFFA_res_2 = torch.optim.Adam(modelFFA_res_2.parameters(),lr=learning_rate,weight_decay=weight_decay)
# optimizerFFA_res_3 = torch.optim.Adam(modelFFA_res_3.parameters(),lr=learning_rate,weight_decay=weight_decay)
loss = nn.MSELoss()
####### training FFA_res_net
for epoch in range(num_epoches):
###adjust learning_rate
if epoch % 70 == 0 or epoch % 80 == 0:
# for param_group in optimizerFFA_res_3.param_groups:
# param_group['lr'] *= 0.1
# for param_group in optimizerFFA_res_2.param_groups:
# param_group['lr'] *= 0.1
# for param_group in optimizerFFA_res_1.param_groups:
# param_group['lr'] *= 0.1
for param_group in optimizerFFA_res_0.param_groups:
param_group['lr'] *= 0.1
for i, (T1, T2, under_T2, under_T2_K) in enumerate(train_loader):
T1 = T1.view(-1, 1, 240, 240)
T1 = T1.cuda()
T2 = T2.view(-1, 1, 240, 240)
T2 = T2.cuda()
under_T2 = under_T2.view(-1, 1, 240, 240)
under_T2 = under_T2.cuda()
under_T2_K = under_T2_K.view(-1, 1, 240, 240, 2)
under_T2_K = under_T2_K.cuda()
T2_0 = modelFFA_res_0(T1, under_T2)
# z_1 = modelupdata_x2(T2_0, under_T2_K)
# T2_1 = modelFFA_res_1(T1, z_1)
# z_2 = modelupdata_x2(T2_1, under_T2_K)
# T2_2 = modelFFA_res_2(T1, z_2)
# z_3 = modelupdata_x2(T2_2, under_T2_K)
# T2_3 = modelFFA_res_2(T1, z_3)
# optimizerFFA_res_3.zero_grad()
# optimizerFFA_res_2.zero_grad()
# optimizerFFA_res_1.zero_grad()
optimizerFFA_res_0.zero_grad()
l_G = loss(T2_0, T2) # or T2_1 or T2_2 or T2_3
l_G.backward()
# optimizerFFA_res_3.step()
# optimizerFFA_res_2.step()
# optimizerFFA_res_1.step()
optimizerFFA_res_0.step()
if i % 10 == 0:
print('epoch:{:d},\tstep:{:d},\tloss:{:.4f}'.format(epoch,i,100*l_G.item()))
if epoch % test_interval == 0:
psnr_z = 0
nrmse_z = 0
psnr = 0
nrmse = 0
with torch.no_grad():
for j, (T1_test, T2_test, under_T2_test, under_T2_test_K) in enumerate(test_loader):
T1_test = T1_test.view(-1, 1, 240, 240, 115)
T1_test = T1_test.cuda()
T2_test = T2_test.view(-1, 1, 240, 240, 115)
T2_test = T2_test.cuda()
under_T2_test = under_T2_test.view(-1, 1, 240, 240, 115)
under_T2_test = under_T2_test.cuda()
under_T2_test_K = under_T2_test_K.view(-1, 1, 240, 240, 115, 2)
under_T2_test_K = under_T2_test_K.cuda()
T2_0_test = torch.zeros(1, 1, 240, 240, 115)
T2_0_test = T2_0_test.cuda()
z_1_test = torch.zeros(1, 1, 240, 240, 115)
z_1_test = z_1_test.cuda()
# T2_1_test = torch.zeros(1, 1, 240, 240, 115)
# T2_1_test = T2_1_test.cuda()
# z_2_test = torch.zeros(1, 1, 240, 240, 115)
# z_2_test = z_2_test.cuda()
# T2_2_test = torch.zeros(1, 1, 240, 240, 115)
# T2_2_test = T2_2_test.cuda()
# z_3_test = torch.zeros(1, 1, 240, 240, 115)
# z_3_test = z_3_test.cuda()
# T2_3_test = torch.zeros(1, 1, 240, 240, 115)
# T2_3_test = T2_3_test.cuda()
# z_4_test = torch.zeros(1, 1, 240, 240, 115)
# z_4_test = z_4_test.cuda()
for k in range(115):
T2_0_test[:, :, :, :, k] = modelFFA_res_0(T1_test[:, :, :, :, k], under_T2_test[:, :, :, :, k])
z_1_test[:, :, :, :, k] = modelupdata_x2(T2_0_test[:, :, :, :, k],
under_T2_test_K[:, :, :, :, k, :])
# T2_1_test[:, :, :, :, k] = modelFFA_res_1(T1_test[:, :, :, :, k], z_1_test[:, :, :, :, k])
# z_2_test[:, :, :, :, k] = modelupdata_x2(T2_1_test[:, :, :, :, k],
# under_T2_test_K[:, :, :, :, k, :])
# T2_2_test[:, :, :, :, k] = modelFFA_res_2(T1_test[:, :, :, :, k], z_2_test[:, :, :, :, k])
# z_3_test[:, :, :, :, k] = modelupdata_x2(T2_2_test[:, :, :, :, k],
# under_T2_test_K[:, :, :, :, k, :])
# T2_3_test[:, :, :, :, k] = modelFFA_res_3(T1_test[:, :, :, :, k], z_3_test[:, :, :, :, k])
# z_4_test[:, :, :, :, k] = modelupdata_x2(T2_3_test[:, :, :, :, k],
# under_T2_test_K[:, :, :, :, k, :])
# l_G_test +=loss(T2_0_test, T2_test)
z_1_test_ = z_1_test.cpu()
z_1_test_ = z_1_test_.data.numpy().squeeze()
T2_0_test_ = T2_0_test.cpu()
T2_0_test_ = T2_0_test_.data.numpy().squeeze()
T2_test_ = T2_test.cpu().numpy().squeeze()
psnr_z += compare_psnr(T2_test_, z_1_test_, data_range=T2_test_.max())
nrmse_z += compare_nrmse(T2_test_, z_1_test_)
psnr_single = compare_psnr(T2_test_, T2_0_test_, data_range=T2_test_.max())
nrmse_single = compare_nrmse(T2_test_, T2_0_test_)
print('epoch:{:d},\tstep:{:d},\ttest_psnr_single:{:.4f},\ttest_nrmse_single:{:.4f}'.format(epoch, j,
psnr_single.item(),
nrmse_single.item()))
psnr += psnr_single
nrmse += nrmse_single
psnr_z /= len(test_loader)
nrmse_z /= len(test_loader)
psnr /= len(test_loader)
nrmse /= len(test_loader)
# l_G_test /= len(test_loader)
print('epoch:{:d},\ttest_psnr:{:.4f},\ttest_nrmse:{:.4f}'.format(epoch, psnr.item(), nrmse.item()))
print('epoch:{:d},\ttest_psnrz:{:.4f},\ttest_nrmsez:{:.4f}'.format(epoch, psnr_z.item(), nrmse_z.item()))
torch.save(modelFFA_res_0, save_path + '/modelFFA_res_0_' + str(epoch) + '.pk')
# torch.save(modelFFA_res_1, save_path + '//modelFFA_res_1_' + str(epoch) + '.pk')
# torch.save(modelFFA_res_2, save_path + '//modelFFA_res_2_' + str(epoch) + '.pk')
# torch.save(modelFFA_res_3, save_path + '//modelFFA_res_3_' + str(epoch) + '.pk')