-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy paths_S2S_cnnTrainS2s0.9Blur1x1.py
158 lines (136 loc) · 5.22 KB
/
s_S2S_cnnTrainS2s0.9Blur1x1.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
# s_DtiNet_cnnTrain.py
#
#
# QT 2019
# %% load modual
from gc import callbacks
import os
import scipy.io as sio
import numpy as np
import nibabel as nb
import glob
from matplotlib import pyplot as plt
import tensorflow as tf
import nibabel as nib
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.callbacks import ModelCheckpoint
import qtlib as qtlib
from s2s_unet1x1 import unet_3d_model
# # Currently, memory growth needs to be the same across GPUs
# %% set up path
dpRoot = os.path.dirname(os.path.abspath('s_S2S_cnnTrainS2s0.9Blur1x1.py'))
os.chdir(dpRoot)
# %% subjects
subjects = sorted(glob.glob(os.path.join(dpRoot, 'mwu*')))
#for ii in np.arange(0, 5):
for ii in np.arange(0, 1):
sj = os.path.basename(subjects[ii])
print(sj)
dpSub = os.path.join(dpRoot, sj);
# %% set up model
nfilter = 64
nin = 1
nout = 1
dtnet = unet_3d_model(nin, nout, filter_num=nfilter)
dtnet.summary()
adam_opt = Adam(lr=0.00001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
dtnet.compile(loss = qtlib.mean_squared_error_weighted, optimizer = adam_opt)
pp = 0.9
cnnname = 'unet-sw1-blur0.91x1-flip'
dpSave = os.path.join(dpSub, cnnname)
if not os.path.exists(dpSave):
os.mkdir(dpSave)
print('create directory')
# %% load data
# fpImg = os.path.join(dpSub, 't1w-norm', sj + '_wave_norm.nii.gz');
# fpMask = os.path.join(dpSub, 'wave-bmask-fs', 'brainmask.nii.gz');
# fpImgBlur = os.path.join(dpSub, 'wave-blur', sj + '_t1w_blurnorm.nii.gz');
fpImg = os.path.join(dpSub,'t1w_sim0.3', 't1w_sim_rep1.nii.gz')
fpMask = os.path.join(dpSub, 't1w','brainmask_fs_dil2.nii.gz')
img = nb.load(fpImg).get_data()
img = np.expand_dims(img, 3)
# imgblur = nb.load(fpImgBlur).get_data()
# imgblur = np.expand_dims(imgblur, 3)
mask = nb.load(fpMask).get_data()
if len(mask.shape)==3:
mask = np.expand_dims(mask, 3)
else:
pass
if len(img.shape)==3:
img = np.expand_dims(img, 3)
else:
pass
imgnorm, _ = qtlib.normalize_image(img,img,mask)
imgnorm = imgnorm * mask
imgblurnorm,kernel = qtlib.blur_image(imgnorm)
imgblurnorm = imgblurnorm*mask
# # ii,jj,kk=150,125,200
# print(imgblurnorm[ii,jj,kk])
# print("--------")
# print(imgnorm[ii+1,jj,kk])
# print(imgnorm[ii-1,jj,kk])
# print(imgnorm[ii,jj+1,kk])
# print(imgnorm[ii,jj-1,kk])
# print(imgnorm[ii,jj,kk+1])
# print(imgnorm[ii,jj,kk-1])
# numbers = [imgnorm[ii+1,jj,kk],imgnorm[ii-1,jj,kk],imgnorm[ii,jj+1,kk],imgnorm[ii,jj-1,kk],imgnorm[ii,jj,kk+1],imgnorm[ii,jj,kk-1]]
# a=0
# for item in numbers:
# a = a+item
# print(a/6)
# %%
# fpBind = os.path.join(dpSub, 'bind-sw1', sj + '_bind_b96.mat')
# bind = sio.loadmat(fpBind)['bind'] - 1
bind, _ = qtlib.block_ind(mask,sz_block=96)
idx_start = 0
idx_step = 17
DataGen = qtlib.maskimageloader(imgnorm,mask,imgblurnorm,shuffle=True)
# idx_step = 2
for jj in np.arange(0, 100):
DataGen.randommask(pp,bind,idx_step,flip=False)
DataGen.index_shuffle()
# %% training
nbatch = 1
fnCp = cnnname + '_ep' + np.str(jj)
fpCp = os.path.join(dpSave, fnCp + '.h5')
checkpoint = ModelCheckpoint(fpCp, monitor='val_loss', save_best_only = True)
# history = dtnet.fit(x = [img_block_all, mask_block_all],
# y = imgres_block_all,
# batch_size = nbatch,
# validation_split=0.2,
# epochs = 1,
# callbacks = [checkpoint],
# verbose = 1,
# shuffle = True)
history = dtnet.fit(DataGen.generate_samples("train",1),
validation_data = DataGen.generate_samples("valid",1),
epochs=1,
callbacks = [checkpoint],
max_queue_size=10,
verbose = 1)
# save loos
fpLoss = os.path.join(dpSave, fnCp + '.mat')
sio.savemat(fpLoss, {'loss_train':history.history['loss'], 'loss_val':history.history['val_loss']})
if jj >= 5: # delete intermediate checkout files, which are large
fnCp = cnnname + '_ep' + np.str(jj - 5)
fpCp = os.path.join(dpSave, fnCp + '.h5')
os.remove(fpCp)
# %% check data
#plt.imshow(img_block_all[5, :, :, 40, 0], clim=(-2., 2.), cmap='gray')
#plt.imshow(imgres_block_all[10, :, :, 40, 0], clim=(0., 1000.), cmap='gray')
#plt.imshow(imgres_block_all[100, :, :, 40, 0] + img_block_all[100, :, :, 40, 0], clim=(0., 1000.), cmap='gray')
#
#plt.imshow(imgres_block_all[100, :, :, 40, 1], clim=(0., 1.), cmap='gray')
#plt.imshow(1. - mask_block_all[100, :, :, 40, 0] + imgres_block_all[100, :, :, 40, 1], clim=(0., 1.), cmap='gray')
#
#
#
#
#tmp = nib.Nifti1Image(img_block_all, np.eye(4))
#nib.save(tmp, 'img_block_all.nii.gz')
#
#tmp = nib.Nifti1Image(imgres_block_all, np.eye(4))
#nib.save(tmp, 'imgres_block_all.nii.gz')
#
#tmp = nib.Nifti1Image(mask_block_all, np.eye(4))
#nib.save(tmp, 'mask_block_all.nii.gz')