-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathnew_train.py
executable file
·158 lines (128 loc) · 5.72 KB
/
new_train.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
# python imports
import os
import glob
import sys
import random
import nibabel as nib
# third-party imports
import tensorflow as tf
import numpy as np
import scipy.io as sio
from keras.backend.tensorflow_backend import set_session
from keras.optimizers import Adam, RMSprop, SGD
from keras.models import load_model, Model
from keras.utils import multi_gpu_model
from keras.losses import mse, binary_crossentropy
# import datagenerators
import refnet
import losses
import DenseDeform
import InverseNet
# from datagenerators import example_gen
from measurment import *
from utility import *
import cv2
import nibabel as nib
import h5py
import pdb
import BatchDataReader
from itertools import permutations
from ConfigReader import *
def train(config_file):
config = Configuration()
config.CreateConfig(config_file)
config.PrintConfiguration()
vol_size = (config.patch_size, config.patch_size, config.patch_size)
batch_size = 4
train_patch_pairs = glob.glob(config.base_directory + "/train/*.h5")
random.shuffle(train_patch_pairs)
model_dir = config.base_directory + config.Model_name
if not os.path.isdir(model_dir):
os.mkdir(model_dir)
gpu = '/gpu:' + str(config.GPU)
os.environ["CUDA_VISIBLE_DEVICES"] = str(config.GPU)
configurnet = tf.ConfigProto()
configurnet.gpu_options.allow_growth = True
configurnet.gpu_options.per_process_gpu_memory_fraction = 0.9
configurnet.allow_soft_placement = True
set_session(tf.Session(config=configurnet))
with tf.device(gpu):
if config.ModelType == 'VM':
nf_enc = [16, 32, 32, 32, 16]
nf_dec = [32, 32, 32, 32, 8]
model = refnet.ref_net((64, 64, 64), nf_enc, nf_dec)
if config.ModelType == 'DDN':
model = DenseDeform.DenseDeformNet(vol_size, config.HalfNet, config.BN, config.Dilation)
model = model.createModel()
if config.ModelType == 'Inverse':
model = InverseNet.InverseNet(vol_size, config.HalfNet, config.BN, config.Dilation)
model = model.createModel()
'''
else:
model = refnet.ref_net(vol_size,nf_enc,nf_dec)
'''
# sgd = SGD(lr=lr, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=Adam(lr=config.Learning_Rate), loss=[losses.cc3D(), losses.gradientLoss('l2')],
loss_weights=[1.0, 1.5])
# model.metrics_tensors = [model.layers[2].output[0,:,:,:,0],model.layers[2].output[0,:,:,:,1],model.layers[166].output[0,:,:,:,0],model.layers[170].output[0,:,:,:,:],
# model.layers[168].output[0,:,:,:,0],model.layers[167].output[0,:,:,:,0]]
# model.metrics_tensors = [model.layers[2].output[0,:,:,:,0],model.layers[2].output[0,:,:,:,1],model.layers[83].output[0,:,:,:,0],model.layers[82].output[0,:,:,:,:]]
if np.int(config.iteration_start) > 0:
model.load_weights(model_dir + '/' + str(config.iteration_start) + '.h5')
'''
input_patch_1 = np.zeros((batch_size,vol_size[0], vol_size[1], vol_size[2], 1),dtype=np.float)
input_patch_2 = np.zeros((batch_size,vol_size[0], vol_size[1], vol_size[2], 1),dtype=np.float)
output_patch = np.zeros((batch_size,vol_size[0], vol_size[1], vol_size[2], 1),dtype=np.float)
deformation = np.zeros((batch_size,vol_size[0], vol_size[1], vol_size[2], 3),dtype=np.float)
'''
inpx_fake = np.zeros((batch_size, vol_size[0], vol_size[1], vol_size[2], 1), dtype=np.float)
inpy_fake = np.zeros((batch_size, vol_size[0], vol_size[1], vol_size[2], 1), dtype=np.float)
f_flow = np.zeros((batch_size, vol_size[0], vol_size[1], vol_size[2], 3), dtype=np.float)
i_flow = np.zeros((batch_size, vol_size[0], vol_size[1], vol_size[2], 3), dtype=np.float)
start = config.iteration_start
iteration = 1734 / batch_size
# n_iteration=iteration*len(train_patch_pairs)
ep_st = 0
step_st = 0
paris_st = 0
if start > 0:
if start > iteration:
step_st = start - np.int((start / iteration)) * iteration
paris_st = np.int(start / np.float(iteration))
else:
step_st = start
# total_pairs = len(train_patch_pairs)*(len(train_patch_pairs)-1)
x = np.arange(len(train_patch_pairs))
total_pairs = list(permutations(x, 2))
random.shuffle(total_pairs)
for pairs in list(total_pairs):
src_im = h5py.File(train_patch_pairs[pairs[0]],"r")
src_im = src_im.get("moving").value
random.shuffle(src_im)
tgt_im = h5py.File(train_patch_pairs[pairs[1]],"r")
tgt_im = tgt_im.get("moving").value
random.shuffle(tgt_im)
print ("{0} Patches are selected".format(len(src_im)))
s = 0
for step in range(step_st, iteration):
src = src_im[s:s + batch_size, :, :, :]
src = np.reshape(src, (batch_size, vol_size[0], vol_size[1], vol_size[2], 1))
tgt = tgt_im[s:s + batch_size, :, :, :]
tgt = np.reshape(tgt, (batch_size, vol_size[0], vol_size[1], vol_size[2], 1))
train_loss = model.train_on_batch([src, tgt], [tgt, f_flow])
print(" Paris : " + str(pairs) + " Step :" + str(step) + " Patch :" + str(s) + " Loss: " + str(
train_loss[0]))
if not isinstance(train_loss, list):
train_loss = [train_loss]
if (start % config.Model_saver == 0):
model.save(model_dir + '/' + str(start) + '.h5')
if (start % 100 == 0):
sess = tf.keras.backend.get_session()
write_summary(start, sess, train_loss, model_dir)
s = s + batch_size
start = start + 1
if __name__ == "__main__":
if (len(sys.argv) != 2):
print "Configuration file name is required"
else:
train(sys.argv[1])