-
Notifications
You must be signed in to change notification settings - Fork 3
/
DeepSCI.py
188 lines (131 loc) · 7.99 KB
/
DeepSCI.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
# -*- coding: utf-8 -*-
"""
Created on Wed Dec 4 11:16:54 2019
@author: xinxi ss
"""
#%% 先加载进必要的文件
import tensorflow as tf
from PIL import Image
import matplotlib.pyplot as plt
import numpy as np
import model_Unet_1
import Myfftshift
tf.reset_default_graph()
#%%输入各种文件的路径
model_save_path = '.\\model_mnist_2\\model_mnist_2.ckpt'
#
# file = '.\\data\\val_imgs\\4360.png'
# ac_save_path = '.\\result\\DL\\' + file
ac_save_path = 'H:\\tzw\\val_imgs_auto\\4360.png'
result_save_path = '.\\result\\mnist_4000\\fine_tune_4360_with_110\\'
label_path = '.\\data\\val_labels_128\\4360.jpg'
auto_result_save_path = '.\\result\\mnist_4000\\fine_tune_4360_auto_with_110\\'
#ac_save_path = '.\\data\\nobacknoise_auto\\test1\\2.png'
#ac_save_path = '.\\data\\data_noise\\face_no_noise.png'
#resut_save_path = '.\\result\\DIP_TL\\data_noise\\'
dim = 128
img_W = 128 # 图像宽度
img_H = 128 # 图像高度
lab_W = 128 # 标签宽度
lab_H = 128 # 标签高度
batch_size = 1 # 每个mini-batch含有的样本数量
keep_prob = 1 # dropout的概率
test_num = 1 # 待测试的样本数量
pad_size = 256 # 将所得结果填充0为这个尺寸
def Measure_step(img_input):
img_input = tf.cast(img_input,dtype = tf.complex64)
FF = tf.fft2d(img_input)
f_mul = tf.multiply(tf.conj(FF),FF)
IM = tf.ifft2d(f_mul)
IM = Myfftshift.ifftshift(IM)
IM = tf.abs(IM)
return IM
with tf.variable_scope('input'):
x = tf.placeholder(tf.float32, shape=[batch_size,img_W,img_H,1],name = 'images')
y_label = tf.placeholder(tf.float32,shape=[batch_size,lab_W,lab_H,1],name = 'labels')
keep_prob = tf.placeholder(tf.float32,name = 'keep_prob') #设置droupout的控制参数
is_training = tf.placeholder(tf.bool,name = 'is_train') #设置BN的控制参数
#rand = tf.placeholder(shape=(1,img_W,img_H,1), dtype=tf.float32)
rand = tf.placeholder(tf.float32,shape = (1,img_W,img_H,1))
# rng = np.random.RandomState(824)
# inpt = tf.constant(rng.uniform(0, 0.1, size=(1,img_W,img_H,1)), dtype = tf.float32) + rand #输入是均匀随机分布的固定值
y_conv = model_Unet_1.inference(x, img_W, img_H, batch_size, is_training)
raw_measure = tf.reshape(x,[img_W,img_H])
raw_measure = (raw_measure - tf.reduce_min(raw_measure))/(tf.reduce_max(raw_measure) - tf.reduce_min(raw_measure))
#加支持域
out = tf.reshape(y_conv,[img_W,img_H])
mask = tf.ones([110,110],dtype=tf.float32)
mask = tf.pad(mask, [[int((dim - 110)/2),int((dim - 110)/2)],[int((dim - 110)/2),int((dim - 110)/2)]], mode='CONSTANT', name=None, constant_values=0)
# mask = plt.imread(path_4360)
# ss = tf.ones([128,128],dtype=tf.float32)
# ss1 = mask * ss
# ss2 = tf.zeros([128,128])
# out_pad = out * (ss - mask)
out_pad = out * mask
# out_pad = tf.pad(out, [[int((pad_size2 - dim)/2),int((pad_size2 - dim)/2)],[int((pad_size2 - dim)/2),int((pad_size2 - dim)/2)]], mode='CONSTANT', name=None, constant_values=0)
out_pad = tf.pad(out_pad, [[int((pad_size - dim)/2),int((pad_size - dim)/2)],[int((pad_size - dim)/2),int((pad_size - dim)/2)]], mode='CONSTANT', name=None, constant_values=0)
#对输出结果补零
# out = tf.reshape(y_conv,[img_W,img_H])
# out_pad = tf.pad(out, [[int((pad_size - dim)/2),int((pad_size - dim)/2)],[int((pad_size - dim)/2),int((pad_size - dim)/2)]], mode='CONSTANT', name=None, constant_values=0)
out_measure = Measure_step(out_pad) # 自相关
out_measure = tf.reshape(out_measure,[1,out_measure.shape[0],out_measure.shape[1],1])
out_measure = tf.image.resize_images(out_measure,[img_W,img_H],tf.image.ResizeMethod.BILINEAR) # 将自相关变为与实测自相关相同维度
out_measure = (out_measure - tf.reduce_min(out_measure))/(tf.reduce_max(out_measure) - tf.reduce_min(out_measure))
out_measure = tf.reshape(out_measure,[img_W,img_H])
loss = tf.losses.mean_squared_error(out_measure, raw_measure)
prediction_error = []
optimizer = tf.train.AdamOptimizer(learning_rate=0.0001)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
with tf.variable_scope('train_step'):
train_op = optimizer.minimize(loss)
init_op = (tf.local_variables_initializer(),tf.global_variables_initializer())
saver = tf.train.Saver()
with tf.Session() as sess:
# 神经网络训练准备工作。变量初始化，线程启动
sess.run(init_op)
# saver.restore(sess,model_save_path)
for i in range(test_num):
# img_ac = plt.imread(ac_save_path + str(i+11003) + '.png')
img_ac = plt.imread(ac_save_path)
img_label = plt.imread(label_path)
img_ac = np.reshape(img_ac,[1,img_W,img_H,1])
saver.restore(sess,model_save_path)
# sess.run(init_op)
for step in range(200):
# new_rand = np.random.uniform(0, 1.0/30.0, size=(1,img_W,img_H,1))
if step % 10 == 0:
# lossval = sess.run(E, feed_dict = {rand: new_rand, x:img_ac, keep_prob:1.0, is_training:True})
lossval,y_pred = sess.run([loss,out], feed_dict = {x:img_ac, keep_prob:1.0, is_training:True})
error = np.sum((img_label/np.max(img_label) - y_pred)**2)/(128*128)
pred_error = error
print('[step:%d measure loss: %f prediction_loss%f]'%(step,lossval,pred_error))
prediction_error.extend([pred_error])
# new_rand = np.random.uniform(0, 0.0/30.0, size=(1,img_W,img_H,1))
# image_out = sess.run(out, feed_dict = {rand: new_rand, x:img_ac, keep_prob:1.0, is_training:True}).reshape(img_W,img_H)
image_out = sess.run(out_pad, feed_dict = {x:img_ac, keep_prob:1.0, is_training:True})
image_out = 255*(image_out - np.min(image_out))/(np.max(image_out) - np.min(image_out))
image_out = Image.fromarray(image_out.astype('uint8')).convert('L')
image_out = image_out.crop((64,64,192,192))
image_out.save(result_save_path+'4360_%d.bmp'%(step))
if step % 10 == 0:
# new_rand = np.random.uniform(0, 0.0/30.0, size=(1,img_W,img_H,1))
# Ac_compute = sess.run(out_measure, feed_dict = {rand: new_rand, x:img_ac, keep_prob:1.0, is_training:True})
# Ac_measure = sess.run(raw_measure, feed_dict = {rand: new_rand, x:img_ac, keep_prob:1.0, is_training:True})
Ac_compute = sess.run(out_measure, feed_dict = {x:img_ac, keep_prob:1.0, is_training:True})
Ac_measure = sess.run(raw_measure, feed_dict = {x:img_ac, keep_prob:1.0, is_training:True})
auto_out = 255*(Ac_compute - np.min(Ac_compute))/(np.max(Ac_compute) - np.min(Ac_compute))
auto_out = Image.fromarray(auto_out.astype('uint8')).convert('L')
auto_out.save(auto_result_save_path+'4360_%d.png'%(step))
plt.subplot(141)
plt.imshow(Ac_measure)
plt.subplot(142)
plt.imshow(Ac_compute)
plt.subplot(143)
plt.imshow(image_out)
plt.subplot(144)
plt.imshow(img_label)
plt.title('step:%d'%(step))
plt.show()
# sess.run(train_op, feed_dict = {rand: new_rand, x:img_ac, keep_prob:1.0, is_training:True})
sess.run(train_op, feed_dict = {x:img_ac, keep_prob:1.0, is_training:True})