-
Notifications
You must be signed in to change notification settings - Fork 3
/
predict_TBEFN.py
174 lines (128 loc) · 7.27 KB
/
predict_TBEFN.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
"""
# -*- coding:utf-8 -*-
# @Time : 2019/5/4 20:24
# @Author : KUN LU @ SXU
# @Email : [email protected]
# @Github : lukun199
"""
from __future__ import division
#import sys
import os, time
# import matplotlib.pyplot as plt
import tensorflow as tf
import tensorflow.contrib.slim as slim
import numpy as np
import glob
import cv2
#from skimage.measure import compare_ssim as ssim
#from skimage.measure import compare_psnr as psnr
checkpoint_dir = './ckpt/' # @lk199_sub1
input_dir = './input_dir/'
result_dir = './results/'
print(tf.__version__) # 1.13.1
def out_acti(x):
return tf.nn.relu(x)-tf.nn.relu(x-1.0)
def denoise_net(input, name): # denoising, dense res linkage
with tf.variable_scope(name):
conv1_out = slim.conv2d(input, 3, [3, 3], rate=1, activation_fn=None, scope='di_conv1')
conv2_in = conv1_out
conv2_out = slim.conv2d(conv2_in, 3, [3, 3], rate=1, activation_fn=None, scope='di_conv2')
conv3_in = conv1_out + conv2_out
conv3_out = slim.conv2d(conv3_in, 3, [3, 3], rate=1, activation_fn=None, scope='di_conv3')
conv4_in = conv3_in + conv3_out
conv4_out = slim.conv2d(conv4_in, 3, [3, 3], rate=1, activation_fn=None, scope='di_conv4')
conv5_in = conv4_in + conv4_out
conv5_out = slim.conv2d(conv5_in, 3, [3, 3], rate=1, activation_fn=None, scope='di_conv5')
return out_acti(input + conv5_out)
def upsample_and_concat(x1, x2, output_channels, in_channels): #x2和out+channel一致。目的：x1卷积之后尺寸变大，维数变小；和x2一致。得�?*x2的输出，32�?
with tf.variable_scope("us_vars"):
pool_size = 2
deconv_filter = tf.Variable(tf.truncated_normal([pool_size, pool_size, output_channels, in_channels], stddev=0.02))
deconv = tf.nn.conv2d_transpose(x1, deconv_filter, tf.shape(x2), strides=[1, pool_size, pool_size, 1])
deconv_output = tf.concat([deconv, x2], 3)
deconv_output.set_shape([None, None, None, output_channels * 2])
return deconv_output
def simple_unet(input,name): #input_1 forward ; input_2 details.
with tf.variable_scope(name):
conv_1 = slim.conv2d(input, 3, [3, 3], rate=1, activation_fn=None, scope='pp_conv1')
conv_2 = slim.conv2d(conv_1, 3, [3, 3], rate=1, activation_fn=None, scope='pp_conv2')
conv_3 = slim.conv2d(conv_2, 3, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='pp_conv3')
conv_4 = slim.conv2d(conv_3, 3, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='pp_conv4')
#fusion
fu_1 = tf.concat([input, conv_4], 3)
# fu_2 = slim.conv2d(fu_1, 3, [3, 3], rate=1, activation_fn=None, scope='fu_conv') #不加了，此时U_net输入是一个6channel的东
conv1 = slim.conv2d(fu_1, 16, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='u_conv1')
pool1 = slim.max_pool2d(conv1, [2, 2], padding='SAME')
conv2 = slim.conv2d(pool1, 32, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='u_conv2')
pool2 = slim.max_pool2d(conv2, [2, 2], padding='SAME')
conv3 = slim.conv2d(pool2, 64, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='u_conv3')
pool3 = slim.max_pool2d(conv3, [2, 2], padding='SAME')
conv4 = slim.conv2d(pool3, 128, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='u_conv4')
up5 = upsample_and_concat(conv4, conv3, 64, 128)
conv5 = slim.conv2d(up5, 64, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='u_conv5')
up6 = upsample_and_concat(conv5, conv2, 32, 64)
conv6 = slim.conv2d(up6, 32, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='u_conv6')
up7 = upsample_and_concat(conv6, conv1, 16, 32)
conv7 = slim.conv2d(up7, 16, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='u_conv7')
conv8 = slim.conv2d(conv7, 3, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='u_conv8') # modified lk199
return conv8
def fusion(input_1,input_2,name):
with tf.variable_scope(name):
fusion_in = tf.concat([input_1, input_2], 3)
out_1 = slim.conv2d(fusion_in, 16, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='fusion_1')
out_2 = slim.conv2d(out_1, 16, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='fusion_2')
out_3 = slim.conv2d(out_2, 3, [3, 3], rate=1, activation_fn=None, scope='fusion_3')
return out_3
def atten(input,name):
with tf.variable_scope(name):
out_1 = slim.conv2d(input, 16, [3, 3], rate=1, activation_fn=tf.nn.relu, scope='atten_1')
out_2 = slim.conv2d(out_1, 16, [3, 3], padding='SAME', rate=2, activation_fn=tf.nn.relu, scope='atten_2')
out_3 = slim.conv2d(out_2, 16, [3, 3], padding='SAME', rate=2, activation_fn=tf.nn.relu, scope='atten_3')
out_4 = slim.conv2d(out_3, 1, [3, 3], rate=1, activation_fn=None, scope='atten_4')
return out_acti(out_4)
def buildmodel(sample):
trans_fun_A_with_1E = simple_unet(sample, name='fun_est_A_with_1E')
enhanced_1E = out_acti(sample * trans_fun_A_with_1E)
denoised_in = denoise_net(sample, name='denoise_net')
trans_fun_B_with_2E = simple_unet(denoised_in, name='fun_est_B_with_2E')
enhanced_2E = out_acti(denoised_in * trans_fun_B_with_2E)
atten_map = atten(sample, name='atten')
fused = atten_map*enhanced_1E + (1-atten_map)*enhanced_2E
enhanced = fusion(fused, sample, name='fusion')
return enhanced
# -----------------------------------------#settings and preparations----------
sess = tf.Session()
in_image = tf.placeholder(tf.float32, [None, None, None, 3])
gt_image = tf.placeholder(tf.float32, [None, None, None, 3]) # The channel dimension of the inputs should be defined. Found `None`
uf_out = buildmodel(in_image)
# =------------------------------updates--------------------------------
time_elapsed = 0
with tf.Session() as sess:
saver = tf.compat.v1.train.Saver()
ckpt = tf.train.get_checkpoint_state(checkpoint_dir)
if ckpt:
print('loaded ' + ckpt.model_checkpoint_path)
saver.restore(sess, ckpt.model_checkpoint_path) # 恢复ckpt用于训练
# stats_graph(sess.graph)
# --------------------------------------------------------------------#
eval_fns = glob.glob(input_dir + '*.*')
for N in range(len(eval_fns)):
temp_train = np.array(cv2.imread(eval_fns[N])) # different!!!
temp_train = temp_train/255.0
# ---------------------------------------------------------------------#
train_data = temp_train.reshape(1, temp_train.shape[0], temp_train.shape[1], temp_train.shape[2])
st = time.time()
[out] = sess.run([uf_out], feed_dict={in_image: train_data})
time_elapsed += time.time() - st
print('%s' % eval_fns[N])
[_, name] = os.path.split(eval_fns[N])
suffix = name[name.find('.') + 1:]
name = name[:name.find('.')]
output = np.array(out[0])
output = output.reshape(output.shape[0], output.shape[1], output.shape[2])
output = output*255.0
output_rueslt = np.array(output)
if not os.path.isdir(result_dir):
os.makedirs(result_dir)
cv2.imwrite(result_dir + name + '_TBEFN.png', output_rueslt)
print('total processing time: ', time_elapsed)