-
Notifications
You must be signed in to change notification settings - Fork 5
/
model_T.py
266 lines (230 loc) · 14.1 KB
/
model_T.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
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
# -*- coding: utf-8 -*-
from utils import (
read_data,
input_setup_MS,
input_setup_PAN,
imsave,
merge,
sobel_gradient,
lrelu,
l2_norm,
linear_map,
lpls_gradient,
lpls_gradient_4,
sobel_gradient_4
)
import time
import os
import matplotlib.pyplot as plt
import numpy as np
import tensorflow as tf
class T_model(object):
def __init__(self,
sess,
image_size_MS=20,
image_size_PAN=80,
batch_size=48,
c_dim=1,
checkpoint_dir=None,
sample_dir=None):
self.sess = sess
self.is_grayscale = (c_dim == 1)
self.image_size_MS = image_size_MS
self.image_size_PAN = image_size_PAN
self.image_size_Label = image_size_PAN
self.batch_size = batch_size
self.c_dim = c_dim
self.checkpoint_dir = checkpoint_dir
self.sample_dir = sample_dir
self.build_model()
def build_model(self):
########## MS Input ######################
with tf.name_scope('MS1_input'):
self.images_MS1 = tf.placeholder(tf.float32, [None, self.image_size_MS, self.image_size_MS, self.c_dim], name='images_MS1')
with tf.name_scope('MS2_input'):
self.images_MS2 = tf.placeholder(tf.float32, [None, self.image_size_MS, self.image_size_MS, self.c_dim], name='images_MS2')
with tf.name_scope('MS3_input'):
self.images_MS3 = tf.placeholder(tf.float32, [None, self.image_size_MS, self.image_size_MS, self.c_dim], name='images_MS3')
with tf.name_scope('MS4_input'):
self.images_MS4 = tf.placeholder(tf.float32, [None, self.image_size_MS, self.image_size_MS, self.c_dim], name='images_MS4')
########## MS Label Input ######################
with tf.name_scope('MS1_Label'):
self.Label_MS1 = tf.placeholder(tf.float32, [None, self.image_size_Label, self.image_size_Label, self.c_dim], name='Label_MS1')
with tf.name_scope('MS2_Label'):
self.Label_MS2 = tf.placeholder(tf.float32, [None, self.image_size_Label, self.image_size_Label, self.c_dim], name='Label_MS2')
with tf.name_scope('MS3_Label'):
self.Label_MS3 = tf.placeholder(tf.float32, [None, self.image_size_Label, self.image_size_Label, self.c_dim], name='Label_MS3')
with tf.name_scope('MS4_Label'):
self.Label_MS4 = tf.placeholder(tf.float32, [None, self.image_size_Label, self.image_size_Label, self.c_dim], name='Label_MS4')
########## PAN Input ######################
with tf.name_scope('PAN_input'):
self.images_PAN = tf.placeholder(tf.float32, [None, self.image_size_PAN, self.image_size_PAN, self.c_dim], name='images_PAN')
with tf.name_scope('input'):
self.input_image_MS1 = self.images_MS1
self.input_image_MS2 = self.images_MS2
self.input_image_MS3 = self.images_MS3
self.input_image_MS4 = self.images_MS4
self.input_image_MS =tf.concat([self.images_MS1,self.images_MS2,self.images_MS3,self.images_MS4],axis=-1)
self.input_image_PAN = self.images_PAN
self.input_Label_MS1 = self.Label_MS1
self.input_Label_MS2 = self.Label_MS2
self.input_Label_MS3 = self.Label_MS3
self.input_Label_MS4 = self.Label_MS4
self.input_Label_MS =tf.concat([self.input_Label_MS1,self.input_Label_MS2,self.input_Label_MS3,self.input_Label_MS4],axis=-1)
with tf.name_scope('Gradient'):
self.Label_MS_gradient_x,self.Label_MS_gradient_y=sobel_gradient_4(self.input_Label_MS)
self.HRPAN_gradient_x,self.HRPAN_gradient_y=sobel_gradient(self.images_PAN)
with tf.name_scope('fusion'):
self.MS2PAN_gradient_x=self.transfer_model(self.Label_MS_gradient_x,reuse=False)
self.MS2PAN_gradient_y=self.transfer_model(self.Label_MS_gradient_y,reuse=True,update_collection='NO_OPS')
with tf.name_scope('t_loss'):
self.t_loss_MS2PAN_grad_x = tf.reduce_mean(tf.square(self.MS2PAN_gradient_x - self.HRPAN_gradient_x))
self.t_loss_MS2PAN_grad_y = tf.reduce_mean(tf.square(self.MS2PAN_gradient_y - self.HRPAN_gradient_y))
self.t_loss_total=100*(self.t_loss_MS2PAN_grad_x+self.t_loss_MS2PAN_grad_y)
tf.summary.scalar('t_loss_MS2PAN_grad_x',self.t_loss_MS2PAN_grad_x)
tf.summary.scalar('t_loss_MS2PAN_grad_y',self.t_loss_MS2PAN_grad_y)
tf.summary.scalar('t_loss_total',self.t_loss_total)
self.saver = tf.train.Saver(max_to_keep=100)
with tf.name_scope('image'):
tf.summary.image('input_Label_MS',tf.expand_dims(self.input_Label_MS[1,:,:,:],0))
tf.summary.image('input_image_PAN',tf.expand_dims(self.input_image_PAN[1,:,:,:],0))
tf.summary.image('MS2PAN_gradient_x',tf.expand_dims(self.MS2PAN_gradient_x[1,:,:,:],0))
tf.summary.image('MS2PAN_gradient_y',tf.expand_dims(self.MS2PAN_gradient_y[1,:,:,:],0))
tf.summary.image('HRPAN_gradient_x',tf.expand_dims(self.HRPAN_gradient_x[1,:,:,:],0))
tf.summary.image('HRPAN_gradient_y',tf.expand_dims(self.HRPAN_gradient_y[1,:,:,:],0))
def train(self, config):
if config.is_train:
input_setup_MS(self.sess, config,"data/Train_data/Train_MS1")
input_setup_MS(self.sess, config,"data/Train_data/Train_MS2")
input_setup_MS(self.sess, config,"data/Train_data/Train_MS3")
input_setup_MS(self.sess, config,"data/Train_data/Train_MS4")
input_setup_PAN(self.sess,config,"data/Train_data/Train_PAN")
input_setup_PAN(self.sess, config,"data/Train_data/Label_MS1")
input_setup_PAN(self.sess, config,"data/Train_data/Label_MS2")
input_setup_PAN(self.sess, config,"data/Train_data/Label_MS3")
input_setup_PAN(self.sess, config,"data/Train_data/Label_MS4")
if config.is_train:
data_dir_MS1 = os.path.join('./{}'.format(config.checkpoint_dir), "data/Train_data/Train_MS1","train.h5")
data_dir_MS2 = os.path.join('./{}'.format(config.checkpoint_dir), "data/Train_data/Train_MS2","train.h5")
data_dir_MS3 = os.path.join('./{}'.format(config.checkpoint_dir), "data/Train_data/Train_MS3","train.h5")
data_dir_MS4 = os.path.join('./{}'.format(config.checkpoint_dir), "data/Train_data/Train_MS4","train.h5")
data_dir_PAN = os.path.join('./{}'.format(config.checkpoint_dir), "data/Train_data/Train_PAN","train.h5")
data_dir_Label_MS1 = os.path.join('./{}'.format(config.checkpoint_dir), "data/Train_data/Label_MS1","train.h5")
data_dir_Label_MS2 = os.path.join('./{}'.format(config.checkpoint_dir), "data/Train_data/Label_MS2","train.h5")
data_dir_Label_MS3 = os.path.join('./{}'.format(config.checkpoint_dir), "data/Train_data/Label_MS3","train.h5")
data_dir_Label_MS4 = os.path.join('./{}'.format(config.checkpoint_dir), "data/Train_data/Label_MS4","train.h5")
train_data_MS1= read_data(data_dir_MS1)
train_data_MS2= read_data(data_dir_MS2)
train_data_MS3= read_data(data_dir_MS3)
train_data_MS4= read_data(data_dir_MS4)
train_data_PAN= read_data(data_dir_PAN)
train_data_Label_MS1= read_data(data_dir_Label_MS1)
train_data_Label_MS2= read_data(data_dir_Label_MS2)
train_data_Label_MS3= read_data(data_dir_Label_MS3)
train_data_Label_MS4= read_data(data_dir_Label_MS4)
t_vars = tf.trainable_variables()
self.trans_vars = [var for var in t_vars if 'transfer_model' in var.name]
print(self.trans_vars)
with tf.name_scope('train_step'):
self.train_trans_op = tf.train.AdamOptimizer(config.learning_rate).minimize(self.t_loss_total,var_list=self.trans_vars)
self.summary_op = tf.summary.merge_all()
self.train_writer = tf.summary.FileWriter(config.summary_dir + '/train',self.sess.graph,flush_secs=60)
tf.initialize_all_variables().run()
counter = 0
start_time = time.time()
if config.is_train:
print("Training...")
for ep in xrange(config.epoch):
# Run by batch images
batch_idxs = len(train_data_PAN) // config.batch_size
for idx in xrange(0, batch_idxs):
batch_images_MS1 = train_data_MS1[idx*config.batch_size : (idx+1)*config.batch_size]
batch_images_MS2 = train_data_MS2[idx*config.batch_size : (idx+1)*config.batch_size]
batch_images_MS3 = train_data_MS3[idx*config.batch_size : (idx+1)*config.batch_size]
batch_images_MS4 = train_data_MS4[idx*config.batch_size : (idx+1)*config.batch_size]
batch_images_PAN = train_data_PAN[idx*config.batch_size : (idx+1)*config.batch_size]
batch_Label_MS1 = train_data_Label_MS1[idx*config.batch_size : (idx+1)*config.batch_size]
batch_Label_MS2 = train_data_Label_MS2[idx*config.batch_size : (idx+1)*config.batch_size]
batch_Label_MS3 = train_data_Label_MS3[idx*config.batch_size : (idx+1)*config.batch_size]
batch_Label_MS4 = train_data_Label_MS4[idx*config.batch_size : (idx+1)*config.batch_size]
counter += 1
_, err_trans,summary_str= self.sess.run([self.train_trans_op, self.t_loss_total,self.summary_op], feed_dict={self.images_MS1: batch_images_MS1,self.images_MS2: batch_images_MS2,self.images_MS3: batch_images_MS3,self.images_MS4: batch_images_MS4,self.images_PAN: batch_images_PAN,self.Label_MS1: batch_Label_MS1,self.Label_MS2: batch_Label_MS2,self.Label_MS3: batch_Label_MS3,self.Label_MS4: batch_Label_MS4})
self.train_writer.add_summary(summary_str,counter)
if counter % 10 == 0:
print("Epoch: [%2d], step: [%2d], time: [%4.4f],loss_trans:[%.8f]" \
% ((ep+1), counter, time.time()-start_time, err_trans))
self.save(config.checkpoint_dir, ep)
def transfer_model(self,img_MS_grad,reuse,update_collection=None):
with tf.variable_scope('transfer_model',reuse=reuse):
#########################################################
#################### grad Layer 1 #######################
#########################################################
with tf.variable_scope('layer1_grad'):
weights=tf.get_variable("w1_grad",[3,3,4,16],initializer=tf.truncated_normal_initializer(stddev=1e-3))
bias=tf.get_variable("b1_grad",[16],initializer=tf.constant_initializer(0.0))
conv1_grad = tf.nn.conv2d(img_MS_grad, weights, strides=[1,1,1,1], padding='SAME') + bias
conv1_grad = lrelu(conv1_grad)
#########################################################
#################### grad Layer 2 ###########################
#########################################################
with tf.variable_scope('layer2_grad'):
weights=tf.get_variable("w2_grad",[3,3,16,16],initializer=tf.truncated_normal_initializer(stddev=1e-3))
bias=tf.get_variable("b2_grad",[16],initializer=tf.constant_initializer(0.0))
conv2_grad = tf.nn.conv2d(conv1_grad, weights, strides=[1,1,1,1], padding='SAME') + bias
conv2_grad = lrelu(conv2_grad)
#########################################################
#################### grad Layer 3 ###########################
#########################################################
with tf.variable_scope('layer3_grad'):
weights=tf.get_variable("w3_grad",[3,3,16,16],initializer=tf.truncated_normal_initializer(stddev=1e-3))
bias=tf.get_variable("b3_grad",[16],initializer=tf.constant_initializer(0.0))
conv3_grad = tf.nn.conv2d(conv2_grad, weights, strides=[1,1,1,1], padding='SAME') + bias
conv3_grad = lrelu(conv3_grad)
#########################################################
#################### grad Layer 4 ###########################
#########################################################
grad_cat_4=tf.concat([conv1_grad,conv3_grad],axis=-1)
with tf.variable_scope('layer4_grad'):
weights=tf.get_variable("w4_grad",[3,3,32,16],initializer=tf.truncated_normal_initializer(stddev=1e-3))
bias=tf.get_variable("b4_grad",[16],initializer=tf.constant_initializer(0.0))
conv4_grad = tf.nn.conv2d(grad_cat_4, weights, strides=[1,1,1,1], padding='SAME') + bias
conv4_grad = lrelu(conv4_grad)
#########################################################
#################### grad Layer 5 #######################
#########################################################
grad_cat_5=tf.concat([img_MS_grad,conv4_grad],axis=-1)
with tf.variable_scope('layer5_grad'):
weights=tf.get_variable("w5_grad",[3,3,20,8],initializer=tf.truncated_normal_initializer(stddev=1e-3))
bias=tf.get_variable("b5_grad",[8],initializer=tf.constant_initializer(0.0))
conv5_grad = tf.nn.conv2d(grad_cat_5, weights, strides=[1,1,1,1], padding='SAME') + bias
conv5_grad = lrelu(conv5_grad)
#########################################################
#################### grad Layer 6 #######################
#########################################################
with tf.variable_scope('layer6_grad'):
weights=tf.get_variable("w6_grad",[3,3,8,1],initializer=tf.truncated_normal_initializer(stddev=1e-3))
bias=tf.get_variable("b6_grad",[1],initializer=tf.constant_initializer(0.0))
conv6_grad = tf.nn.conv2d(conv5_grad, weights, strides=[1,1,1,1], padding='SAME') + bias
conv6_grad = tf.nn.tanh(conv6_grad)*2
return conv6_grad
def save(self, checkpoint_dir, step):
model_name = "T_model.model"
model_dir = "%s" % ("T_model")
checkpoint_dir = os.path.join(checkpoint_dir, model_dir)
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
self.saver.save(self.sess,
os.path.join(checkpoint_dir, model_name),
global_step=step)
def load(self, checkpoint_dir):
print(" [*] Reading checkpoints...")
model_dir = "%s" % ("T_model")
checkpoint_dir = os.path.join(checkpoint_dir, model_dir)
ckpt = tf.train.get_checkpoint_state(checkpoint_dir)
if ckpt and ckpt.model_checkpoint_path:
ckpt_name = os.path.basename(ckpt.model_checkpoint_path)
print(ckpt_name)
self.saver.restore(self.sess, os.path.join(checkpoint_dir,ckpt_name))
return True
else:
return False