forked from kweisamx/TensorFlow-VDSR
-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
149 lines (122 loc) · 6.19 KB
/
model.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
import tensorflow as tf
import numpy as np
import time
import os
from utils import (
input_setup,
checkpoint_dir,
read_data,
merge,
checkimage,
imsave
)
class VDSR(object):
def __init__(self,
sess,
image_size,
label_size,
layer,
c_dim):
self.sess = sess
self.image_size = image_size
self.label_size = label_size
self.layer = layer
self.c_dim = c_dim
self.build_model()
def build_model(self):
self.images = tf.placeholder(tf.float32, [None, self.image_size, self.image_size, self.c_dim], name='images')
self.labels = tf.placeholder(tf.float32, [None, self.label_size, self.label_size, self.c_dim], name='labels')
self.weights = {
'w_start': tf.Variable(tf.random_normal([3, 3, self.c_dim, 64], stddev =np.sqrt(2.0/9)), name='w_start'),
'w_end': tf.Variable(tf.random_normal([3, 3, 64, self.c_dim], stddev=np.sqrt(2.0/9/64)), name='w_end')
}
self.biases = {
'b_start': tf.Variable(tf.zeros([64], name='b_start')),
'b_end': tf.Variable(tf.zeros([self.c_dim], name='b_end'))
}
# Create very deep layer weight and bias
for i in range(2, self.layer): #except start and end
self.weights.update({'w_%d' % i: tf.Variable(tf.random_normal([3, 3, 64, 64], stddev= np.sqrt(2.0/9/64)), name='w_%d' % i) })
self.biases.update({'b_%d' % i: tf.Variable(tf.zeros([64], name='b_%d' % i)) })
self.pred = self.model()
# residul = labels - images
self.loss = tf.reduce_mean(tf.square(self.labels - self.images - self.pred))
self.saver = tf.train.Saver() # To save checkpoint
def model(self):
conv = []
conv.append(tf.nn.relu(tf.nn.conv2d(self.images, self.weights['w_start'], strides=[1,1,1,1], padding='SAME') + self.biases['b_start']))
for i in range(2, self.layer):
conv.append(tf.nn.relu(tf.nn.conv2d(conv[i-2], self.weights['w_%d' % i], strides=[1,1,1,1], padding='SAME') + self.biases['b_%d' % i]))
#conv2 = tf.nn.relu(tf.nn.conv2d(conv[0], self.weights['w_2'], strides=[1,1,1,1], padding='SAME') + self.biases['b_2'])
conv_end = tf.nn.conv2d(conv[i-1], self.weights['w_end'], strides=[1,1,1,1], padding='SAME') + self.biases['b_end'] # This layer don't need ReLU
return conv_end
def train(self, config):
# NOTE : if train, the nx, ny are ingnored
nx, ny = input_setup(config)
data_dir = checkpoint_dir(config)
input_, label_ = read_data(data_dir)
# Stochastic gradient descent with the standard backpropagation
# NOTE: learning rate decay
global_step = tf.Variable(0, trainable=False)
#learning_rate = tf.train.exponential_decay(config.learning_rate, global_step * config.batch_size, len(input_)*100, 0.1, staircase=True)
# NOTE: Clip gradient
opt = tf.train.AdamOptimizer(learning_rate=config.learning_rate)
grad_and_value = opt.compute_gradients(self.loss)
clip = tf.Variable(config.clip_grad, name='clip')
capped_gvs = [(tf.clip_by_value(grad, -(clip), clip), var) for grad, var in grad_and_value]
self.train_op = opt.apply_gradients(capped_gvs, global_step=global_step)
#self.train_op = tf.train.AdamOptimizer(learning_rate=config.learning_rate).minimize(self.loss)
tf.initialize_all_variables().run()
counter = 0
time_ = time.time()
self.load(config.checkpoint_dir)
# Train
if config.is_train:
print("Now Start Training...")
for ep in range(config.epoch):
# Run by batch images
batch_idxs = len(input_) // config.batch_size
for idx in range(0, batch_idxs):
batch_images = input_[idx * config.batch_size : (idx + 1) * config.batch_size]
batch_labels = label_[idx * config.batch_size : (idx + 1) * config.batch_size]
counter += 1
_, err = self.sess.run([self.train_op, self.loss], feed_dict={self.images: batch_images, self.labels: batch_labels})
if counter % 10 == 0:
print("Epoch: [%2d], step: [%2d], time: [%4.4f], loss: [%.8f]" % ((ep+1), counter, time.time()-time_, err ))
if counter % 500 == 0:
self.save(config.checkpoint_dir, counter)
# Test
else:
print("Now Start Testing...")
result = self.pred.eval({self.images: input_}) + input_
image = merge(result, [nx, ny], self.c_dim)
checkimage(merge(result, [nx, ny], self.c_dim))
#checkimage(image_LR)
imsave(image, config.result_dir+'/result.png', config)
def load(self, checkpoint_dir):
"""
To load the checkpoint use to test or pretrain
"""
print("\nReading Checkpoints.....\n\n")
model_dir = "%s_%s_%slayer" % ("vdsr", self.label_size,self.layer)# give the model name by label_size
checkpoint_dir = os.path.join(checkpoint_dir, model_dir)
ckpt = tf.train.get_checkpoint_state(checkpoint_dir)
# Check the checkpoint is exist
if ckpt and ckpt.model_checkpoint_path:
ckpt_path = str(ckpt.model_checkpoint_path) # convert the unicode to string
self.saver.restore(self.sess, os.path.join(os.getcwd(), ckpt_path))
print("\n Checkpoint Loading Success! %s\n\n"% ckpt_path)
else:
print("\n! Checkpoint Loading Failed \n\n")
def save(self, checkpoint_dir, step):
"""
To save the checkpoint use to test or pretrain
"""
model_name = "VDSR.model"
model_dir = "%s_%s_%slayer" % ("vdsr", self.label_size,self.layer)
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)