tf-gen/superres.py at master · Hello1024/tf-gen · GitHub

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
import tensorflow as tf
import utils
import time
import sys
from datetime import datetime
import matplotlib.pyplot as plt
import random

import superres_model

BATCH_SIZE = 1

with tf.name_scope("input"):
  jpg = tf.image.decode_jpeg(open('city.jpg', 'r').read())
  jpg.set_shape([100,100,3])  # Bodge to bypass checks later...
  jpg = tf.image.per_image_whitening(jpg)
  crops = []
  for i in range(BATCH_SIZE):
    crops.append(tf.image.random_crop(jpg, [512, 512]))
  jpg = tf.pack(crops)
  #jpg = tf.image.resize_bilinear(jpg, [128, 128])

mix = tf.random_uniform([1]);
net_in = mix*jpg + (1.0-mix)*tf.truncated_normal(jpg.get_shape(), dtype=tf.float32, stddev=1)
l10 = net_in


all_images = []

with tf.variable_scope("gen", reuse=None) as scope:
  with tf.name_scope("1"):
    l10 = superres_model.superres_model(l10)
  all_images.append(l10)
with tf.variable_scope("gen", reuse=True) as scope:
  for i in range(2,8):
    with tf.name_scope(str(i)):
      l10 = superres_model.superres_model(l10)
    all_images.append(l10)

measure_images = tf.concat(0, [image[:,:,:,0:3] for image in [all_images[i] for i in [0, 2, 4, 6]]])
feedback_images = tf.concat(0, [image[:,:,:,0:3] for image in [all_images[i] for i in [6]]])
show_images = tf.concat(0, [image[0:1,:,:,0:3] for image in all_images])


tf.image_summary("gen", measure_images, max_images=1)
tf.image_summary("real", jpg, max_images=1)

adv_inp = tf.concat(0, [measure_images, jpg])
answers = tf.concat(0, [
  tf.tile(tf.constant([[0.0,1.0]]), [BATCH_SIZE*4,1]),
  tf.tile(tf.constant([[1.0,0.0]]), [BATCH_SIZE,1])
])

with tf.name_scope("adv") as scope:
  al = utils._conv(adv_inp, 5, 5, 10)
  al = utils._half(al)
  al = utils._conv(al, 5, 5, 15)
  al = utils._half(al)
  al = utils._conv(al, 5, 5, 25)
  al = utils._half(al)
  al = utils._conv(al, 5, 5, 25)
  al = utils._half(al)
  al = utils._conv(al, 5, 5, 35)
  al = utils._half(al)
  al = utils._conv(al, 5, 5, 35)
  al = utils._half(al)
  al = utils._conv(al, 5, 5, 35)
  al = utils._half(al)
  al = utils._conv(al, 3, 3, 35)
  al = utils._fc(al, 2)
  al = tf.nn.softmax(al)


def normclip(grads_and_vars):
  #for g,v in grads_and_vars:
  #  print "G: " + str(g)
  #  print "V: " + str(v.name)
  #return grads_and_vars
  return [ (tf.clip_by_norm(g, 1.0),v) for g, v in grads_and_vars]


adv_entropy = -tf.reduce_sum(answers*tf.log(tf.clip_by_value(al, 1e-10, 1.0))) / BATCH_SIZE
tf.scalar_summary("adv_entropy", adv_entropy)
adv_opt = tf.train.AdagradOptimizer(2e-2)
adv_train_step = adv_opt.apply_gradients(normclip(adv_opt.compute_gradients(adv_entropy, var_list=[x for x in tf.trainable_variables() if "adv" in x.name])))


gen_entropy = -tf.reduce_sum((1.0-answers)*tf.log(tf.clip_by_value(al, 1e-10, 1.0))) / BATCH_SIZE
tf.scalar_summary("gen_entropy", gen_entropy)
gen_opt = tf.train.AdagradOptimizer(2e-2)
gen_train_step = gen_opt.apply_gradients(normclip(gen_opt.compute_gradients(gen_entropy, var_list=[x for x in tf.trainable_variables() if "gen" in x.name])))


saver = tf.train.Saver()

plt.ion()

grad_op = tf.gradients(gen_entropy, adv_inp)[0]*300

config = tf.ConfigProto()
config.gpu_options.allocator_type = 'BFC'

with tf.Session(config=config) as sess:
  log_path = '/tmp/superres/'+datetime.now().strftime("%Y/%m/%d/%H/%M/%S")
  merged_summary_op = tf.merge_all_summaries()
  summary_writer = tf.train.SummaryWriter(log_path, sess.graph_def)

  init = tf.initialize_all_variables()
  sess.run(init)
  print "variables initialized"

  if len(sys.argv)>1:
    saver.restore(sess, sys.argv[1])


  ae=1.0

  feedback_images_res = None

  for i in range(1000000):

     if feedback_images_res is not None and random.random()<0.90:
       feed_dict={net_in: feedback_images_res}
     else:
       print "new"
       feed_dict={}

     if not (i%10):
       summary_str, out, ae, ge, gg, dif, show_images_res = sess.run([merged_summary_op, adv_inp, adv_entropy, gen_entropy, al, grad_op, show_images], feed_dict=feed_dict)
       summary_writer.add_summary(summary_str, i)

       save_path = saver.save(sess, log_path+ "/model.ckpt")

       print gg

       #print "adv entropy: " + str(ae)
       #print "gen entropy: " + str(ge)

       #for b in range(4):
       #  plt.subplot(4, 2, b*2+1)
       #  plt.imshow((out[b*BATCH_SIZE,:,:,0:3]/3+0.5).clip(min=0, max=1))
       #  plt.subplot(4, 2, b*2+2)
       #  plt.imshow((dif[b*BATCH_SIZE,:,:,0:3]+0.5).clip(min=0, max=1))
       for b in range(show_images_res.shape[0]):
         plt.subplot(2, 5, b+1)
         plt.imshow((show_images_res[b,:,:,:]/3+0.5).clip(min=0, max=1))

       plt.pause(0.1)


     if ae*5 < ge:
       chosen_step = gen_train_step
     else:
       chosen_step = adv_train_step

     ae, ge, feedback_images_res, _ = sess.run([adv_entropy, gen_entropy, feedback_images, chosen_step], feed_dict=feed_dict)
     print i, ae, ge


     #for j in range(len(gv)):
     #  print "-----------------"
     #  print ddd[j].name
     #  print gv[j]

     # summary_str = sess.run(merged_summary_op, feed_dict={inp: testin.eval()})
     # summary_writer.add_summary(summary_str, i)

#print prob