-
Notifications
You must be signed in to change notification settings - Fork 0
/
picsnet.py
184 lines (96 loc) · 4.59 KB
/
picsnet.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
import tensorflow as tf
import numpy as np
from process_image import ProcessImage
class NeuralTransfer(object):
def __init__(self, style_image,
content_image,
height,
width,
style_weight,
content_weight,
tv_weight,
learning_rate,
iterations = 100,
blank_image=False):
self.style_image = style_image
self.content_image = content_image
self.height = height
self.width = width
self.style_weight = style_weight
self.content_weight = content_weight
self.tv_weight = tv_weight
self.learning_rate = learning_rate
self.iterations = iterations
if blank_image:
self.tinker_image = np.random.uniform(0, 255, (1, self.height, self.width, 3))
else:
self.tinker_image = content_image
self.layers_for_opt = ["block1_conv1", "block2_conv1", "block2_conv2", "block3_conv1",
"block3_conv2", "block4_conv1", "block4_conv3", "block5_conv1", "block5_conv3"]
@staticmethod
def gram_matrix(model, layer):
dimension = model.get_layer(layer).output.shape[3].value
permute = tf.transpose(model.get_layer(layer).output[0], perm=[2, 0, 1])
reshaped = tf.reshape(permute, [dimension, -1])
final = tf.matmul(reshaped, reshaped, transpose_b=True)
return final
def content_loss(self, input_image, content_image):
increment_loss = tf.reduce_sum(tf.square(input_image - content_image))/(self.height*self.width)
return increment_loss
def style_loss(self, input_gramm, style_gramm, weight):
increment_loss = tf.reduce_sum((tf.square(input_gramm - style_gramm)))
increment_loss = weight * increment_loss
return increment_loss
def tv_loss(self, image):
increment_loss = (tf.reduce_sum(tf.abs(image[:, 1:, :, :] - image[:, :-1, :, :])) +
tf.reduce_sum(tf.abs(image[:, :, 1:, :] - image[:, :, :-1, :])))
return increment_loss
def __make_style_grams__(self):
sess1 = tf.InteractiveSession()
style_tensor = tf.constant(self.style_image)
model = tf.keras.applications.VGG16(include_top=False, weights="imagenet", input_tensor=style_tensor)
model.trainable = False
self.grams = {}
for layer in self.layers_for_opt:
self.grams[layer] = sess1.run(self.gram_matrix(model, layer))
sess1.close()
tf.reset_default_graph()
def __stylize__(self):
sess2 = tf.InteractiveSession()
tinkered_tensor = tf.Variable(self.content_image, trainable=True)
content_tensor = tf.constant(self.content_image)
model2 = tf.keras.applications.VGG16(include_top=False, weights="imagenet", input_tensor=tinkered_tensor)
model2.trainable = False
model3 = tf.keras.applications.VGG16(include_top=False, weights="imagenet", input_tensor=content_tensor)
model3.trainable = False
loss = tf.Variable(initial_value=tf.zeros(1))
loss.initializer.run()
c_loss = self.content_loss(model3.get_layer("block1_conv2").output, model2.get_layer("block1_conv2").output)
t_loss = self.tv_loss(tinkered_tensor)
s_loss = 0
for layer in self.layers_for_opt:
current_gram = self.gram_matrix(model2, layer)
s_loss += self.style_loss(self.grams[layer], current_gram, 1)
loss = self.content_weight * c_loss + self.style_weight * s_loss + self.tv_weight * t_loss
opt = tf.train.AdamOptimizer(learning_rate=self.learning_rate)
run_it = opt.minimize(loss, var_list=tinkered_tensor)
tinkered_tensor.initializer.run()
sess2.run(tf.variables_initializer(opt.variables()))
for i in range(self.iterations):
_, current_loss = sess2.run([run_it, loss])
print(i, current_loss)
if i != 0 and i % 10 == 0:
final_image = sess2.run(tinkered_tensor)[0]
img = ProcessImage.unprocess_image(final_image)
img.save("temp_output.jpg")
img.show()
continue_optimizing = input("Continue optimizing (yes / no)? ")
if continue_optimizing == "no":
break
else:
continue
self.final_image = sess2.run(tinkered_tensor)[0]
sess2.close()
def run_stylizing(self):
self.__make_style_grams__()
self.__stylize__()