前一帖生成音乐，本帖生成图片。本文使用TensorFlow实现论文《Conditional Image Generation with PixelCNN Decoders》，它是基于PixelCNN架构的模型，最早出现在《Pixel Recurrent Neural Networks》一文。
使用的图片数据
我本想使用ImageNet做为图片来源，就像论文中使用的。ImageNet图像有现成的分类，抓取也容易，但是由于很多源都被防火墙屏蔽，下载速度堪忧。《OpenCV之使用Haar Cascade进行对象识别》
我看到网上有很多爬妹纸图的Python脚本，额，我爬了几天几夜的妹纸图（特别暴露那种），额，我就想看看PixelCNN最后能生成什么鬼。
如果你懒的爬图片，可以使用我抓取的图片（分成两部分）：
- https://pan.baidu.com/s/1kVSA8z9 (密码: atqm)
- https://pan.baidu.com/s/1ctbd9O (密码: kubu)
数据预处理
下载的图片分布在多个目录，把图片汇总到一个新目录：
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
|
import
os
old_dir
=
'images'
new_dir
=
'girls'
if
not
os.path
.
exists
(
new_dir
)
:
os
.
makedirs
(
new_dir
)
count
=
0
for
(
dirpath
,
dirnames
,
filenames
)
in
os
.
walk
(
old_dir
)
:
for
filename
in
filenames
:
if
filename
.
endswith
(
'.jpg'
)
:
new_filename
=
str
(
count
)
+
'.jpg'
os
.
rename
(
os
.
sep
.
join
(
[
dirpath
,
filename
]
)
,
os
.
sep
.
join
(
[
new_dir
,
new_filename
]
)
)
print
(
os
.
sep
.
join
(
[
dirpath
,
filename
]
)
)
count
+=
1
print
(
"Total Picture: "
,
count
)
|
使用《open_nsfw: 基于Caffe的成人图片识别模型》剔除掉和妹子图不相关的图片，给open_nsfw输入要检测的图片，它会返回图片评级（0-1），等级越高，图片越黄越暴力。使用OpenCV应该也不难。
为了减小计算量，我把图像缩放为64×64像素：
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
|
import
os
import
cv2
import
numpy
as
np
image_dir
=
'girls'
new_girl_dir
=
'little_girls'
if
not
os.path
.
exists
(
new_girl_dir
)
:
os
.
makedirs
(
new_girl_dir
)
for
img_file
in
os
.
listdir
(
image_dir
)
:
img_file_path
=
os.path
.
join
(
image_dir
,
img_file
)
img
=
cv2
.
imread
(
img_file_path
)
if
img
is
None
:
print
(
"image read fail"
)
continue
height
,
weight
,
channel
=
img
.
shape
if
height
<
200
or
weight
<
200
or
channel
!=
3
:
continue
# 你也可以转为灰度图片(channel=1)，加快训练速度
# 把图片缩放为64x64
img
=
cv2
.
resize
(
img
,
(
64
,
64
)
)
new_file
=
os.path
.
join
(
new_girl_dir
,
img_file
)
cv2
.
imwrite
(
new_file
,
img
)
print
(
new_file
)
|
去除重复图片：
|
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
|
import
os
import
cv2
import
numpy
as
np
# 判断两张图片是否完全一样（使用哈希应该要快很多）
def
is_same_image
(
img_file1
,
img_file2
)
:
img1
=
cv2
.
imread
(
img_file1
)
img2
=
cv2
.
imread
(
img_file2
)
if
img1
is
None
or
img2
is
None
:
return
False
if
img1
.
shape
==
img2
.
shape
and
not
(
np
.
bitwise_xor
(
img1
,
img2
)
.
any
(
)
)
:
return
True
else
:
return
False
# 去除重复图片
file_list
=
os
.
listdir
(
'little_girls'
)
try
:
for
img1
in
file_list
:
print
(
len
(
file_list
)
)
for
img2
in
file_list
:
if
img1
!=
img2
:
if
is_same_image
(
'little_girls/'
+
img1
,
'little_girls/'
+
img2
)
is
True
:
print
(
img1
,
img2
)
os
.
remove
(
'little_girls/'
+
img1
)
file_list
.
remove
(
img1
)
except
Exception
as
e
:
print
(
e
)
|
PixelCNN生成妹纸图完整代码
下面代码只实现了unconditional模型（无条件），没有实现conditional和autoencoder模型。详细信息，请参看论文。
|
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
|
# -*- coding: utf-8 -*-
import
tensorflow
as
tf
import
numpy
as
np
import
os
import
cv2
# 如果使用mnist数据集，把MNIST设置为True
MNIST
=
False
if
MNIST
==
True
:
from
tensorflow
.
examples
.
tutorials
.
mnist
import
input_data
data
=
input_data
.
read_data_sets
(
'/tmp/'
)
image_height
=
28
image_width
=
28
image_channel
=
1
batch_size
=
128
n_batches
=
data
.
train
.
num_examples
/
/
batch_size
else
:
picture_dir
=
'little_girls'
picture_list
=
[
]
# 建议不要把图片一次加载到内存，为了节省内存，最好边加载边使用
for
(
dirpath
,
dirnames
,
filenames
)
in
os
.
walk
(
picture_dir
)
:
for
filename
in
filenames
:
if
filename
.
endswith
(
'.jpg'
)
:
picture_list
.
append
(
os
.
sep
.
join
(
[
dirpath
,
filename
]
)
)
print
(
"图像总数: "
,
len
(
picture_list
)
)
# 图像大小和Channel
image_height
=
64
image_width
=
64
image_channel
=
3
# 每次使用多少样本训练
batch_size
=
128
n_batches
=
len
(
picture_list
)
/
/
batch_size
#图片格式对应输入X
img_data
=
[
]
for
img_file
in
picture_list
:
img_data
.
append
(
cv2
.
imread
(
img_file
)
)
img_data
=
np
.
array
(
img_data
)
img_data
=
img_data
/
255.0
#print(img_data.shape) # (44112, 64, 64, 3)
X
=
tf
.
placeholder
(
tf
.
float32
,
shape
=
[
None
,
image_height
,
image_width
,
image_channel
]
)
def
gated_cnn
(
W_shape_
,
fan_in
,
gated
=
True
,
payload
=
None
,
mask
=
None
,
activation
=
True
)
:
W_shape
=
[
W_shape_
[
0
]
,
W_shape_
[
1
]
,
fan_in
.
get_shape
(
)
[
-
1
]
,
W_shape_
[
2
]
]
b_shape
=
W_shape_
[
2
]
def
get_weights
(
shape
,
name
,
mask
=
None
)
:
weights_initializer
=
tf
.
contrib
.
layers
.
xavier_initializer
(
)
W
=
tf
.
get_variable
(
name
,
shape
,
tf
.
float32
,
weights_initializer
)
if
mask
:
filter_mid_x
=
shape
[
0
]
/
/
2
filter_mid_y
=
shape
[
1
]
/
/
2
mask_filter
=
np
.
ones
(
shape
,
dtype
=
np
.
float32
)
mask_filter
[
filter_mid_x
,
filter_mid_y
+
1
:
,
:
,
:
]
=
0.
mask_filter
[
filter_mid_x
+
1
:
,
:
,
:
,
:
]
=
0.
if
mask
==
'a'
:
mask_filter
[
filter_mid_x
,
filter_mid_y
,
:
,
:
]
=
0.
W
*=
mask_filter
return
W
if
gated
:
W_f
=
get_weights
(
W_shape
,
"v_W"
,
mask
=
mask
)
W_g
=
get_weights
(
W_shape
,
"h_W"
,
mask
=
mask
)
b_f
=
tf
.
get_variable
(
"v_b"
,
b_shape
,
tf
.
float32
,
tf
.
zeros_initializer
)
b_g
=
tf
.
get_variable
(
"h_b"
,
b_shape
,
tf
.
float32
,
tf
.
zeros_initializer
)
conv_f
=
tf
.
nn
.
conv2d
(
fan_in
,
W_f
,
strides
=
[
1
,
1
,
1
,
1
]
,
padding
=
'SAME'
)
conv_g
=
tf
.
nn
.
conv2d
(
fan_in
,
W_g
,
strides
=
[
1
,
1
,
1
,
1
]
,
padding
=
'SAME'
)
if
payload
is
not
None
:
conv_f
+=
payload
conv_g
+=
payload
fan_out
=
tf
.
mul
(
tf
.
tanh
(
conv_f
+
b_f
)
,
tf
.
sigmoid
(
conv_g
+
b_g
)
)
else
:
W
=
get_weights
(
W_shape
,
"W"
,
mask
=
mask
)
b
=
tf
.
get_variable
(
"b"
,
b_shape
,
tf
.
float32
,
tf
.
zeros_initializer
)
conv
=
tf
.
nn
.
conv2d
(
fan_in
,
W
,
strides
=
[
1
,
1
,
1
,
1
]
,
padding
=
'SAME'
)
if
activation
:
fan_out
=
tf
.
nn
.
relu
(
tf
.
add
(
conv
,
b
)
)
else
:
fan_out
=
tf
.
add
(
conv
,
b
)
return
fan_out
def
pixel_cnn
(
layers
=
12
,
f_map
=
32
)
:
v_stack_in
,
h_stack_in
=
X
,
X
for
i
in
range
(
layers
)
:
filter_size
=
3
if
i
>
0
else
7
mask
=
'b'
if
i
>
0
else
'a'
residual
=
True
if
i
>
0
else
False
i
=
str
(
i
)
with
tf
.
variable_scope
(
"v_stack"
+
i
)
:
v_stack
=
gated_cnn
(
[
filter_size
,
filter_size
,
f_map
]
,
v_stack_in
,
mask
=
mask
)
v_stack_in
=
v_stack
with
tf
.
variable_scope
(
"v_stack_1"
+
i
)
:
v_stack_1
=
gated_cnn
(
[
1
,
1
,
f_map
]
,
v_stack_in
,
gated
=
False
,
mask
=
mask
)
with
tf
.
variable_scope
(
"h_stack"
+
i
)
:
h_stack
=
gated_cnn
(
[
1
,
filter_size
,
f_map
]
,
h_stack_in
,
payload
=
v_stack_1
,
mask
=
mask
)
with
tf
.
variable_scope
(
"h_stack_1"
+
i
)
:
h_stack_1
=
gated_cnn
(
[
1
,
1
,
f_map
]
,
h_stack
,
gated
=
False
,
mask
=
mask
)
if
residual
:
h_stack_1
+=
h_stack_in
h_stack_in
=
h_stack_1
with
tf
.
variable_scope
(
"fc_1"
)
:
fc1
=
gated_cnn
(
[
1
,
1
,
f_map
]
,
h_stack_in
,
gated
=
False
,
mask
=
'b'
)
color
=
256
with
tf
.
variable_scope
(
"fc_2"
)
:
fc2
=
gated_cnn
(
[
1
,
1
,
image_channel
*
color
]
,
fc1
,
gated
=
False
,
mask
=
'b'
,
activation
=
False
)
fc2
=
tf
.
reshape
(
fc2
,
(
-
1
,
color
)
)
return
fc2
def
train_pixel_cnn
(
)
:
output
=
pixel_cnn
(
)
loss
=
tf
.
reduce_mean
(
tf
.
nn
.
sparse_softmax_cross_entropy_with_logits
(
output
,
tf
.
cast
(
tf
.
reshape
(
X
,
[
-
1
]
)
,
dtype
=
tf
.
int32
)
)
)
trainer
=
tf
.
train
.
RMSPropOptimizer
(
1e
-
3
)
gradients
=
trainer
.
compute_gradients
(
loss
)
clipped_gradients
=
[
(
tf
.
clip_by_value
(
_
[
0
]
,
-
1
,
1
)
,
_
[
1
]
)
for
_
in
gradients
]
optimizer
=
trainer
.
apply_gradients
(
clipped_gradients
)
with
tf
.
Session
(
)
as
sess
:
sess
.
run
(
tf
.
initialize_all_variables
(
)
)
saver
=
tf
.
train
.
Saver
(
tf
.
trainable_variables
(
)
)
for
epoch
in
range
(
50
)
:
for
batch
in
range
(
n_batches
)
:
if
MNIST
==
True
:
batch_X
,
_
=
data
.
train
.
next_batch
(
batch_size
)
batch_X
=
batch_X
.
reshape
(
[
batch_size
,
image_height
,
image_width
,
image_channel
]
)
else
:
batch_X
=
img_data
[
batch_size
*
batch
:
batch_size
*
(
batch
+
1
)
]
_
,
cost
=
sess
.
run
(
[
optimizer
,
loss
]
,
feed_dict
=
{
X
:
batch_X
}
)
print
(
"epoch:"
,
epoch
,
' batch:'
,
batch
,
' cost:'
,
cost
)
if
epoch
%
7
==
0
:
saver
.
save
(
sess
,
"girl.ckpt"
,
global_step
=
epoch
)
# 训练
train_pixel_cnn
(
)
def
generate_girl
(
)
:
output
=
pixel_cnn
(
)
predict
=
tf
.
reshape
(
tf
.
multinomial
(
tf
.
nn
.
softmax
(
output
)
,
num_samples
=
1
,
seed
=
100
)
,
tf
.
shape
(
X
)
)
#predict_argmax = tf.reshape(tf.argmax(tf.nn.softmax(output), dimension=tf.rank(output) - 1), tf.shape(X))
with
tf
.
Session
(
)
as
sess
:
sess
.
run
(
tf
.
initialize_all_variables
(
)
)
saver
=
tf
.
train
.
Saver
(
tf
.
trainable_variables
(
)
)
saver
.
restore
(
sess
,
'girl.ckpt-49'
)
pics
=
np
.
zeros
(
(
1
*
1
,
image_height
,
image_width
,
image_channel
)
,
dtype
=
np
.
float32
)
for
i
in
range
(
image_height
)
:
for
j
in
range
(
image_width
)
:
for
k
in
range
(
image_channel
)
:
next_pic
=
sess
.
run
(
predict
,
feed_dict
=
{
X
:
pics
}
)
pics
[
:
,
i
,
j
,
k
]
=
next_pic
[
:
,
i
,
j
,
k
]
cv2
.
imwrite
(
'girl.jpg'
,
pics
[
0
]
)
print
(
'生成妹子图: girl.jpg'
)
# 生成图像
generate_girl
(
)
|
额，妹子图正在训练中…
补充练习：使用OpenCV提取图像中的脸，然后使用上面模型进行训练，看看能生成什么。
- Deep Generative Image Models using a Laplacian Pyramid of Adversarial Networks
- https://github.com/awentzonline/image-analogies
- https://github.com/ericjang/draw
7790
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
