forked from hanna-xu/MEF-GAN
-
Notifications
You must be signed in to change notification settings - Fork 8
/
Discriminator.py
149 lines (122 loc) · 5.88 KB
/
Discriminator.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 npimport
import tensorflow as tf
from tensorflow.python import pywrap_tensorflow
import numpy as np
import tensorflow as tf
from tensorflow.python import pywrap_tensorflow
import numpy as np
WEIGHT_INIT_STDDEV = 0.1
device = "/gpu:1"
class Discriminator(object):
def __init__(self, scope_name):
self.weight_vars = []
self.scope = scope_name
with tf.variable_scope(scope_name):
self.weight_vars.append(self._create_variables(3, 64, 3, scope = 'conv1'))
self.weight_vars.append(self._create_variables(64, 64, 3, scope = 'conv2'))
self.weight_vars.append(self._create_variables(64, 96, 3, scope = 'conv3'))
self.weight_vars.append(self._create_variables(96, 128, 3, scope = 'conv4'))
self.weight_vars.append(self._create_variables(128, 256, 3, scope = 'conv5'))
self.weight_vars.append(self._create_variables(256, 512, 3, scope = 'conv6'))
# self.weight_vars.append(self._create_variables(128, 256, 3, scope = 'conv5'))
# self.weight_vars.append(self._create_variables(256, 512, 3, scope = 'conv6'))
# self.weight_vars.append(self._create_variables(512, 512, 3, scope = 'conv7'))
# self.weight_vars.append(self._create_variables(12, 1, 3, scope = 'conv6'))
def _create_variables(self, input_filters, output_filters, kernel_size, scope):
shape = [kernel_size, kernel_size, input_filters, output_filters]
with tf.device("/cpu:0"):
with tf.variable_scope(scope):
kernel = tf.Variable(tf.truncated_normal(shape, stddev = WEIGHT_INIT_STDDEV), name = 'kernel')
bias = tf.Variable(tf.zeros([output_filters]), name = 'bias')
return (kernel, bias)
def discrim(self, img, reuse):
with tf.device(device):
conv_num = len(self.weight_vars)
if len(img.shape) != 4:
img = tf.expand_dims(img, -1)
out = img
for i in range(conv_num):
kernel, bias = self.weight_vars[i]
if i == 0:
out = conv2d(out, kernel, bias, [1, 2, 2, 1], use_relu = True, use_BN = False, sn=True,
Scope = self.scope + '/b' + str(i), Reuse = reuse)
# elif i == conv_num - 1:
# out = tf.nn.conv2d(out, kernel, [1, 1, 1, 1], padding = 'VALID')
# out = tf.nn.bias_add(out, bias)
# out = tf.nn.tanh(out)
# out = out / 2 + 0.5
elif i == conv_num - 1:
out = conv2d(out, kernel, bias, [1, 1, 1, 1], use_relu = True, use_BN = False, sn= False,
Scope = self.scope + '/b' + str(i), Reuse = reuse)
else:
out = conv2d(out, kernel, bias, [1, 2, 2, 1], use_relu = True, use_BN = False, sn = True,
Scope = self.scope + '/b' + str(i), Reuse = reuse)
# out = self_attention(out, channel_factor = 8, scope_name=self.scope, name = 'self_attention4', reuse= reuse)
out = tf.reshape(out, [-1, int(out.shape[1]) * int(out.shape[2]) * int(out.shape[3])])
with tf.variable_scope(self.scope):
with tf.variable_scope('flatten1'):
# out = tf.layers.dense(out, 512, activation = tf.nn.relu, use_bias = True, trainable = True,
# # reuse = reuse)
# out = tf.layers.batch_normalization(out, training = True, reuse = reuse)
# # with tf.variable_scope('flatten2'):
out = tf.layers.dense(out, 1, activation = tf.nn.tanh, use_bias = True, trainable = True,
reuse = reuse)
out = out / 2 + 0.5
return out
def conv2d(x, kernel, bias, strides, use_relu = True, use_BN = True, Scope = None, sn=True, Reuse = None):
with tf.device(device):
# padding image with reflection mode
x_padded = tf.pad(x, [[0, 0], [1, 1], [1, 1], [0, 0]], mode = 'REFLECT')
# conv and add bias
if sn:
out = tf.nn.conv2d(input=x_padded, filter = spectral_norm(kernel, scope_name = Scope, reuse=Reuse), strides=strides, padding = 'VALID')
else:
out = tf.nn.conv2d(input = x_padded, filter = kernel, strides = strides, padding = 'VALID')
out = tf.nn.bias_add(out, bias)
if use_BN:
with tf.variable_scope(Scope):
out = tf.layers.batch_normalization(out, training = True, reuse = Reuse)
if use_relu:
# out = tf.nn.relu(out)
out = tf.maximum(out, 0.2 * out)
return out
def self_attention(inputs, channel_factor = 8, scope_name = None, name = 'self_attention', reuse=False):
num_filters = inputs.shape[-1].value // channel_factor
with tf.variable_scope(scope_name):
with tf.variable_scope(name, reuse= reuse):
flat_inputs = tf.reshape(inputs, shape = [int(inputs.shape[0]), int(inputs.shape[1])*int(inputs.shape[2]), int(inputs.shape[-1])])
print('flat_inputs shape:', flat_inputs.shape)
f = tf.layers.conv1d(flat_inputs, kernel_size = 1, filters = num_filters)
g = tf.layers.conv1d(flat_inputs, kernel_size = 1, filters = num_filters)
h = tf.layers.conv1d(flat_inputs, kernel_size = 1, filters = inputs.shape[-1])
beta = tf.nn.softmax(tf.matmul(f, g, transpose_b = True))
o = tf.matmul(beta, h)
gamma = tf.get_variable('gamma', [], initializer = tf.zeros_initializer)
y = gamma * o + flat_inputs
y = tf.reshape(y, inputs.shape)
print('attention output shape:', y.shape)
return inputs
def spectral_norm(w, iteration = 1, scope_name = None, reuse=False):
w_shape = w.shape.as_list()
w = tf.reshape(w, [-1, w_shape[-1]])
with tf.variable_scope(scope_name, reuse=reuse):
u = tf.get_variable("u", [1, w_shape[-1]], initializer = tf.truncated_normal_initializer(), trainable = False)
u_hat = u
v_hat = None
for i in range(1):
"""
power iteration
Usually iteration = 1 will be enough
"""
v_ = tf.matmul(u_hat, tf.transpose(w))
v_hat = l2_norm(v_)
u_ = tf.matmul(v_hat, w)
u_hat = l2_norm(u_)
sigma = tf.matmul(tf.matmul(v_hat, w), tf.transpose(u_hat))
w_norm = w / sigma
with tf.control_dependencies([u.assign(u_hat)]):
w_norm = tf.reshape(w_norm, w_shape)
return w_norm
def l2_norm(v, eps=1e-12):
return v / (tf.reduce_sum(v ** 2) ** 0.5 + eps)