-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathops.py
221 lines (187 loc) · 8.84 KB
/
ops.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
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
import math
import numpy as np
import tensorflow as tf
from tensorflow.python.framework import ops
from utils import *
class VBN(object):
"""
Virtual Batch Normalization
"""
def __init__(self, x, name, epsilon=1e-5, half=None):
"""
x is the reference batch
"""
assert isinstance(epsilon, float)
self.half = half
shape = x.get_shape().as_list()
needs_reshape = len(shape) != 4
if needs_reshape:
orig_shape = shape
if len(shape) == 2:
x = tf.reshape(x, [shape[0], 1, 1, shape[1]])
elif len(shape) == 1:
x = tf.reshape(x, [shape[0], 1, 1, 1])
else:
assert False, shape
shape = x.get_shape().as_list()
with tf.variable_scope(name) as scope:
assert name.startswith("d_") or name.startswith("g_")
self.epsilon = epsilon
self.name = name
if self.half is None:
half = x
elif self.half == 1:
half = tf.slice(x, [0, 0, 0, 0],
[shape[0] // 2, shape[1], shape[2], shape[3]])
elif self.half == 2:
half = tf.slice(x, [shape[0] // 2, 0, 0, 0],
[shape[0] // 2, shape[1], shape[2], shape[3]])
else:
assert False
self.mean = tf.reduce_mean(half, [0, 1, 2], keep_dims=True)
self.mean_sq = tf.reduce_mean(tf.square(half), [0, 1, 2], keep_dims=True)
self.batch_size = int(half.get_shape()[0])
assert x is not None
assert self.mean is not None
assert self.mean_sq is not None
out = self._normalize(x, self.mean, self.mean_sq, "reference")
if needs_reshape:
out = tf.reshape(out, orig_shape)
self.reference_output = out
def __call__(self, x):
shape = x.get_shape().as_list()
needs_reshape = len(shape) != 4
if needs_reshape:
orig_shape = shape
if len(shape) == 2:
x = tf.reshape(x, [shape[0], 1, 1, shape[1]])
elif len(shape) == 1:
x = tf.reshape(x, [shape[0], 1, 1, 1])
else:
assert False, shape
shape = x.get_shape().as_list()
with tf.variable_scope(self.name) as scope:
new_coeff = 1. / (self.batch_size + 1.)
old_coeff = 1. - new_coeff
new_mean = tf.reduce_mean(x, [1, 2], keep_dims=True)
new_mean_sq = tf.reduce_mean(tf.square(x), [0, 1, 2], keep_dims=True)
mean = new_coeff * new_mean + old_coeff * self.mean
mean_sq = new_coeff * new_mean_sq + old_coeff * self.mean_sq
out = self._normalize(x, mean, mean_sq, "live")
if needs_reshape:
out = tf.reshape(out, orig_shape)
return out
def _normalize(self, x, mean, mean_sq, message):
# make sure this is called with a variable scope
shape = x.get_shape().as_list()
assert len(shape) == 4
self.gamma = tf.get_variable("gamma", [shape[-1]],
initializer=tf.random_normal_initializer(1., 0.02))
gamma = tf.reshape(self.gamma, [1, 1, 1, -1])
self.beta = tf.get_variable("beta", [shape[-1]],
initializer=tf.constant_initializer(0.))
beta = tf.reshape(self.beta, [1, 1, 1, -1])
assert self.epsilon is not None
assert mean_sq is not None
assert mean is not None
std = tf.sqrt(self.epsilon + mean_sq - tf.square(mean))
out = x - mean
out = out / std
# out = tf.Print(out, [tf.reduce_mean(out, [0, 1, 2]),
# tf.reduce_mean(tf.square(out - tf.reduce_mean(out, [0, 1, 2], keep_dims=True)), [0, 1, 2])],
# message, first_n=-1)
out = out * gamma
out = out + beta
return out
class batch_norm(object):
"""Code modification of http://stackoverflow.com/a/33950177"""
def __init__(self, epsilon=1e-5, momentum = 0.9, name="batch_norm"):
with tf.variable_scope(name):
self.epsilon = epsilon
self.momentum = momentum
self.ema = tf.train.ExponentialMovingAverage(decay=self.momentum)
self.name = name
def __call__(self, x, train=True):
shape = x.get_shape().as_list()
if train:
with tf.variable_scope(self.name) as scope:
self.beta = tf.get_variable("beta", [shape[-1]],
initializer=tf.constant_initializer(0.))
self.gamma = tf.get_variable("gamma", [shape[-1]],
initializer=tf.random_normal_initializer(1., 0.02))
try:
batch_mean, batch_var = tf.nn.moments(x, [0, 1, 2], name='moments')
except:
batch_mean, batch_var = tf.nn.moments(x, [0, 1], name='moments')
ema_apply_op = self.ema.apply([batch_mean, batch_var])
self.ema_mean, self.ema_var = self.ema.average(batch_mean), self.ema.average(batch_var)
with tf.control_dependencies([ema_apply_op]):
mean, var = tf.identity(batch_mean), tf.identity(batch_var)
else:
mean, var = self.ema_mean, self.ema_var
normed = tf.nn.batch_norm_with_global_normalization(
x, mean, var, self.beta, self.gamma, self.epsilon, scale_after_normalization=True)
return normed
def binary_cross_entropy(preds, targets, name=None):
"""Computes binary cross entropy given `preds`.
For brevity, let `x = `, `z = targets`. The logistic loss is
loss(x, z) = - sum_i (x[i] * log(z[i]) + (1 - x[i]) * log(1 - z[i]))
Args:
preds: A `Tensor` of type `float32` or `float64`.
targets: A `Tensor` of the same type and shape as `preds`.
"""
eps = 1e-12
with ops.op_scope([preds, targets], name, "bce_loss") as name:
preds = ops.convert_to_tensor(preds, name="preds")
targets = ops.convert_to_tensor(targets, name="targets")
return tf.reduce_mean(-(targets * tf.log(preds + eps) +
(1. - targets) * tf.log(1. - preds + eps)))
def conv_cond_concat(x, y):
"""Concatenate conditioning vector on feature map axis."""
x_shapes = x.get_shape()
y_shapes = y.get_shape()
print(x_shapes, y_shapes)
return tf.concat(3, [x, y*tf.ones([x_shapes[0], x_shapes[1], x_shapes[2], y_shapes[3]])])
def conv2d(input_, output_dim,
k_h=5, k_w=5, d_h=2, d_w=2, stddev=0.02,
name="conv2d"):
with tf.variable_scope(name):
w = tf.get_variable('w', [k_h, k_w, input_.get_shape()[-1], output_dim],
initializer=tf.truncated_normal_initializer(stddev=stddev))
conv = tf.nn.conv2d(input_, w, strides=[1, d_h, d_w, 1], padding='SAME')
biases = tf.get_variable('biases', [output_dim], initializer=tf.constant_initializer(0.0))
conv = tf.reshape(tf.nn.bias_add(conv, biases), conv.get_shape())
return conv
def deconv2d(input_, output_shape,
k_h=5, k_w=5, d_h=2, d_w=2, stddev=0.02,
name="deconv2d", with_w=False):
with tf.variable_scope(name):
# filter : [height, width, output_channels, in_channels]
w = tf.get_variable('w', [k_h, k_w, output_shape[-1], input_.get_shape()[-1]],
initializer=tf.random_normal_initializer(stddev=stddev))
try:
deconv = tf.nn.conv2d_transpose(input_, w, output_shape=output_shape,
strides=[1, d_h, d_w, 1])
# Support for verisons of TensorFlow before 0.7.0
except AttributeError:
deconv = tf.nn.deconv2d(input_, w, output_shape=output_shape,
strides=[1, d_h, d_w, 1])
biases = tf.get_variable('biases', [output_shape[-1]], initializer=tf.constant_initializer(0.0))
deconv = tf.reshape(tf.nn.bias_add(deconv, biases), deconv.get_shape())
if with_w:
return deconv, w, biases
else:
return deconv
def lrelu(x, leak=0.2, name="lrelu"):
return tf.maximum(x, leak*x)
def linear(input_, output_size, scope=None, stddev=0.02, bias_start=0.0, with_w=False):
shape = input_.get_shape().as_list()
with tf.variable_scope(scope or "Linear"):
matrix = tf.get_variable("Matrix", [shape[1], output_size], tf.float32,
tf.random_normal_initializer(stddev=stddev))
bias = tf.get_variable("bias", [output_size],
initializer=tf.constant_initializer(bias_start))
if with_w:
return tf.matmul(input_, matrix) + bias, matrix, bias
else:
return tf.matmul(input_, matrix) + bias