MobileNet4Style/model.py at master · daniel985/MobileNet4Style · 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
import tensorflow as tf

def conv2d(x, input_filters, output_filters, kernel, strides, mode='REFLECT'):
    with tf.variable_scope('conv') as scope:
        shape = [kernel, kernel, input_filters, output_filters]
        #weight = tf.Variable(tf.truncated_normal(shape, stddev=0.1))
        initializer = tf.contrib.layers.xavier_initializer()
        weight = tf.Variable(initializer(shape))
        #x_padded = tf.pad(x, [[0, 0], [kernel / 2, kernel / 2], [kernel / 2, kernel / 2], [0, 0]], mode=mode)
        return tf.nn.conv2d(x, weight, strides=[1, strides, strides, 1], padding='SAME', name='conv')

def conv2d_dw(x, input_filters, multiplier, kernel, strides, mode='REFLECT'):
    with tf.variable_scope('conv_dw') as scope:
        shape = [kernel, kernel, input_filters, multiplier]
        #weight = tf.Variable(tf.truncated_normal(shape, stddev=0.1))
        initializer = tf.contrib.layers.xavier_initializer()
        weight = tf.Variable(initializer(shape))
        #x_padded = tf.pad(x, [[0, 0], [kernel / 2, kernel / 2], [kernel / 2, kernel / 2], [0, 0]], mode=mode)
        conv_dw = tf.nn.depthwise_conv2d(x, weight, strides=[1, strides, strides, 1], padding='SAME', name='conv_dw')
        return conv_dw

def conv2d_transpose(x, input_filters, output_filters, kernel, strides):
    with tf.variable_scope('conv_transpose') as scope:

        shape = [kernel, kernel, output_filters, input_filters]
        #weight = tf.Variable(tf.truncated_normal(shape, stddev=0.1))
        initializer = tf.contrib.layers.xavier_initializer()
        weight = tf.Variable(initializer(shape))

        batch_size = tf.shape(x)[0]
        height = tf.shape(x)[1] * strides
        width = tf.shape(x)[2] * strides
        output_shape = tf.stack([batch_size, height, width, output_filters])
        return tf.nn.conv2d_transpose(x, weight, output_shape, strides=[1, strides, strides, 1], name='conv_transpose')


def resize_conv2d(x, input_filters, output_filters, kernel, strides, training):
    '''
    An alternative to transposed convolution where we first resize, then convolve.
    See http://distill.pub/2016/deconv-checkerboard/

    For some reason the shape needs to be statically known for gradient propagation
    through tf.image.resize_images, but we only know that for fixed image size, so we
    plumb through a "training" argument
    '''
    with tf.variable_scope('conv_transpose') as scope:
        height = x.get_shape()[1].value if training else tf.shape(x)[1]
        width = x.get_shape()[2].value if training else tf.shape(x)[2]

        new_height = height * strides * 2
        new_width = width * strides * 2

        x_resized = tf.image.resize_images(x, [new_height, new_width], tf.image.ResizeMethod.NEAREST_NEIGHBOR)

        return conv2d(x_resized, input_filters, output_filters, kernel, strides)

def resize(x, strides, training):
    with tf.variable_scope('conv_transpose') as scope:
        height = x.get_shape()[1].value if training else tf.shape(x)[1]
        width = x.get_shape()[2].value if training else tf.shape(x)[2]

        new_height = height * strides * 2
        new_width = width * strides * 2
        x_resized = tf.image.resize_images(x, [new_height, new_width], tf.image.ResizeMethod.NEAREST_NEIGHBOR)
        return x_resized

def instance_norm(x):
    epsilon = 1e-9

    mean, var = tf.nn.moments(x, [1, 2], keep_dims=True)

    return tf.div(tf.subtract(x, mean), tf.sqrt(tf.add(var, epsilon)))

def batch_norm(x, size, training, decay=0.999):
    beta = tf.Variable(tf.zeros([size]), name='beta')
    scale = tf.Variable(tf.ones([size]), name='scale')
    pop_mean = tf.Variable(tf.zeros([size]))
    pop_var = tf.Variable(tf.ones([size]))
    epsilon = 1e-3

    batch_mean, batch_var = tf.nn.moments(x, [0, 1, 2])
    train_mean = tf.assign(pop_mean, pop_mean * decay + batch_mean * (1 - decay))
    train_var = tf.assign(pop_var, pop_var * decay + batch_var * (1 - decay))

    def batch_statistics():
        with tf.control_dependencies([train_mean, train_var]):
            return tf.nn.batch_normalization(x, batch_mean, batch_var, beta, scale, epsilon, name='batch_norm')

    def population_statistics():
        return tf.nn.batch_normalization(x, pop_mean, pop_var, beta, scale, epsilon, name='batch_norm')

    return tf.cond(training, batch_statistics, population_statistics)


def residual(x, filters, kernel, strides):
    with tf.variable_scope('residual') as scope:
        conv1 = conv2d(x, filters, filters, kernel, strides)
        conv2 = conv2d(tf.nn.relu(conv1), filters, filters, kernel, strides)

        residual = x + conv2

        return residual

def mresidual(x, filters, kernel, strides):
    with tf.variable_scope('mresidual') as scope:
        conv1_dw = conv2d_dw(x, filters, 1, kernel, strides)
        conv1 = conv2d(conv1_dw, filters, filters, 1, 1)

        #conv2_dw = conv2d_dw(tf.nn.relu(conv1), filters, 1, kernel, strides)
        #conv2 = conv2d(conv2_dw, filters, filters, 1, 1)
        return x + conv1

def net(image, alpha):
    with tf.variable_scope('conv1'):
        mnet = tf.nn.relu(instance_norm(conv2d(image, 3, int(32*alpha), 3, 1)))

    with tf.variable_scope('conv2_dw'):
        mnet = tf.nn.relu(instance_norm(conv2d_dw(mnet, int(32*alpha), 1, 3, 2)))
    with tf.variable_scope('conv2'):
        mnet = tf.nn.relu(instance_norm(conv2d(mnet, int(32*alpha), int(64*alpha), 1, 1)))
    with tf.variable_scope('conv3_dw'):
        mnet = tf.nn.relu(instance_norm(conv2d_dw(mnet, int(64*alpha), 1, 3, 2)))
    with tf.variable_scope('conv3'):
        mnet = tf.nn.relu(instance_norm(conv2d(mnet, int(64*alpha), int(128*alpha), 1, 1)))

    #with tf.variable_scope('res1'):
    #    mnet = mresidual(mnet, int(128*alpha), 3, 1)
    #with tf.variable_scope('res2'):
    #    mnet = mresidual(mnet, int(128*alpha), 3, 1)
    #with tf.variable_scope('res3'):
    #    mnet = mresidual(mnet, int(128*alpha), 3, 1)
    #with tf.variable_scope('res4'):
    #    mnet = mresidual(mnet, int(128*alpha), 3, 1)
    #with tf.variable_scope('res5'):
    #    mnet = mresidual(mnet, int(128*alpha), 3, 1)

    with tf.variable_scope('conv4_dw'):
        mnet = tf.nn.relu(instance_norm(conv2d_dw(mnet, int(128*alpha), 1, 3, 1)))
    with tf.variable_scope('conv4'):
        mnet = tf.nn.relu(instance_norm(conv2d(mnet, int(128*alpha), int(128*alpha), 1, 1)))
    with tf.variable_scope('conv5_dw'):
        mnet = tf.nn.relu(instance_norm(conv2d_dw(mnet, int(128*alpha), 1, 3, 1)))
    with tf.variable_scope('conv5'):
        mnet = tf.nn.relu(instance_norm(conv2d(mnet, int(128*alpha), int(128*alpha), 1, 1)))
    with tf.variable_scope('conv6_dw'):
        mnet = tf.nn.relu(instance_norm(conv2d_dw(mnet, int(128*alpha), 1, 3, 1)))
    with tf.variable_scope('conv6'):
        mnet = tf.nn.relu(instance_norm(conv2d(mnet, int(128*alpha), int(128*alpha), 1, 1)))

    with tf.variable_scope('deconv1_dw'):
        mnet = resize(mnet, 2, training)
        mnet = tf.nn.relu(instance_norm(conv2d_dw(mnet, int(128*alpha), 1, 3, 2)))
    with tf.variable_scope('deconv1'):
        mnet = tf.nn.relu(instance_norm(conv2d(mnet, int(128*alpha), int(64*alpha), 1, 1)))
    with tf.variable_scope('deconv2_dw'):
        mnet = resize(mnet, 2, training)
        mnet = tf.nn.relu(instance_norm(conv2d_dw(mnet, int(64*alpha), 1, 3, 2)))
    with tf.variable_scope('deconv2'):
        mnet = tf.nn.relu(instance_norm(conv2d(mnet, int(64*alpha), int(32*alpha), 1, 1)))

    with tf.variable_scope('final'):
        mnet = tf.nn.tanh(instance_norm(conv2d(mnet, int(32*alpha), 3, 3, 1)))

    y = (mnet + 1) * 127.5

    return y