import tensorflow as tf
import numpy as np
import cPickle
import ipdb
class Model():
def __init__(self):
pass
def new_conv_layer( self, bottom, filter_shape, activation=tf.identity, padding='SAME', stride=1, name=None ):
with tf.variable_scope( name ):
w = tf.get_variable(
"W",
shape=filter_shape,
initializer=tf.random_normal_initializer(0., 0.005))
b = tf.get_variable(
"b",
shape=filter_shape[-1],
initializer=tf.constant_initializer(0.))
conv = tf.nn.conv2d( bottom, w, [1,stride,stride,1], padding=padding)
bias = activation(tf.nn.bias_add(conv, b))
return bias #relu
def new_deconv_layer(self, bottom, filter_shape, output_shape, activation=tf.identity, padding='SAME', stride=1, name=None):
with tf.variable_scope(name):
W = tf.get_variable(
"W",
shape=filter_shape,
initializer=tf.random_normal_initializer(0., 0.005))
b = tf.get_variable(
"b",
shape=filter_shape[-2],
initializer=tf.constant_initializer(0.))
deconv = tf.nn.conv2d_transpose( bottom, W, output_shape, [1,stride,stride,1], padding=padding)
bias = activation(tf.nn.bias_add(deconv, b))
return bias
def new_fc_layer( self, bottom, output_size, name ):
shape = bottom.get_shape().as_list()
dim = np.prod( shape[1:] )
x = tf.reshape( bottom, [-1, dim])
input_size = dim
with tf.variable_scope(name):
w = tf.get_variable(
"W",
shape=[input_size, output_size],
initializer=tf.random_normal_initializer(0., 0.005))
b = tf.get_variable(
"b",
shape=[output_size],
initializer=tf.constant_initializer(0.))
fc = tf.nn.bias_add( tf.matmul(x, w), b)
return fc
def channel_wise_fc_layer(self, input, name): # bottom: (7x7x512)
_, width, height, n_feat_map = input.get_shape().as_list()
input_reshape = tf.reshape( input, [-1, width*height, n_feat_map] )
input_transpose = tf.transpose( input_reshape, [2,0,1] )
with tf.variable_scope(name):
W = tf.get_variable(
"W",
shape=[n_feat_map,width*height, width*height], # (512,49,49)
initializer=tf.random_normal_initializer(0., 0.005))
output = tf.batch_matmul(input_transpose, W)
output_transpose = tf.transpose(output, [1,2,0])
output_reshape = tf.reshape( output_transpose, [-1, height, width, n_feat_map] )
return output_reshape
def leaky_relu(self, bottom, leak=0.1):
return tf.maximum(leak*bottom, bottom)
def batchnorm(self, bottom, is_train, epsilon=1e-8, name=None):
bottom = tf.clip_by_value( bottom, -100., 100.)
depth = bottom.get_shape().as_list()[-1]
with tf.variable_scope(name):
gamma = tf.get_variable("gamma", [depth], initializer=tf.constant_initializer(1.))
beta = tf.get_variable("beta" , [depth], initializer=tf.constant_initializer(0.))
batch_mean, batch_var = tf.nn.moments(bottom, [0,1,2], name='moments')
ema = tf.train.ExponentialMovingAverage(decay=0.5)
def update():
with tf.control_dependencies([ema_apply_op]):
return tf.identity(batch_mean), tf.identity(batch_var)
ema_apply_op = ema.apply([batch_mean, batch_var])
ema_mean, ema_var = ema.average(batch_mean), ema.average(batch_var)
mean, var = tf.cond(
is_train,
update,
lambda: (ema_mean, ema_var) )
normed = tf.nn.batch_norm_with_global_normalization(bottom, mean, var, beta, gamma, epsilon, False)
return normed
def build_reconstruction( self, images, is_train ):
batch_size = images.get_shape().as_list()[0]
with tf.variable_scope('GEN'):
conv1 = self.new_conv_layer(images, [4,4,3,64], stride=2, name="conv1" )
bn1 = self.leaky_relu(self.batchnorm(conv1, is_train, name='bn1'))
conv2 = self.new_conv_layer(bn1, [4,4,64,64], stride=2, name="conv2" )
bn2 = self.leaky_relu(self.batchnorm(conv2, is_train, name='bn2'))
conv3 = self.new_conv_layer(bn2, [4,4,64,128], stride=2, name="conv3")
bn3 = self.leaky_relu(self.batchnorm(conv3, is_train, name='bn3'))
conv4 = self.new_conv_layer(bn3, [4,4,128,256], stride=2, name="conv4")
bn4 = self.leaky_relu(self.batchnorm(conv4, is_train, name='bn4'))
conv5 = self.new_conv_layer(bn4, [4,4,256,512], stride=2, name="conv5")
bn5 = self.leaky_relu(self.batchnorm(conv5, is_train, name='bn5'))
conv6 = self.new_conv_layer(bn5, [4,4,512,4000], stride=2, padding='VALID', name='conv6')
bn6 = self.leaky_relu(self.batchnorm(conv6, is_train, name='bn6'))
deconv4 = self.new_deconv_layer( bn6, [4,4,512,4000], conv5.get_shape().as_list(), padding='VALID', stride=2, name="deconv4")
debn4 = tf.nn.relu(self.batchnorm(deconv4, is_train, name='debn4'))
deconv3 = self.new_deconv_layer( debn4, [4,4,256,512], conv4.get_shape().as_list(), stride=2, name="deconv3")
debn3 = tf.nn.relu(self.batchnorm(deconv3, is_train, name='debn3'))
deconv2 = self.new_deconv_layer( debn3, [4,4,128,256], conv3.get_shape().as_list(), stride=2, name="deconv2")
debn2 = tf.nn.relu(self.batchnorm(deconv2, is_train, name='debn2'))
deconv1 = self.new_deconv_layer( debn2, [4,4,64,128], conv2.get_shape().as_list(), stride=2, name="deconv1")
debn1 = tf.nn.relu(self.batchnorm(deconv1, is_train, name='debn1'))
recon = self.new_deconv_layer( debn1, [4,4,3,64], [batch_size,64,64,3], stride=2, name="recon")
return bn1, bn2, bn3, bn4, bn5, bn6, debn4, debn3, debn2, debn1, recon, tf.nn.tanh(recon)
def build_adversarial(self, images, is_train, reuse=None):
with tf.variable_scope('DIS', reuse=reuse):
conv1 = self.new_conv_layer(images, [4,4,3,64], stride=2, name="conv1" )
bn1 = self.leaky_relu(self.batchnorm(conv1, is_train, name='bn1'))
conv2 = self.new_conv_layer(bn1, [4,4,64,128], stride=2, name="conv2")
bn2 = self.leaky_relu(self.batchnorm(conv2, is_train, name='bn2'))
conv3 = self.new_conv_layer(bn2, [4,4,128,256], stride=2, name="conv3")
bn3 = self.leaky_relu(self.batchnorm(conv3, is_train, name='bn3'))
conv4 = self.new_conv_layer(bn3, [4,4,256,512], stride=2, name="conv4")
bn4 = self.leaky_relu(self.batchnorm(conv4, is_train, name='bn4'))
output = self.new_fc_layer( bn4, output_size=1, name='output')
return output[:,0]
