def transpose_deconvolution_layer(input_tensor, used_weights, new_shape,
stride, scope_name):
with tf.varaible_scope(scope_name):
output = tf.nn.conv2d_transpose(input_tensor,
used_weights,
output_shape=new_shape,
strides=[1, stride, stride, 1],
padding='SAME')
output = tf.nn.relu(output)
return output
def SwitchNorm(input,data_format="NHWC",eps=1e-5):
x = copy.deepcopy(input)
if data_format =="NHWC":
x = tf.transpose(x,[0,3,1,2])
n,c,h,w = x.get_shape().as_list()
x = tf.reshape(x,[n,c,-1])
else:
n,c,h,w = x.get_shape().as_list()
x = tf.rehape(x,[n,c,-1])
weight = tf.get_variable("weight",[1,c,1,1],dtype=tf.float32,initializer=tf.ones_initializer())
bias = tf.get_variable("bias",[1,c,1,1],dtype=tf.float32,initializer=tf.zeros_initializer())
mean_weight = tf.get_variable("mean_weight",[3],dtype=tf.float32,initializer=tf.ones_initialzier())
var_weight = tf.get_variable("var_weight",[3],dtype=tf.float32,initializer=tf.ones_initialzier())
with tf.variable_scope("in"):
# w.h set to 1
mean_in,var_in = tf.nn.moments(x,axes=-1,keep_dims=True)
#diffenrent from pytorch version,which use approximation
#mean_ln = mean_in.mean(1, keepdim=True)
#temp = var_in + mean_in ** 2
#var_ln = temp.mean(1, keepdim=True) - mean_ln ** 2
with tf.varaible_scope("ln"):
# c set to 1
mean_ln,var_ln = tf.nn.moments(x,axes=1,keep_dims=True)
with tf.variable_scope("bn"):
#set n,w,h to 1
mean_bn,var_bn = tf.nn.moments(x,axes=[0,2],keep_dims=1)
mean_weight = tf.nn.softmax(mean_weight)
var_weight = tf.nn.softmax(var_weight)
mean = mean_weight[0]*mean_in + mean_weight[1]*mean_ln + mean_weight[2]*mean_bn
var = var_weight[0]*var_in + var_weight[1]*var_ln + var_weight[2]*var_bn
x = (x-mean)/tf.sqrt((var+eps))
x = tf.reshape(x,[n,h,w,c])
x = x*weight+bias
return x
def share_generator(self, x, is_training=True, reuse=False, scope='share_generator'):
channel = self.ch * pow(2, self.n_encoder - 1)
with tensorflow.varaible_scope(scope, reuse=reuse):
for i in range(0, self.n_gen_share):
x = res_block(x, channel, kernel=3, stride=1, pad=1, dropout_ratio=self.res_dropout, normal_weight_init=self.normal_weight_init, is_training=is_training, norm_fn=self.norm, scope='res_block' + str(i))
return x