def up_convolution(layer_input, output_shape, in_channels, dropout_keep_prob):
with tf.variable_scope('up_convolution'):
x = deconvolution_3d(layer_input,
[2, 2, 2, in_channels // 2, in_channels],
output_shape, [1, 2, 2, 2, 1])
x = prelu(x)
x = tf.nn.dropout(x, dropout_keep_prob)
return x
def _up_convolution(input_, output_shape, in_channels):
"""
:param input_: data shape [batch, depth, height, width, in_channels]
:param output_shape: must match input dimensions [batch, depth, height, width, in_channels]
:param in_channels:
:return:
"""
with tf.variable_scope('up_convolution'):
return deconvolution_3d(input_, [2, 2, 2, in_channels, in_channels],
output_shape, [1, 2, 2, 2, 1])
def _up_convolution(layer_input, output_shape, in_channels):
with tf.variable_scope('up_convolution'):
return tf.nn.relu(
deconvolution_3d(layer_input,
[2, 2, 2, in_channels // 2, in_channels],
output_shape, [1, 2, 2, 2, 1]))