def convolution(
inputs,
num_outputs,
kernel_size,
stride=1,
padding='SAME',
data_format=None,
rate=1,
activation_fn=nn.relu,
normalizer_fn=None,
normalizer_params=None,
weights_initializer=initializers.xavier_initializer(),
weights_regularizer=None,
biases_initializer=init_ops.zeros_initializer(),
biases_regularizer=None,
reuse=None,
variables_collections=None,
outputs_collections=None,
trainable=True,
scope=None,
):
scope = _default_scope(scope, 'Conv')
if data_format not in [None, 'NHWC', 'NCHW']:
raise ValueError('Invalid data_format: %r' % (data_format,))
data_format = 'channels_first' if data_format == 'NCHW' else 'channels_last'
input_rank = inputs.get_shape().ndims
with vs.variable_scope(scope, reuse=reuse) as sc:
if input_rank == 4:
return layers.conv2d(
inputs=inputs,
filters=num_outputs,
kernel_size=kernel_size,
strides=stride,
padding=padding,
data_format=data_format,
dilation_rate=rate,
activation=activation_fn,
use_bias=True if biases_initializer is not None else False,
kernel_initializer=weights_initializer,
bias_initializer=biases_initializer,
bias_regularizer=biases_regularizer,
activity_regularizer=None,
trainable=trainable,
reuse=reuse,
)
def convolution(inputs,
num_outputs,
kernel_size,
stride=1,
padding='SAME',
data_format=None,
rate=1,
activation_fn=nn.relu,
normalizer_fn=None,
normalizer_params=None,
weights_initializer=initializers.xavier_initializer(),
weights_regularizer=None,
biases_initializer=init_ops.zeros_initializer(),
biases_regularizer=None,
reuse=None,
variables_collections=None,
outputs_collections=None,
trainable=True,
scope=None):
scope = _default_scope(scope, 'CONVOLUTION', 'Conv')
if data_format not in [None, 'NHWC', 'NCHW']:
raise ValueError('Invalid data_format: %r' % (data_format,))
data_format = 'channels_first' if data_format == 'NCHW' else 'channels_last'
input_rank = inputs.get_shape().ndims
with vs.variable_scope(scope, reuse=reuse) as sc:
if input_rank == 4:
return layers.conv2d(
inputs=inputs,
filters=num_outputs,
kernel_size=kernel_size,
strides=stride,
padding=padding,
data_format=data_format,
dilation_rate=rate,
activation=activation_fn,
use_bias=True if biases_initializer is not None else False,
kernel_initializer=weights_initializer,
bias_initializer=biases_initializer,
bias_regularizer=biases_regularizer,
activity_regularizer=None,
trainable=trainable,
reuse=reuse)