dawsonia.ml.network.layers

Gated implementations GatedConv2D: Introduce a Conv2D layer (same number of filters) to multiply with its sigmoid activation.

FullGatedConv2D: Introduce a Conv2D to extract features (linear and
sigmoid), making a full gated process.  This process will double number of
filters to make one convolutional process.

Module Contents

Classes

GatedConv2D	Gated Convolutional Class.
FullGatedConv2D	Gated Convolutional Class.
OctConv2D	Octave Convolutional Class.

API

class dawsonia.ml.network.layers.GatedConv2D(**kwargs)

Bases: tensorflow.keras.layers.Conv2D

Gated Convolutional Class.

Initialization

call(inputs)

Apply gated convolution.

get_config()

Return the config of the layer.

class dawsonia.ml.network.layers.FullGatedConv2D(filters, **kwargs)

Bases: tensorflow.keras.layers.Conv2D

Gated Convolutional Class.

Initialization

call(inputs)

Apply gated convolution.

compute_output_shape(input_shape)

Compute shape of layer output.

get_config()

Return the config of the layer.

class dawsonia.ml.network.layers.OctConv2D(filters, alpha, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_initializer='glorot_uniform', kernel_regularizer=None, kernel_constraint=None, **kwargs)

Bases: tensorflow.keras.layers.Layer

Octave Convolutional Class.

Initialization

build(input_shape)

call(inputs)

compute_output_shape(input_shapes)

get_config()