DeepD3 model

builder

deepd3.model.builder.DeepD3_Model(filters=32, input_shape=(128, 128, 1), layers=4, activation='swish')

DeepD3 TensorFlow Keras Model. It defines the architecture, together with the single encoder and dual decoders.

Parameters

• filters (int, optional) – Base filter multiplier. Defaults to 32.

• input_shape (tuple, optional) – Image shape for training. Defaults to (128, 128, 1).

• layers (int, optional) – Network depth layers. Defaults to 4.

• activation (str, optional) – Activation function used in convolutional layers. Defaults to “swish”.

Returns

function TensorFlow/Keras model

Return type

Model

deepd3.model.builder.convlayer(x, filters, activation, name, residual=None, use_batchnorm=True)

Convolutional layer with normalization and residual connection

Parameters

• x (Keras.layer) – input layer

• filters (int) – filters used in convolutional layer

• activation (str) – Activation function

• name (str) – Description of layer

• residual (Keras.layer, optional) – Residual layer. Defaults to None.

• use_batchnorm (bool, optional) – Use of batch normalization. Defaults to True.

Returns

Full convolutional procedure

Return type

Keras.layer

deepd3.model.builder.decoder(x, filters, layers, to_concat, name, activation)

Decoder for neural network.

Parameters

• x (Keras layer) – Start of decoder, normally the latent space

• filters (int) – The filter multiplier

• layers (int) – Depth layers to be used for upsampling

• to_concat (list) – Encoder layers to be concatenated

• name (str) – Description of the decoder

• activation (str) – Activation function used in Decoder

Returns

Full decoder across layers

Return type

Keras layer

deepd3.model.builder.identity(x, filters, name)

Identity layer for residual layers

Parameters

• x (Keras.layer) – Keras layer

• filters (int) – Used filters

• name (str) – Layer description

Returns

Identity layer

Return type

Keras.layer