UNet¶

Inheritance Diagram

class ashpy.models.convolutional.unet.UNet(input_res, min_res, kernel_size, initial_filters, filters_cap, channels, use_dropout_encoder=True, use_dropout_decoder=True, dropout_prob=0.3, encoder_non_linearity=<class 'tensorflow.python.keras.layers.advanced_activations.LeakyReLU'>, decoder_non_linearity=<class 'tensorflow.python.keras.layers.advanced_activations.ReLU'>, normalization_layer=<class 'ashpy.layers.instance_normalization.InstanceNormalization'>, last_activation=<function tanh>, use_attention=False)[source]¶

Bases: ashpy.models.convolutional.interfaces.Conv2DInterface

UNet Architecture.

Architecture similar to the one found in “Image-to-Image Translation with Conditional Adversarial Nets” [1].

Originally proposed in “U-Net: Convolutional Networks for Biomedical Image Segmentation” [2].

Examples

Direct Usage:

x = tf.ones((1, 512, 512, 3))
u_net = UNet(input_res = 512,
             min_res=4,
             kernel_size=4,
             initial_filters=64,
             filters_cap=512,
             channels=3)
y = u_net(x)
print(y.shape)
print(len(u_net.trainable_variables)>0)

(1, 512, 512, 3)
True

[1]	Image-to-Image Translation with Conditional Adversarial Nets - https://arxiv.org/abs/1611.07004

[2]	U-Net: Convolutional Networks for Biomedical Image Segmentation - https://arxiv.org/abs/1505.04597

Methods

`__init__`(input_res, min_res, kernel_size, …)	Initialize the UNet.
`call`(inputs[, training])	Forward pass of the UNet model.
`get_decoder_block`(filters[, use_bn, …])	Return a block to be used in the decoder part of the UNET.
`get_encoder_block`(filters[, use_bn, …])	Return a block to be used in the encoder part of the UNET.

Attributes

`activity_regularizer`	Optional regularizer function for the output of this layer.
`dtype`
`dynamic`
`inbound_nodes`	Deprecated, do NOT use! Only for compatibility with external Keras.
`input`	Retrieves the input tensor(s) of a layer.
`input_mask`	Retrieves the input mask tensor(s) of a layer.
`input_shape`	Retrieves the input shape(s) of a layer.
`input_spec`	Gets the network’s input specs.
`layers`
`losses`	Losses which are associated with this Layer.
`metrics`	Returns the model’s metrics added using compile, add_metric APIs.
`metrics_names`	Returns the model’s display labels for all outputs.
`name`	Returns the name of this module as passed or determined in the ctor.
`name_scope`	Returns a tf.name_scope instance for this class.
`non_trainable_variables`
`non_trainable_weights`
`outbound_nodes`	Deprecated, do NOT use! Only for compatibility with external Keras.
`output`	Retrieves the output tensor(s) of a layer.
`output_mask`	Retrieves the output mask tensor(s) of a layer.
`output_shape`	Retrieves the output shape(s) of a layer.
`run_eagerly`	Settable attribute indicating whether the model should run eagerly.
`sample_weights`
`state_updates`	Returns the updates from all layers that are stateful.
`stateful`
`submodules`	Sequence of all sub-modules.
`trainable`
`trainable_variables`	Sequence of variables owned by this module and it’s submodules.
`trainable_weights`
`updates`
`variables`	Returns the list of all layer variables/weights.
`weights`	Returns the list of all layer variables/weights.

__init__(input_res, min_res, kernel_size, initial_filters, filters_cap, channels, use_dropout_encoder=True, use_dropout_decoder=True, dropout_prob=0.3, encoder_non_linearity=<class 'tensorflow.python.keras.layers.advanced_activations.LeakyReLU'>, decoder_non_linearity=<class 'tensorflow.python.keras.layers.advanced_activations.ReLU'>, normalization_layer=<class 'ashpy.layers.instance_normalization.InstanceNormalization'>, last_activation=<function tanh>, use_attention=False)[source]¶

Initialize the UNet.

Parameters:

input_res (int) – input resolution.
min_res (int) – minimum resolution reached after decode.
kernel_size (int) – kernel size used in the network.
initial_filters (int) – number of filter of the initial convolution.
filters_cap (int) – maximum number of filters.
channels (int) – number of output channels.
use_dropout_encoder (bool) – whether to use dropout in the encoder module.
use_dropout_decoder (bool) – whether to use dropout in the decoder module.
dropout_prob (float) – probability of dropout.
encoder_non_linearity (Type[Layer]) – non linearity of encoder.
decoder_non_linearity (Type[Layer]) – non linearity of decoder.
last_activation (<module 'tensorflow_core.keras.activations' from '/home/docs/checkouts/readthedocs.org/user_builds/ashpy/envs/master/lib/python3.7/site-packages/tensorflow_core/python/keras/api/_v2/keras/activations/__init__.py'>) – last activation function, tanh or softmax (for semantic images).
use_attention (bool) – whether to use attention.

call(inputs, training=False)[source]¶: Forward pass of the UNet model.

get_decoder_block(filters, use_bn=True, use_dropout=False, use_attention=False)[source]¶

Return a block to be used in the decoder part of the UNET.

Parameters:	filters – number of filters use_bn – whether to use batch normalization use_dropout – whether to use dropout use_attention – whether to use attention
Returns:	A block to be used in the decoder part

get_encoder_block(filters, use_bn=True, use_attention=False)[source]¶

Return a block to be used in the encoder part of the UNET.

Parameters:	filters – number of filters. use_bn – whether to use batch normalization. use_attention – whether to use attention.
Returns:	A block to be used in the encoder part.