niftynet.layer.base_layer module

class niftynet.layer.base_layer.DataDependentLayer(name='data_dependent_op')

Bases: niftynet.layer.base_layer.Layer

Some layers require a one-pass training through the training set to determine their internal models, this abstract provides interfaces for training these internal models and querying the status.

is_ready()

train(*args, **kwargs)

class niftynet.layer.base_layer.Invertible

Bases: object

interface of Invertible data

inverse_op(*args, **kwargs)

class niftynet.layer.base_layer.Layer(name='untitled_op')

Bases: object

layer_op(*args, **kwargs)

layer_scope()

to_string()

class niftynet.layer.base_layer.LayerFromCallable(layer_op, name='untitled_op')

Bases: niftynet.layer.base_layer.Layer

Module wrapping a function provided by the user. Analogous to snt.Module

layer_op(*args, **kwargs)

class niftynet.layer.base_layer.RandomisedLayer(name='randomised_op')

Bases: niftynet.layer.base_layer.Layer

The layers require a randomisation process, to randomly change some of the layer’s states on the fly.

randomise(*args, **kwargs)

class niftynet.layer.base_layer.TrainableLayer(name='trainable_op')

Bases: niftynet.layer.base_layer.Layer

Extends the Layer object to have trainable parameters, adding intiailizers and regularizers.

initializers

num_trainable_params()

regularizer_loss()

regularizers

restore_from_checkpoint(checkpoint_name, scope=None)

to_string()

trainable_variables()