niftynet.layer.activation module

class niftynet.layer.activation.ActiLayer(func, regularizer=None, name='activation')

Bases: niftynet.layer.base_layer.TrainableLayer

Apply an element-wise non-linear activation function. ‘Prelu’ uses trainable parameters and those are initialised to zeros Dropout function is also supported

layer_op(input_tensor, keep_prob=None)

niftynet.layer.activation.leakyRelu(x, name)

niftynet.layer.activation.prelu(f_in, channelwise_params)

niftynet.layer.activation.selu(x, name)