niftynet.layer.deconvolution module

default_w_initializer()[source]
default_b_initializer()[source]
infer_output_dims(input_dims, strides, kernel_sizes, padding)[source]

infer output dims from list, the dim can be different in different directions. Note: dilation is not considered here.

class DeconvLayer(n_output_chns, kernel_size=3, stride=1, padding='SAME', with_bias=False, w_initializer=None, w_regularizer=None, b_initializer=None, b_regularizer=None, name='deconv')[source]

Bases: niftynet.layer.base_layer.TrainableLayer

This class defines a simple deconvolution with an optional bias term. Please consider DeconvolutionalLayer if batch_norm and activation are also used.

layer_op(input_tensor)[source]
class DeconvolutionalLayer(n_output_chns, kernel_size=3, stride=1, padding='SAME', with_bias=False, feature_normalization='batch', group_size=-1, acti_func=None, w_initializer=None, w_regularizer=None, b_initializer=None, b_regularizer=None, moving_decay=0.9, eps=1e-05, name='deconv')[source]

Bases: niftynet.layer.base_layer.TrainableLayer

This class defines a composite layer with optional components:

deconvolution -> batch_norm -> activation -> dropout

The b_initializer and b_regularizer are applied to the DeconvLayer The w_initializer and w_regularizer are applied to the DeconvLayer, the batch normalisation layer, and the activation layer (for ‘prelu’)

layer_op(input_tensor, is_training=None, keep_prob=None)[source]