niftynet.layer.upsample_res_block module

class UpBlock(n_output_chns=4, kernel_size=3, upsample_stride=2, acti_func='relu', w_initializer=None, w_regularizer=None, is_residual_upsampling=True, type_string='bn_acti_conv', name='res-upsample')[source]

Bases: niftynet.layer.base_layer.TrainableLayer

layer_op(inputs, forwarding=None, is_training=True)[source]

Consists of:

(inputs)--upsampling-+-o--conv_1--conv_2--+--(conv_res)--
                     | |                  |
(forwarding)---------o o------------------o

where upsampling method could be DeconvolutionalLayer or ResidualUpsampleLayer