block_double: Block Double

In neuroimaginador/dl4ni: Deep Learning Flows for Neuroimaging

Description

This function creates a block of convolutional layers with upsampling such that the dimension of the outputs is doubled after each layer.

Usage

block_double(object, initial_filters = NULL, kernel_size = c(3, 3, 3),
  num_steps = NULL, batch_normalization = FALSE, dropout = 0,
  use_upsampling = TRUE, activation = "relu")

Arguments

object	(keras object) Object to which append this block
initial_filters	(numeric) Number of filters in the first convolutional layer, Default: 2 ^ num_steps
kernel_size	(list or vector) size of the convolution kernels, Default: c(3, 3, 3)
num_steps	(integer) Number of steps to perform downsampling, Default: 1 if kernel_size is a vector or its length if it's a list.
activation	(character) Activation function in the block layers, Default: 'relu'
params	(list) List of parameters to apply, if not listed in the previous ones.

Details

In each step, the number of filters is halved wrt the previous step. Thus, if num_steps == 3, the number of filters in the layers in this block is: 8, 4, 2.

Value

The composed object.

See Also

block_downsample block_unet

Other blocks: block_half

neuroimaginador/dl4ni documentation built on May 3, 2019, 5:47 p.m.