Merge branch 'master' of https://github.com/ANTsX/ANTsRNet

R/createCustomUnetModel.R

@@ -312,6 +312,8 @@ createHippMapp3rUnetModel3D <- function( inputImageSize,
 #'
 #' @param numberOfModalities Specifies number of channels in the 
 #' architecture.
+#' @param numberOfOutputs Specifies the number of outputs per voxel.  
+#' Determines final activation function (1 = sigmoid, >1 = softmax).
 #' @return a u-net keras model
 #' @author Tustison NJ
 #' @examples
@@ -322,7 +324,8 @@ createHippMapp3rUnetModel3D <- function( inputImageSize,
 #' }
 #' @import keras
 #' @export
-createShivaUnetModel3D <- function( numberOfModalities = 1 )
+createShivaUnetModel3D <- function( numberOfModalities = 1,
+                                    numberOfOutputs = 1 )
 {
   K <- tensorflow::tf$keras$backend
 
@@ -424,7 +427,12 @@ createShivaUnetModel3D <- function( numberOfModalities = 1 )
   outputs <- outputs %>% layer_batch_normalization()
   outputs <- outputs %>% layer_activation( "swish" )
 
-  outputs <- outputs %>% layer_conv_3d( 1, kernel_size = 1L, activation = "sigmoid", padding = 'same' )
+  activation = 'sigmoid'
+  if( numberOfOutputs > 1 )
+    {
+    activation = 'softmax'
+    } 
+  outputs <- outputs %>% layer_conv_3d( 1, kernel_size = numberOfOutputs, activation = activation, padding = 'same' )
 
   unetModel <- keras_model( inputs = inputs, outputs = outputs )