- Conv2D and Conv3D now take dilation as parameter

inferno/extensions/layers/convolutional.py

@@ -238,12 +238,14 @@ def __init__(self, in_channels, out_channels, kernel_size, dilation=2):
 class Conv2D(ConvActivation):
     """
     2D convolutional layer with same padding and orthogonal weight initialization.
-    This layer does not apply an activation function.
+    By default, this layer does not apply an activation function.
     """
-    def __init__(self, in_channels, out_channels, kernel_size, activation=None):
+    def __init__(self, in_channels, out_channels, kernel_size, dilation=1,
+                 activation=None):
         super(Conv2D, self).__init__(in_channels=in_channels,
                                      out_channels=out_channels,
                                      kernel_size=kernel_size,
+                                     dilation=dilation,
                                      dim=2,
                                      activation=activation,
                                      initialization=OrthogonalWeightsZeroBias())
@@ -252,12 +254,14 @@ def __init__(self, in_channels, out_channels, kernel_size, activation=None):
 class Conv3D(ConvActivation):
     """
     3D convolutional layer with same padding and orthogonal weight initialization.
-    This layer does not apply an activation function.
+    By default, this layer does not apply an activation function.
     """
-    def __init__(self, in_channels, out_channels, kernel_size, activation=None):
+    def __init__(self, in_channels, out_channels, kernel_size, dilation=1,
+                 activation=None):
         super(Conv3D, self).__init__(in_channels=in_channels,
                                      out_channels=out_channels,
                                      kernel_size=kernel_size,
+                                     dilation=dilation,
                                      dim=3,
                                      activation=activation,
                                      initialization=OrthogonalWeightsZeroBias())