spatial size calc for activations

sparseconvnet/activations.py

@@ -20,7 +20,8 @@ class Sigmoid(Module):
         output.metadata = input.metadata
         output.spatial_size = input.spatial_size
         return output
-
+    def input_spatial_size(self, out_size):
+        return out_size
 
 class LeakyReLU(Module):
     def __init__(self,leak=1/3):
@@ -32,6 +33,8 @@ class LeakyReLU(Module):
         output.metadata = input.metadata
         output.spatial_size = input.spatial_size
         return output
+    def input_spatial_size(self, out_size):
+        return out_size
 
 
 class Tanh(Module):
@@ -41,6 +44,8 @@ class Tanh(Module):
         output.metadata = input.metadata
         output.spatial_size = input.spatial_size
         return output
+    def input_spatial_size(self, out_size):
+        return out_size
 
 
 class ReLU(Module):
@@ -50,6 +55,8 @@ class ReLU(Module):
         output.metadata = input.metadata
         output.spatial_size = input.spatial_size
         return output
+    def input_spatial_size(self, out_size):
+        return out_size
 
 
 class ELU(Module):
@@ -59,6 +66,8 @@ class ELU(Module):
         output.metadata = input.metadata
         output.spatial_size = input.spatial_size
         return output
+    def input_spatial_size(self, out_size):
+        return out_size
 
 class SELU(Module):
     def forward(self, input):
@@ -67,6 +76,8 @@ class SELU(Module):
         output.metadata = input.metadata
         output.spatial_size = input.spatial_size
         return output
+    def input_spatial_size(self, out_size):
+        return out_size
 
 def BatchNormELU(nPlanes, eps=1e-4, momentum=0.9):
     return sparseconvnet.Sequential().add(BatchNormalization(nPlanes,eps,momentum)).add(ELU())