Handle filter_dilation in the abstractconv_grad* perform methods

theano/tensor/nnet/abstract_conv.py

@@ -972,6 +972,9 @@ class AbstractConv2d_gradWeights(BaseAbstractConv2d):
             mode = (shape[0] // 2, shape[1] // 2)
         if isinstance(mode, tuple):
             pad_h, pad_w = map(int, mode)
+            pad_h = (pad_h - 1) * self.filter_dilation[0] + 1
+            pad_w = (pad_w - 1) * self.filter_dilation[1] + 1
+
             mode = "valid"
             new_img = numpy.zeros((img.shape[0], img.shape[1],
                                    img.shape[2] + 2 * pad_h,
@@ -980,9 +983,11 @@ class AbstractConv2d_gradWeights(BaseAbstractConv2d):
             img = new_img
 
         if self.subsample[0] > 1 or self.subsample[1] > 1:
+            dil_shape = ((shape[0] - 1) * self.filter_dilation[0] + 1,
+                         (shape[1] - 1) * self.filter_dilation[1] + 1)
             new_shape = (topgrad.shape[0], topgrad.shape[1],
-                         img.shape[2] - shape[0] + 1,
-                         img.shape[3] - shape[1] + 1)
+                         img.shape[2] - dil_shape[0] + 1,
+                         img.shape[3] - dil_shape[1] + 1)
             new_topgrad = numpy.zeros((new_shape), dtype=topgrad.dtype)
             new_topgrad[:, :, ::self.subsample[0], ::self.subsample[1]] = topgrad
             topgrad = new_topgrad
@@ -990,6 +995,8 @@ class AbstractConv2d_gradWeights(BaseAbstractConv2d):
         topgrad = topgrad.transpose(1, 0, 2, 3)[:, :, ::-1, ::-1]
         img = img.transpose(1, 0, 2, 3)
         kern = self.conv2d(img, topgrad, mode="valid")
+        if self.filter_dilation[0] > 1 or self.filter_dilation[1] > 1:
+            kern = kern[:, :, ::self.filter_dilation[0], ::self.filter_dilation[1]]
         if self.filter_flip:
             kern = kern.transpose(1, 0, 2, 3)[:, :, ::-1, ::-1]
         else:
@@ -1103,24 +1110,26 @@ class AbstractConv2d_gradInputs(BaseAbstractConv2d):
                 '"valid", "full", "half", an integer or a pair of'
                 ' integers'.format(mode))
 
+        dil_kernshp = ((kern.shape[2] - 1) * self.filter_dilation[0] + 1,
+                       (kern.shape[3] - 1) * self.filter_dilation[1] + 1)
         pad_h, pad_w = 0, 0
         if mode == "full":
-            pad_h, pad_w = (kern.shape[2] - 1, kern.shape[3] - 1)
+            pad_h, pad_w = (dil_kernshp[0] - 1, dil_kernshp[0] - 1)
         elif mode == "half":
-            pad_h, pad_w = (kern.shape[2] // 2, kern.shape[3] // 2)
+            pad_h, pad_w = (dil_kernshp[1] // 2, dil_kernshp[1] // 2)
         elif isinstance(mode, tuple):
             pad_h, pad_w = map(int, self.border_mode)
         if self.subsample[0] > 1 or self.subsample[1] > 1:
             new_shape = (topgrad.shape[0], topgrad.shape[1],
-                         shape[0] + 2 * pad_h - kern.shape[2] + 1,
-                         shape[1] + 2 * pad_w - kern.shape[3] + 1)
+                         shape[0] + 2 * pad_h - dil_kernshp[0] + 1,
+                         shape[1] + 2 * pad_w - dil_kernshp[1] + 1)
             new_topgrad = numpy.zeros((new_shape), dtype=topgrad.dtype)
             new_topgrad[:, :, ::self.subsample[0], ::self.subsample[1]] = topgrad
             topgrad = new_topgrad
         kern = kern.transpose(1, 0, 2, 3)
         if self.filter_flip:
             topgrad = topgrad[:, :, ::-1, ::-1]
-        img = self.conv2d(topgrad, kern, mode="full")
+        img = self.conv2d(topgrad, kern, mode="full", dilation=self.filter_dilation)
         if self.filter_flip:
             img = img[:, :, ::-1, ::-1]
         if pad_h > 0 or pad_w > 0: