Merge pull request #2590 from lamblin/fix_pool_infer_shape

theano/sandbox/cuda/dnn.py

@@ -835,11 +835,12 @@ class GpuDnnPool(DnnBase):
         desc = node.inputs[1].owner.op
         kh, kw = desc.ws
         sh, sw = desc.stride
+        padh, padw = desc.pad
         return [(
             shape[0][0],
             shape[0][1],
-            (shape[0][2] - kh)//sh + 1,
-            (shape[0][3] - kw)//sw + 1
+            (shape[0][2] + 2*padh - kh)//sh + 1,
+            (shape[0][3] + 2*padw - kw)//sw + 1
         )]
 
     def c_support_code_struct(self, node, name):

theano/sandbox/cuda/tests/test_dnn.py

@@ -73,11 +73,17 @@ def test_pooling():
         if pad != (0, 0) and cuda.dnn.version() < 20:
             continue
 
+        if pad != (0, 0) and func is T.mean:
+            continue
+
         for ws in (4, 2, 5):
             for stride in (2, 3):
                 if stride > ws:
                     continue
                 if func is T.max:
+                    if pad[0] > stride or pad[1] > stride:
+                        # Not implemented
+                        continue
                     # We will check that the opt introduced it.
                     out1 = max_pool_2d(x, (ws, ws),
                                        st=(stride, stride),
@@ -117,6 +123,9 @@ def test_pooling():
 
             ws = 2
             stride = 2
+            if pad[0] > stride or pad[1] > stride:
+                # Not implemented
+                continue
 
             # This test the CPU grad + opt + GPU implemtentation
             def fn(x):

theano/tensor/signal/downsample.py

@@ -257,7 +257,7 @@ class DownsampleFactorMax(Op):
 
     def infer_shape(self, node, in_shapes):
         shp = self.out_shape(in_shapes[0], self.ds,
-                             self.ignore_border, self.st)
+                             self.ignore_border, self.st, self.padding)
         return [shp]
 
     def grad(self, inp, grads):