[Relay][Frontend][Pytorch] Fixed ConvTranspose2D parsing (#5157)

* Fixed conv transpose parsing.

* small format change.

* Chage test module names.

* Simplified test syntax.

python/tvm/relay/frontend/pytorch.py

@@ -251,7 +251,7 @@ def _impl(inputs, input_types):
 def _convolution():
     def _impl(inputs, input_types):
         # Use transpose or normal
-        use_transpose = True if inputs[6] == "1" else False
+        use_transpose = True if inputs[6] == 1 else False
 
         data = inputs[0]
         weight = inputs[1]
@@ -268,6 +268,10 @@ def _impl(inputs, input_types):
         else:
             assert "data type {} could not be parsed in conv op" % (type(weight))
 
+        # Transposed convolutions have IOHW layout.
+        if use_transpose:
+            weight_shape[0], weight_shape[1] = weight_shape[1], weight_shape[0]
+
         channels = weight_shape[0]
         groups = int(inputs[8])
 

tests/python/frontend/pytorch/test_forward.py

@@ -448,6 +448,14 @@ def forward(self, *args):
                  input_data=torch.randn((1, 8, 16, 16)))
 
 
+def test_forward_conv_transpose():
+    torch.set_grad_enabled(False)
+    input_shape = [1, 3, 10, 10]
+    input_data = torch.rand(input_shape).float()
+    verify_model(torch.nn.ConvTranspose2d(3, 6, 7, bias=True), input_data=input_data)
+    verify_model(torch.nn.ConvTranspose2d(3, 12, 3, bias=False), input_data=input_data)
+
+
 def test_forward_threshold():
     torch.set_grad_enabled(False)
     input_shape = [1, 3]
@@ -1050,6 +1058,7 @@ def forward(self, xs):
     test_forward_maxpool1d()
     test_forward_hardtanh()
     test_forward_conv()
+    test_forward_conv_transpose()
     test_forward_threshold()
     test_forward_contiguous()
     test_forward_batchnorm()