[Relay][Frontend][Onnx] Enable group_conv1d import through conv2d conversion.

python/tvm/relay/frontend/onnx.py

@@ -451,6 +451,7 @@ class Conv(OnnxOpConverter):
     def _impl_v1(cls, inputs, attr, params):
         # Use shape of input to determine convolution type.
         data = inputs[0]
+        kernel = inputs[1]
         input_shape = infer_shape(data)
         ndim = len(input_shape)
 
@@ -473,13 +474,32 @@ def _impl_v1(cls, inputs, attr, params):
                     mode=attr["auto_pad"],
                 )
             elif attr["auto_pad"] == "VALID":
-                attr["pads"] = tuple([0 for i in range(ndim - 2)])
+                attr["pads"] = [0 for i in range(ndim - 2)]
             elif attr["auto_pad"] == "NOTSET":
                 pass
             else:
                 msg = 'Value {} in attribute "auto_pad" of operator Conv is invalid.'
                 raise tvm.error.OpAttributeInvalid(msg.format(attr["auto_pad"]))
             attr.pop("auto_pad")
+
+        # Check if the requested convolution is a group conv1d, if so convert it to conv2d.
+        # TODO(jwfromm) Remove once proper group_conv1d is supported.
+        group_conv1d = False
+        if dimension_picker("conv")(attr) == "conv1d" and attr.get("group") != 1:
+            group_conv1d = True
+            # Expand input from NCW to NCHW
+            data = _op.expand_dims(data, axis=2)
+            # Expand kernel from OIW to OIHW
+            kernel = _op.expand_dims(kernel, axis=2)
+            # Add new value to kernel_shape, strices, dilation, pads, if needed
+            attr["kernel_shape"] = [1] + list(attr["kernel_shape"])
+            if "strides" in attr:
+                attr["strides"] = [1] + list(attr["strides"])
+            if "dilations" in attr:
+                attr["dilations"] = [1] + list(attr["dilations"])
+            if "pads" in attr:
+                attr["pads"] = [0, attr["pads"][0], 0, attr["pads"][1]]
+
         out = AttrCvt(
             op_name=dimension_picker("conv"),
             transforms={
@@ -489,7 +509,11 @@ def _impl_v1(cls, inputs, attr, params):
                 "group": ("groups", 1),
             },
             custom_check=dimension_constraint(),
-        )([data, inputs[1]], attr, params)
+        )([data, kernel], attr, params)
+
+        # If this was a group_conv1d, squish output back to NCW.
+        if group_conv1d:
+            out = _op.squeeze(out, axis=[2])
 
         use_bias = len(inputs) == 3
         if use_bias:

tests/python/frontend/onnx/test_forward.py

@@ -2410,6 +2410,7 @@ def verify_conv(
     kernel_shape,
     strides,
     dilations,
+    group=1,
     auto_pad="NOTSET",
     unset_pad=False,
 ):
@@ -2422,7 +2423,7 @@ def verify_conv(
             # Default values for other attributes:
             strides=strides,
             dilations=dilations,
-            # groups=1
+            group=group,
         )
     elif padding is None:
         ## autopadding with unset default attributes
@@ -2438,6 +2439,7 @@ def verify_conv(
             outputs=["y"],
             # Default values for other attributes:
             auto_pad=auto_pad,
+            group=group,
             **kwargs,
         )
     else:
@@ -2449,7 +2451,7 @@ def verify_conv(
             # Default values for other attributes:
             strides=strides,
             dilations=dilations,
-            # groups=1
+            group=group,
             pads=padding,
         )
 
@@ -2559,6 +2561,20 @@ def repeat(N, D):
             repeat(2, D),
         )
 
+    # TODO(jwfromm): Merge with other tests once group_conv3d is supported.
+    for D in [1, 2]:
+        # Group Convolution
+        verify_conv(
+            (1, 8) + repeat(5, D),
+            (8, 1) + repeat(3, D),
+            (1, 8) + repeat(5, D),
+            2 * repeat(1, D),
+            repeat(3, D),
+            repeat(1, D),
+            repeat(1, D),
+            group=8,
+        )
+
 
 def verify_convtranspose_with_padding(
     x_shape,