[v1.x] Onnx Reshpe support for special caes

python/mxnet/contrib/onnx/mx2onnx/_op_translations.py

@@ -1598,6 +1598,7 @@ def convert_floor(node, **kwargs):
     """
     return create_basic_op_node('Floor', node, kwargs)
 
+
 # Changing shape and type.
 @mx_op.register("Reshape")
 def convert_reshape(node, **kwargs):
@@ -1612,6 +1613,17 @@ def convert_reshape(node, **kwargs):
     reverse = attrs.get('reverse', 'False')
     targ_shape = convert_string_to_list(attrs["shape"])
 
+    # In general -2, -3, -4 in the target shape are not supoorted, but there are
+    # a few special cases that we can convert to supported scenarios
+
+    # If -2 and -4 are not used, then we can just remove the -4
+    if -4 in targ_shape and -3 not in targ_shape and -2 not in targ_shape:
+        targ_shape = [i for i in targ_shape if i != -4]
+
+    if targ_shape == [-3, 0]:
+        targ_shape = [-1, 0]
+        reverse = 'True'
+
     not_supported_shape = [-2, -3, -4]
     for val in targ_shape:
         if val in not_supported_shape:

tests/python-pytest/onnx/test_operators.py

@@ -216,6 +216,23 @@ def test_onnx_export_reshape(tmp_path, dtype):
     op_export_test('reshape_3', M3, [x], tmp_path)
 
 
+@pytest.mark.parametrize('dtype', ['float32', 'float64', 'int32', 'int64'])
+def test_onnx_export_reshape_special_cases(tmp_path, dtype):
+    x1 = mx.nd.ones((8, 9), dtype=dtype)
+    M1 = def_model('reshape', shape=(0, -4, 1, -1))
+    op_export_test('reshape_spec_1', M1, [x1], tmp_path)
+
+    x2 = mx.nd.ones((8, 9, 10), dtype=dtype)
+
+    M2 = def_model('reshape', shape=(0, -4, 3, -1, 10))
+    op_export_test('reshape_spec_2', M2, [x2], tmp_path)
+    M3 = def_model('reshape', shape=(-4, 2, -1, 10, 9))
+    op_export_test('reshape_spec_3', M3, [x2], tmp_path)
+
+    M4 = def_model('reshape', shape=(-3, 0))
+    op_export_test('reshape_spec_4', M4, [x2], tmp_path)
+
+
 @pytest.mark.parametrize('dtype', ['int32', 'int64'])
 def test_onnx_export_embedding(tmp_path, dtype):
     x = mx.nd.array([[ 1.,  3.],