Fix issue onnx#2102: Set reduction axis of mean to height and width for adjust_contrast op

Signed-off-by: cosine <[email protected]>

Author: cosineFish

tests/test_backend.py

@@ -3280,13 +3280,13 @@ def func(x, x_new_size_):
         self._run_test_case(func, [_OUTPUT], {_INPUT: x_val, _INPUT1: x_new_size})
 
     def test_adjust_contrast(self):
-        x_shape = [4, 3, 2]
-        x_val = np.arange(1, 1 + np.prod(x_shape), dtype=np.float32).reshape(x_shape)
-        y_val = np.array(2.1, np.float32)
         def func(x, y):
             x_ = tf.image.adjust_contrast(x, y)
             return tf.identity(x_, name=_TFOUTPUT)
-        self._run_test_case(func, [_OUTPUT], {_INPUT: x_val, _INPUT1: y_val})
+        for x_shape in [[4, 3, 2], [2, 3, 4, 5], [3, 4, 2, 4, 3]]:
+            x_val = np.arange(1, 1 + np.prod(x_shape), dtype=np.float32).reshape(x_shape)
+            y_val = np.array(2.1, np.float32)
+            self._run_test_case(func, [_OUTPUT], {_INPUT: x_val, _INPUT1: y_val})
 
     @check_opset_min_version(11, "GatherElements")
     def test_adjust_saturation(self):

tf2onnx/onnx_opset/nn.py

@@ -1505,8 +1505,8 @@ def version_1(cls, ctx, node, **kwargs):
             contrast_factor = ctx.make_node("Cast", [dtype], attr={'to': dtype}).output[0]
         rank = ctx.get_rank(images)
         utils.make_sure(rank is not None, "AdjustContrastv2 requires input of known rank")
-        # Reduce everything except channels
-        axes_to_reduce = list(range(rank))[:-1]
+        # Reduce height and width only
+        axes_to_reduce = list(range(rank))[-3:-1]
         mean = ctx.make_node("ReduceMean", [images], attr={'axes': axes_to_reduce, 'keepdims': True},
                              op_name_scope=node.name).output[0]
         diff = ctx.make_node("Sub", [images, mean], op_name_scope=node.name).output[0]