From c8ee99ace5431ca8300d1aa7e3380c7e54a7d43a Mon Sep 17 00:00:00 2001
From: Joe Evans <joeev@amazon.com>
Date: Fri, 5 Feb 2021 17:48:59 +0000
Subject: [PATCH] Replace deprecated ConstantFill onnx operator with Identity,
 add commented out models that work with onnx export but fail with
 onnxruntime.

---
 .../contrib/onnx/mx2onnx/_op_translations.py  |  8 ++++----
 tests/python-pytest/onnx/test_onnxruntime.py  | 19 ++++++++++++++++++-
 2 files changed, 22 insertions(+), 5 deletions(-)

diff --git a/python/mxnet/contrib/onnx/mx2onnx/_op_translations.py b/python/mxnet/contrib/onnx/mx2onnx/_op_translations.py
index 3576242e0d77..a9a63950e5de 100644
--- a/python/mxnet/contrib/onnx/mx2onnx/_op_translations.py
+++ b/python/mxnet/contrib/onnx/mx2onnx/_op_translations.py
@@ -783,10 +783,10 @@ def convert_copy(node, **kwargs):
 
 @mx_op.register("identity")
 def convert_identity(node, **kwargs):
-    """Map MXNet's identity operator attributes to onnx's ConstantFill operator
+    """Map MXNet's identity operator attributes to onnx's Identity operator
     and return the created node.
     """
-    return create_basic_op_node('ConstantFill', node, kwargs)
+    return create_basic_op_node('Identity', node, kwargs)
 
 @mx_op.register("InstanceNorm")
 def convert_instancenorm(node, **kwargs):
@@ -1013,12 +1013,12 @@ def convert_logistic_regression_output(node, **kwargs):
 @mx_op.register("BlockGrad")
 def convert_blockgrad(node, **kwargs):
     """ Skip operator  """
-    return create_basic_op_node('ConstantFill', node, kwargs)
+    return create_basic_op_node('Identity', node, kwargs)
 
 @mx_op.register("MakeLoss")
 def convert_makeloss(node, **kwargs):
     """ Skip operator  """
-    return create_basic_op_node('ConstantFill', node, kwargs)
+    return create_basic_op_node('Identity', node, kwargs)
 
 @mx_op.register("Concat")
 def convert_concat(node, **kwargs):
diff --git a/tests/python-pytest/onnx/test_onnxruntime.py b/tests/python-pytest/onnx/test_onnxruntime.py
index 151c5bebd9b2..319d65348bb8 100644
--- a/tests/python-pytest/onnx/test_onnxruntime.py
+++ b/tests/python-pytest/onnx/test_onnxruntime.py
@@ -246,7 +246,24 @@ def normalize_image(imgfile):
     'center_net_resnet101_v1b_voc',
     'center_net_resnet18_v1b_coco',
     'center_net_resnet50_v1b_coco',
-    'center_net_resnet101_v1b_coco'
+    'center_net_resnet101_v1b_coco',
+    # the following models are failing due to onnxruntime errors
+    #'ssd_300_vgg16_atrous_voc',
+    #'ssd_512_vgg16_atrous_voc',
+    #'ssd_512_resnet50_v1_voc',
+    #'ssd_512_mobilenet1.0_voc',
+    #'faster_rcnn_resnet50_v1b_voc',
+    #'yolo3_darknet53_voc',
+    #'yolo3_mobilenet1.0_voc',
+    #'ssd_300_vgg16_atrous_coco',
+    #'ssd_512_vgg16_atrous_coco',
+    #'ssd_300_resnet34_v1b_coco',
+    #'ssd_512_resnet50_v1_coco',
+    #'ssd_512_mobilenet1.0_coco',
+    #'faster_rcnn_resnet50_v1b_coco',
+    #'faster_rcnn_resnet101_v1d_coco',
+    #'yolo3_darknet53_coco',
+    #'yolo3_mobilenet1.0_coco',
 ])
 def test_obj_detection_model_inference_onnxruntime(tmp_path, model):
     def normalize_image(imgfile):