IntenseWebs/openvino
3
0 Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix the issue in values_from_const method in OVTF integration with TF FE (#10225)

Browse Source 
* Fix the issue in values_from_const method in OVTF integration with TF FE

* fix comment
This commit is contained in:
Ivan Tikhonov authored and GitHub committed 2022-02-10 11:33:42 +03:00
1 parent 334e9e994e
commit 3f0e532dce
1 file changed
+8 -1
Download Patch File Download Diff File Expand all files Collapse all files
src/frontends/tensorflow/src/utils.hpp
+8 -1
View File
@@ -93,7 +93,14 @@ void values_from_const_node(const NodeContext& node, ov::Shape* const_tensor_sha
TENSORFLOW_OP_VALIDATION(node, node.get_op_type() == "Const", "Node is expected to be Constant.");
const auto* decoder = node.get_decoder();
auto dt = decoder->get_native_attribute("dtype").as<::tensorflow::DataType>();
auto tensor_proto = decoder->get_native_attribute("value").as<::tensorflow::TensorProto>();
// TODO: investigate why as<>() && method using std::move leads to the issue (75371) in OVTF integration with
// tensorflow frontend. The current fix: replace it with as<>() & method. But in fact, both
// approaches should work the same way.
// auto tensor_proto = decoder->get_native_attribute("value").as<::tensorflow::TensorProto>();
auto value = decoder->get_native_attribute("value");
auto tensor_proto = value.as<::tensorflow::TensorProto>();
const ::tensorflow::TensorShapeProto& shape = tensor_proto.tensor_shape();
ov::PartialShape pshape;
tf_shape_to_ov_shape(shape, &pshape);