Use ONNX Frontend instead of ONNX Reader

ngraph/frontend/onnx/frontend/include/onnx_frontend/frontend.hpp

@@ -22,6 +22,9 @@ namespace ngraph
             std::shared_ptr<ngraph::Function> convert(InputModel::Ptr model) const override;
             void convert(std::shared_ptr<ngraph::Function> partially_converted) const override;
             std::shared_ptr<ngraph::Function> decode(InputModel::Ptr model) const override;
+            std::string get_name() const override;
+            bool supported_impl(
+                const std::vector<std::shared_ptr<Variant>>& variants) const override;
 
         protected:
             InputModel::Ptr

ngraph/frontend/onnx/frontend/src/frontend.cpp

@@ -5,6 +5,7 @@
 #include <frontend_manager/frontend_exceptions.hpp>
 #include <frontend_manager/frontend_manager.hpp>
 #include <input_model.hpp>
+#include <ngraph/file_util.hpp>
 #include <onnx_frontend/frontend.hpp>
 #include <onnx_import/onnx.hpp>
 #include <utils/onnx_internal.hpp>
@@ -54,3 +55,21 @@ std::shared_ptr<ngraph::Function> FrontEndONNX::decode(InputModel::Ptr model) co
     NGRAPH_CHECK(model_onnx != nullptr, "Invalid input model");
     return model_onnx->decode();
 }
+
+std::string FrontEndONNX::get_name() const
+{
+    return "onnx_experimental";
+}
+
+bool FrontEndONNX::supported_impl(const std::vector<std::shared_ptr<Variant>>& variants) const
+{
+    if (variants.size() > 0 && is_type<VariantWrapper<std::string>>(variants[0]))
+    {
+        const auto path = as_type_ptr<VariantWrapper<std::string>>(variants[0])->get();
+        if (file_util::get_file_ext(path) == ".onnx")
+        {
+            return true;
+        }
+    }
+    return false;
+}

ngraph/python/tests/test_frontend/test_frontend_onnx.py

@@ -96,3 +96,18 @@ def test_decode_and_convert():
     b = np.array([[2, 3], [4, 5]], dtype=np.float32)
     expected = np.array([[1.5, 5], [10.5, 18]], dtype=np.float32)
     run_function(decoded_function, a, b, expected=[expected])
+
+
+def test_load_by_model():
+    skip_if_onnx_frontend_is_disabled()
+
+    fe = fem.load_by_model(onnx_model_filename)
+    assert fe
+    assert fe.get_name() == "onnx_experimental"
+    model = fe.load(onnx_model_filename)
+    assert model
+    decoded_function = fe.decode(model)
+    assert decoded_function
+
+    assert not fem.load_by_model("test.xx")
+    assert not fem.load_by_model("onnx.yy")