[ONNX] Add ArgMin/Max operators (#1898)

* Add ArgMin operator

* Add ArgMax and a basic test case

* Rename variables

* Apply workaround for problems with Reshape on i64

* Review comments

* Review comments

src/ngraph/frontend/onnx_import/CMakeLists.txt

@@ -38,6 +38,10 @@ add_library(onnx_import STATIC
         op/acos.hpp
         op/add.hpp
         op/and.hpp
+        op/argmax.cpp
+        op/argmax.hpp
+        op/argmin.cpp
+        op/argmin.hpp
         op/asin.hpp
         op/atan.hpp
         op/average_pool.cpp

src/ngraph/frontend/onnx_import/op/argmax.cpp

+//*****************************************************************************
+// Copyright 2017-2018 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+
+#include "ngraph/op/argmax.hpp"
+#include "core/node.hpp"
+#include "ngraph/node_vector.hpp"
+#include "utils/reduction.hpp"
+
+namespace ngraph
+{
+    namespace onnx_import
+    {
+        namespace op
+        {
+            namespace set_1
+            {
+                NodeVector argmax(const Node& node)
+                {
+                    return {reduction::make_ng_index_reduction_op<ngraph::op::ArgMax>(node)};
+                }
+
+            } // namespace set_1
+
+        } //namespace op
+
+    } // namespace onnx_import
+
+} // namespace ngraph

src/ngraph/frontend/onnx_import/op/argmax.hpp

+//*****************************************************************************
+// Copyright 2017-2018 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+
+#pragma once
+
+#include "core/node.hpp"
+#include "ngraph/node_vector.hpp"
+
+namespace ngraph
+{
+    namespace onnx_import
+    {
+        namespace op
+        {
+            namespace set_1
+            {
+                /// \brief Convert ONNX ArgMax operation to an nGraph node.
+                ///
+                /// \param node   The ONNX node object representing this operation.
+                ///
+                /// \return The vector containing an Ngraph node which produces the output
+                ///         of an ONNX ArgMax operation.
+                NodeVector argmax(const Node& node);
+
+            } // namespace set_1
+
+        } //namespace op
+
+    } // namespace onnx_import
+
+} // namespace ngraph

src/ngraph/frontend/onnx_import/op/argmin.cpp

+//*****************************************************************************
+// Copyright 2017-2018 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+
+#include "ngraph/op/argmin.hpp"
+#include "core/node.hpp"
+#include "ngraph/node_vector.hpp"
+#include "utils/reduction.hpp"
+
+namespace ngraph
+{
+    namespace onnx_import
+    {
+        namespace op
+        {
+            namespace set_1
+            {
+                NodeVector argmin(const Node& node)
+                {
+                    return {reduction::make_ng_index_reduction_op<ngraph::op::ArgMin>(node)};
+                }
+
+            } // namespace set_1
+
+        } //namespace op
+
+    } // namespace onnx_import
+
+} // namespace ngraph

src/ngraph/frontend/onnx_import/op/argmin.hpp

+//*****************************************************************************
+// Copyright 2017-2018 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+
+#pragma once
+
+#include "core/node.hpp"
+#include "ngraph/node_vector.hpp"
+
+namespace ngraph
+{
+    namespace onnx_import
+    {
+        namespace op
+        {
+            namespace set_1
+            {
+                /// \brief Convert ONNX ArgMin operation to an nGraph node.
+                ///
+                /// \param node   The ONNX node object representing this operation.
+                ///
+                /// \return The vector containing an Ngraph node which produces the output
+                ///         of an ONNX ArgMin operation.
+                NodeVector argmin(const Node& node);
+
+            } // namespace set_1
+
+        } //namespace op
+
+    } // namespace onnx_import
+
+} // namespace ngraph

src/ngraph/frontend/onnx_import/ops_bridge.cpp

@@ -25,6 +25,8 @@
 #include "op/acos.hpp"
 #include "op/add.hpp"
 #include "op/and.hpp"
+#include "op/argmax.hpp"
+#include "op/argmin.hpp"
 #include "op/asin.hpp"
 #include "op/atan.hpp"
 #include "op/average_pool.hpp"
@@ -145,6 +147,8 @@ namespace ngraph
             REGISTER_OPERATOR("Acos", 1, acos);
             REGISTER_OPERATOR("Add", 1, add);
             REGISTER_OPERATOR("And", 1, logical_and);
+            REGISTER_OPERATOR("ArgMin", 1, argmin);
+            REGISTER_OPERATOR("ArgMax", 1, argmax);
             REGISTER_OPERATOR("Asin", 1, asin);
             REGISTER_OPERATOR("Atan", 1, atan);
             REGISTER_OPERATOR("AveragePool", 1, average_pool);

src/ngraph/frontend/onnx_import/utils/reduction.hpp

@@ -27,6 +27,7 @@
 
 #include "ngraph/axis_set.hpp"
 #include "ngraph/node_vector.hpp"
+#include "ngraph/op/convert.hpp"
 #include "ngraph/op/reshape.hpp"
 #include "ngraph/op/util/arithmetic_reduction.hpp"
 #include "ngraph/shape.hpp"
@@ -101,6 +102,38 @@ namespace ngraph
                     reshape::get_default_axis_vector(op_node->get_shape().size()),
                     Shape{output_shape});
             }
+
+            template <class IndexReduction>
+            std::shared_ptr<ngraph::Node> make_ng_index_reduction_op(const Node& node)
+            {
+                auto axis = node.get_attribute_value<int64_t>("axis", 0);
+                auto keepdims = node.get_attribute_value<int64_t>("keepdims", 1);
+                auto input_node = node.get_ng_inputs().at(0);
+
+                auto op_node = std::make_shared<IndexReduction>(input_node, axis, element::i64);
+
+                if (keepdims == 0)
+                {
+                    return op_node;
+                }
+
+                // WORKAROUND FOR PROBLEMS WITH RESHAPE ON i64 @TODO: remove
+                auto convert_node = std::make_shared<ngraph::op::Convert>(op_node, element::f32);
+
+                auto output_shape = input_node->get_shape();
+                output_shape.at(axis) = 1;
+                auto reshape_node = std::make_shared<ngraph::op::Reshape>(
+                    convert_node,
+                    reshape::get_default_axis_vector(op_node->get_shape().size()),
+                    Shape{output_shape});
+
+                // WORKAROUND FOR PROBLEMS WITH RESHAPE ON i64 @TODO: remove
+                auto reconvert_node =
+                    std::make_shared<ngraph::op::Convert>(reshape_node, element::i64);
+
+                return reconvert_node;
+            }
+
         } // namespace  reduction
     }     // namespace onnx_import
 } // namespace ngraph

test/onnx_import.cpp

@@ -17,6 +17,7 @@
 #include <cstdint>
 #include <fstream>
 #include <sstream>
+#include <vector>
 
 #include "gtest/gtest.h"
 #include "ngraph/frontend/onnx_import/onnx.hpp"
@@ -74,6 +75,18 @@ TEST(onnx, model_addmul_abc)
     EXPECT_TRUE(test::all_close_f(expected_output, result_vectors.front()));
 }
 
+TEST(onnx, model_argmin_no_keepdims)
+{
+    auto function = onnx_import::import_onnx_function(
+        file_util::path_join(SERIALIZED_ZOO, "onnx/argmin_no_keepdims.onnx"));
+
+    Inputs inputs{test::NDArray<float, 2>{{2, 1}, {3, 10}}.get_vector()};
+    std::vector<std::vector<int64_t>> expected_output{{1, 0}};
+    std::vector<std::vector<int64_t>> result{
+        execute<float, int64_t>(function, inputs, "INTERPRETER")};
+    EXPECT_EQ(expected_output, result);
+}
+
 TEST(onnx, model_split_equal_parts_default)
 {
     Model model{onnx_import::load_onnx_model(