[ONNX] Add Tile op (#2990)

* [ONNX] Add Tile op

* Fix test model output shape

* Add static test for Tile

* Test Case: use expected shape, not computed output shape

* Exclude dynamic Tile op tests

* Add tests using BackendMode::DYNAMIC

* Code review comments
Co-Authored-By: Katarzyna Mitrus <katarzyna.mitrus@intel.com>

* Apply suggestions from code review
Co-Authored-By: Katarzyna Mitrus <katarzyna.mitrus@intel.com>

* Code review
Co-authored-by: Katarzyna Mitrus <katarzyna.mitrus@intel.com>
Co-authored-by: Scott Cyphers <diyessi@users.noreply.github.com>
Co-authored-by: Sang Ik Lee <sang.ik.lee@intel.com>

src/ngraph/frontend/onnx_import/CMakeLists.txt

@@ -212,6 +212,8 @@ add_library(onnx_import STATIC
         op/tanh.hpp
         op/thresholded_relu.cpp
         op/thresholded_relu.hpp
+        op/tile.cpp
+        op/tile.hpp
         op/topk.cpp
         op/topk.hpp
         op/transpose.cpp

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

+//*****************************************************************************
+// Copyright 2017-2020 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 <memory>
+
+#include "core/node.hpp"
+#include "default_opset.hpp"
+#include "tile.hpp"
+
+namespace ngraph
+{
+    namespace onnx_import
+    {
+        namespace op
+        {
+            namespace set_1
+            {
+                NodeVector tile(const Node& node)
+                {
+                    auto input = node.get_ng_inputs().at(0);
+                    auto repeats = node.get_ng_inputs().at(1);
+
+                    // Workaround for backends which require repeats to be i64.
+                    // Remove the following line when no longer needed.
+                    repeats = std::make_shared<default_opset::Convert>(repeats, element::i64);
+
+                    return {std::make_shared<default_opset::Tile>(input, repeats)};
+                }
+
+            } // namespace set_1
+
+        } // namespace op
+
+    } // namespace onnx_import
+
+} // namespace ngraph

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

+//*****************************************************************************
+// Copyright 2017-2020 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.hpp"
+
+namespace ngraph
+{
+    namespace onnx_import
+    {
+        namespace op
+        {
+            namespace set_1
+            {
+                /// \brief Performs ONNX Tile operation.
+                ///
+                /// \param node The ONNX node object representing this operation.
+                /// \return The vector containing nGraph a node producing the output of the Tile op.
+                NodeVector tile(const Node& node);
+
+            } // namespace set_1
+
+        } // namespace op
+
+    } // namespace onnx_import
+
+} // namespace ngraph

src/ngraph/frontend/onnx_import/ops_bridge.cpp

@@ -125,6 +125,7 @@
 #include "op/tan.hpp"
 #include "op/tanh.hpp"
 #include "op/thresholded_relu.hpp"
+#include "op/tile.hpp"
 #include "op/topk.hpp"
 #include "op/transpose.hpp"
 #include "op/unsqueeze.hpp"
@@ -362,6 +363,7 @@ namespace ngraph
             REGISTER_OPERATOR("Tan", 1, tan);
             REGISTER_OPERATOR("Tanh", 1, tanh);
             REGISTER_OPERATOR("ThresholdedRelu", 1, thresholded_relu);
+            REGISTER_OPERATOR("Tile", 1, tile);
             REGISTER_OPERATOR("TopK", 1, topk);
             REGISTER_OPERATOR("TopK", 10, topk);
             REGISTER_OPERATOR("TopK", 11, topk);

test/models/onnx/tile.prototxt

+ir_version: 3
+producer_name: "nGraph ONNX Importer"
+graph {
+  node {
+    input: "input"
+    input: "repeats"
+    output: "output"
+    op_type: "Tile"
+  }
+  name: "test_tile"
+  input {
+    name: "input"
+    type {
+      tensor_type {
+        elem_type: 5
+        shape {
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 3
+          }
+        }
+      }
+    }
+  }
+  input {
+    name: "repeats"
+    type {
+      tensor_type {
+        elem_type: 5
+        shape {
+          dim {
+            dim_value: 2
+          }
+        }
+      }
+    }
+  }
+  output {
+    name: "output"
+    type {
+      tensor_type {
+        elem_type: 5
+        shape {
+          dim {
+            dim_value: 4
+          }
+          dim {
+            dim_value: 3
+          }
+        }
+      }
+    }
+  }
+}
+opset_import {
+  version: 8
+}

test/models/onnx/tile_static.prototxt

+ir_version: 3
+producer_name: "nGraph ONNX Importer"
+graph {
+  node {
+    input: "input"
+    input: "repeats"
+    output: "output"
+    op_type: "Tile"
+  }
+  name: "test_tile"
+  initializer {
+    dims: 2
+    data_type: 7
+    raw_data: "\002\000\000\000\000\000\000\000\002\000\000\000\000\000\000\000"
+    name: "repeats"
+  }
+  input {
+    name: "input"
+    type {
+      tensor_type {
+        elem_type: 5
+        shape {
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 3
+          }
+        }
+      }
+    }
+  }
+  input {
+    name: "repeats"
+    type {
+      tensor_type {
+        elem_type: 7
+        shape {
+          dim {
+            dim_value: 2
+          }
+        }
+      }
+    }
+  }
+  output {
+    name: "output"
+    type {
+      tensor_type {
+        elem_type: 5
+        shape {
+          dim {
+            dim_value: 4
+          }
+          dim {
+            dim_value: 6
+          }
+        }
+      }
+    }
+  }
+}
+opset_import {
+  version: 8
+}

test/onnx/onnx_import_dyn_shapes.in.cpp

@@ -550,3 +550,29 @@ NGRAPH_TEST(onnx_dyn_shapes_${BACKEND_NAME}, expand_uint16_dyn_shape)
 
     test_case.run();
 }
+
+NGRAPH_TEST(onnx_${BACKEND_NAME}, model_tile)
+{
+    auto function =
+        onnx_import::import_onnx_model(file_util::path_join(SERIALIZED_ZOO, "onnx/tile.prototxt"));
+
+    auto test_case =
+        ngraph::test::NgraphTestCase(function, "${BACKEND_NAME}", BackendMode::DYNAMIC);
+    test_case.add_input<std::int16_t>({0, 1, 2, 3, 4, 5}); // input
+    test_case.add_input<std::int16_t>({2, 1});             // repeats
+    test_case.add_expected_output<std::int16_t>(Shape{4, 3}, {0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5});
+    test_case.run();
+}
+
+NGRAPH_TEST(onnx_${BACKEND_NAME}, model_tile_static)
+{
+    auto function = onnx_import::import_onnx_model(
+        file_util::path_join(SERIALIZED_ZOO, "onnx/tile_static.prototxt"));
+
+    auto test_case =
+        ngraph::test::NgraphTestCase(function, "${BACKEND_NAME}", BackendMode::DYNAMIC);
+    test_case.add_input<std::int16_t>({0, 1, 2, 3, 4, 5}); // input
+    test_case.add_expected_output<std::int16_t>(
+        Shape{4, 6}, {0, 1, 2, 0, 1, 2, 3, 4, 5, 3, 4, 5, 0, 1, 2, 0, 1, 2, 3, 4, 5, 3, 4, 5});
+    test_case.run();
+}