Support hardware acceleration on non-cuda platforms (#21)

* Support hardware acceleration on non-cuda platforms

* Ensure hardware accel is used on DML

* Fix accidental delete

* Fix AMD

Author: ooeygui

.gitignore

@@ -0,0 +1 @@
+ros_msft_onnx/testdata/model.onnx

ros_msft_onnx/CMakeLists.txt

@@ -17,6 +17,9 @@ if(MSVC)
 
 endif()
 
+add_definitions(-DBOOST_BIND_GLOBAL_PLACEHOLDERS)
+
+
 option(CUDA_SUPPORT "use CUDA support onnxruntime library" OFF)
 
 find_package(Eigen3 REQUIRED)
@@ -65,9 +68,9 @@ if(CUDA_SUPPORT)
   set(PACKAGE_URL "https://www.nuget.org/api/v2/package/Microsoft.ML.OnnxRuntime.Gpu/1.7.1")
   set(PACKAGE_SHA512 "41112118007aae34fcc38100152df6e6fa5fc567e61aa4ded42a26d39751f1be7ec225c0d73799f065015e284f0fb9bd7e0835c733e9abad5b0243a391411f8d")
 else()
-  set(ONNX_RUNTIME "Microsoft.ML.OnnxRuntime.1.7.0")
-  set(PACKAGE_URL "https://www.nuget.org/api/v2/package/Microsoft.ML.OnnxRuntime/1.7.0")
-  set(PACKAGE_SHA512 "1fc15386bdfa455f457e50899e3c9c454aafbdc345799dcf4ecfd6990a9dbd8cd7f0b1f3bf412c47c900543c535f95aa1cb1e14e9851cb9b600c60a981f38a50")
+  set(ONNX_RUNTIME "Microsoft.ML.OnnxRuntime.DirectML.1.7.0")
+  set(PACKAGE_URL "https://www.nuget.org/api/v2/package/Microsoft.ML.OnnxRuntime.DirectML/1.7.0")
+  set(PACKAGE_SHA512 "2e5bd2c0ade72444d4efdfbd6a75571aaa72045769f9b5847186129c9e5e667ad080d5d2b9a12cce88c9eee68302be89cdb7030ccefa3d572e591b1c453c7340")
 endif()
 
 file(DOWNLOAD

ros_msft_onnx/launch/tracker.launch

@@ -3,6 +3,8 @@
 
   <node pkg="ros_msft_onnx" type="ros_msft_onnx_node" name="ros_msft_onnx" output="screen">
     <param name="image_topic" value="$(eval '/camera/image_raw' if os_windows_arg else '/cv_camera/image_raw')"/>
+     <param name="tensor_width" value="416"/>
+     <param name="tensor_height" value="416"/>
   </node>
 
   <!-- The camera node will be selected based on os. ros_msft_camera for Windows and cv_camera for others. -->

ros_msft_onnx/src/main.cpp

@@ -6,17 +6,17 @@
 
 #include "ros_msft_onnx/ros_msft_onnx.h"
 
+#ifdef _WIN32
+#include <objbase.h>
+#endif
+
 using namespace std;
 
 int main(int argc, char **argv)
 {
-    /*
-    ROS_WARN("ONNX: Waiting for Debugger");
-    while (!IsDebuggerPresent())
-    {
-        Sleep(5);
-    }
-    */
+    #ifdef _WIN32
+    HRESULT hr = CoInitializeEx(NULL, COINIT_MULTITHREADED);
+    #endif
 
     ros::init(argc, argv, "ros_msft_onnx");
 
@@ -39,5 +39,7 @@ int main(int argc, char **argv)
         return 1;
     }
 
-
+#ifdef _WIN32
+    CoUninitialize();
+#endif
 }

ros_msft_onnx/src/ros_msft_onnx.cpp

@@ -1,4 +1,5 @@
 #include <ros/ros.h>
+#include <boost/bind/bind.hpp>
 #include <cv_bridge/cv_bridge.h>
 #include <image_transport/image_transport.h>
 #include <visualization_msgs/MarkerArray.h>
@@ -13,6 +14,9 @@
 #include <codecvt>
 #include <fstream>
 #include <sstream>
+#ifdef _WIN32
+#include "dml_provider_factory.h"
+#endif
 
 using namespace std;
 
@@ -141,11 +145,16 @@ bool OnnxProcessor::init(ros::NodeHandle &nh, ros::NodeHandle &nhPrivate)
         // initialize session options if needed
         Ort::SessionOptions session_options;
         session_options.SetIntraOpNumThreads(1);
+        session_options.DisableMemPattern();    // Required for DirectML
+        session_options.SetExecutionMode(ExecutionMode::ORT_SEQUENTIAL);    // Required for DirectML
         session_options.SetGraphOptimizationLevel(GraphOptimizationLevel::ORT_ENABLE_EXTENDED);
 
         std::string modelString = _onnxModel;
 
 #ifdef _WIN32
+        // Select Device 0
+        OrtSessionOptionsAppendExecutionProvider_DML((OrtSessionOptions*)session_options, 0);
+
         auto modelFullPath = to_wstring(modelString).c_str();
 #else
         auto modelFullPath = _onnxModel.c_str();
@@ -290,7 +299,7 @@ bool OnnxTracker::init(ros::NodeHandle &nh, ros::NodeHandle &nhPrivate)
     _nh = nh;
     _nhPrivate = nhPrivate;
 
-    f = boost::bind(&OnnxTracker::callback, this, _1, _2);
+    f = boost::bind(&OnnxTracker::callback, this, boost::placeholders::_1, boost::placeholders::_2);
     server.setCallback(f);
 
     return _status;

ros_msft_onnx/src/yolo_box.cpp

@@ -112,7 +112,6 @@ namespace yolo
 
                     if (_debug)
                     {
-                        ROS_INFO("ONNX: %s detected!", label.c_str());
                         // Draw a bounding box on the CV image
                         cv::Scalar color(255, 255, 0);
                         cv::Rect box;