Working switch between cuda and non cuda; code not working

Author: MatthijsBurgh

CMakeLists.txt

@@ -20,37 +20,48 @@ find_package(catkin REQUIRED
 )
 
 
+set(${PROJECT_NAME}_CUDA_ENABLED ${onnxruntime_ros_CUDA_ENABLED})
+if(onnxruntime_ros_CUDA_ENABLED)
+  find_package(CUDAToolkit REQUIRED)
+endif()
+
+configure_file(include/${PROJECT_NAME}/config.hpp.in ${CATKIN_DEVEL_PREFIX}/${CATKIN_GLOBAL_INCLUDE_DESTINATION}/${PROJECT_NAME}/config.hpp)
+# add_custom_target(generate_config_hpp
+#   DEPENDS ${CATKIN_DEVEL_PREFIX}/${CATKIN_GLOBAL_INCLUDE_DESTINATION}/${PROJECT_NAME}/config.hpp
+# )
+
 # ------------------------------------------------------------------------------------------------
 #                                        CATKIN EXPORT
 # ------------------------------------------------------------------------------------------------
 
 catkin_package(
-  INCLUDE_DIRS include
+  INCLUDE_DIRS include ${CATKIN_DEVEL_PREFIX}/${CATKIN_GLOBAL_INCLUDE_DESTINATION}
   LIBRARIES ${PROJECT_NAME}
   CATKIN_DEPENDS
-  DEPENDS OpenCV
+  DEPENDS CUDAToolkit OpenCV
 )
 
+
 # ------------------------------------------------------------------------------------------------
 #                                           BUILD
 # ------------------------------------------------------------------------------------------------
 
 include_directories(
   include
+  ${CATKIN_DEVEL_PREFIX}/${CATKIN_GLOBAL_INCLUDE_DESTINATION}
   SYSTEM
+  ${CUDAToolkit_INCLUDE_DIRS}
   ${OpenCV_INCLUDE_DIRS}
   ${catkin_INCLUDE_DIRS}
 )
 
 # -------------- ONNXRuntime  ------------------#
 
-# -------------- Cuda ------------------#
-add_definitions(-DUSE_CUDA=1)
-
 add_library(${PROJECT_NAME}
   src/yolo_inference.cpp
 )
-target_link_libraries(${PROJECT_NAME} ${OpenCV_LIBRARIES} ${catkin_LIBRARIES})
+# add_dependencies(${PROJECT_NAME} generate_config_hpp)
+target_link_libraries(${PROJECT_NAME} CUDA::toolkit ${OpenCV_LIBRARIES} ${catkin_LIBRARIES})
 
 add_executable(test_${PROJECT_NAME}
   src/main.cpp
@@ -65,6 +76,35 @@ if (WIN32)
             $<TARGET_FILE_DIR:${PROJECT_NAME}>)
 endif ()
 
+
+# ------------------------------------------------------------------------------------------------
+#                                               INSTALL
+# ------------------------------------------------------------------------------------------------
+
+install(FILES
+  ${CATKIN_DEVEL_PREFIX}/${CATKIN_PACKAGE_INCLUDE_DESTINATION}/config.hpp
+  DESTINATION ${CATKIN_PACKAGE_INCLUDE_DESTINATION}
+)
+
+install(
+  DIRECTORY include/${PROJECT_NAME}/
+  DESTINATION ${CATKIN_PACKAGE_INCLUDE_DESTINATION}
+)
+
+install(
+  TARGETS
+    ${PROJECT_NAME}
+  ARCHIVE DESTINATION ${CATKIN_PACKAGE_LIB_DESTINATION}
+  LIBRARY DESTINATION ${CATKIN_PACKAGE_LIB_DESTINATION}
+  RUNTIME DESTINATION ${CATKIN_GLOBAL_BIN_DESTINATION}
+)
+
+install(
+  TARGETS
+    ${PROJECT_NAME}
+  DESTINATION ${CATKIN_PACKAGE_BIN_DESTINATION}
+)
+
 # Download https://raw.githubusercontent.com/ultralytics/ultralytics/main/ultralytics/cfg/datasets/coco.yaml
 # and put it in the same folder of the executable file
 configure_file(data/coco.yaml ${CMAKE_CURRENT_BINARY_DIR}/coco.yaml COPYONLY)

include/yolo_onnx_ros/config.hpp.in

@@ -0,0 +1,7 @@
+#ifndef YOLO_ONNX_ROS_CONFIG_HPP_
+#define YOLO_ONNX_ROS_CONFIG_HPP_
+
+//Set which version of the Tree Interface to use
+#define YOLO_ONNX_ROS_CUDA_ENABLED @onnx_yolo_ros_CUDA_ENABLED@
+
+#endif //#define YOLO_ONNX_ROS_CONFIG_HPP_

include/yolo_onnx_ros/yolo_inference.h

@@ -14,7 +14,9 @@
 #include <opencv2/opencv.hpp>
 #include <onnxruntime_cxx_api.h>
 
-#ifdef USE_CUDA
+#include <yolo_onnx_ros/config.hpp>
+
+#ifdef YOLO_ONNX_ROS_CUDA_ENABLED
 #include <cuda_fp16.h>
 #endif
 
@@ -75,21 +77,21 @@ class YOLO_V8
 
     // Note: The logic is on the .cpp file since its a private method.
     template<typename N>
-    char* TensorProcess(clock_t& starttime_1, const cv::Mat& iImg, N& blob, std::vector<int64_t>& inputNodeDims,
+    char* TensorProcess(clock_t& starttime_1, N& blob, std::vector<int64_t>& inputNodeDims,
         std::vector<DL_RESULT>& oResult);
 
     char* PreProcess(const cv::Mat& iImg, std::vector<int> iImgSize, cv::Mat& oImg);
 
-    Ort::Env env;
-    std::unique_ptr<Ort::Session> session;
-    bool cudaEnable;
-    Ort::RunOptions options;
-    std::vector<const char*> inputNodeNames;
-    std::vector<const char*> outputNodeNames;
-
-    MODEL_TYPE modelType;
-    std::vector<int> imgSize;
-    float rectConfidenceThreshold;
-    float iouThreshold;
-    float resizeScales;//letterbox scale
+    Ort::Env env_;
+    std::unique_ptr<Ort::Session> session_;
+    bool cudaEnable_;
+    Ort::RunOptions options_;
+    std::vector<const char*> inputNodeNames_;
+    std::vector<const char*> outputNodeNames_;
+
+    MODEL_TYPE modelType_;
+    std::vector<int> imgSize_;
+    float rectConfidenceThreshold_;
+    float iouThreshold_;
+    float resizeScales_; //letterbox scale
 };

src/main.cpp

@@ -1,6 +1,7 @@
 #include <iostream>
 #include <iomanip>
 #include "yolo_onnx_ros/yolo_inference.h"
+#include <yolo_onnx_ros/config.hpp>
 #include <filesystem>
 #include <fstream>
 #include <random>
@@ -76,7 +77,7 @@ void Classifier(std::unique_ptr<YOLO_V8>& p)
             const char* ret = p->RunSession(img, res);
 
             float positionY = 50;
-            for (int i = 0; i < res.size(); i++)
+            for (uint i = 0; i < res.size(); i++)
             {
                 int r = dis(gen);
                 int g = dis(gen);
@@ -157,7 +158,7 @@ void DetectTest()
     params.iouThreshold = 0.5;
     params.modelPath = "yolo11m.onnx";
     params.imgSize = { 640, 640 };
-#ifdef USE_CUDA
+#ifdef YOLO_ONNX_ROS_CUDA_ENABLED
     params.cudaEnable = true;
 
     // GPU FP32 inference