一. yolov11 python環境安裝

0. 基礎環境
   CUDA：cuda_11.8.0_522.06_windows
   cudnn：cudnn-windows-x86_64-8.6.0.163_cuda11-archive
1. 創建python環境

   conda create --name yolov11 python=3.10 -y
   

2. 安裝pytorch

   pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu118
   

3. 安裝yolov11

   pip install ultralytics -i https://pypi.mirrors.ustc.edu.cn/simple/
   

4. 安裝必要的庫

   pip install -r requirements.txt -i https://pypi.mirrors.ustc.edu.cn/simple/
   

二. windows10下yolov11 tensorrtx推理加速

1. 官網下載tensorrtx

   git clone https://github.com/wang-xinyu/tensorrtx.git
   

2. 進入yolov11文件夾，轉換模型.pt轉.wts

   python gen_wts.py -w D:\code\ultralytics-main\yolo11n.pt -o yolo11n.wts -t detect
   

   
   注意：pytorch2.6需要修改代碼中的torch.load，在里面添加weigths_only=False
   

3. 修改cmakeList.txt文件
   根據自己的opencv，tensort，dirent所在目錄路徑，修改以下文件路徑

   cmake_minimum_required(VERSION 3.10)project(yolov11)add_definitions(-std=c++11)
   add_definitions(-DAPI_EXPORTS)
   add_compile_definitions(NOMINMAX)set(CMAKE_CXX_STANDARD 11)
   set(CMAKE_BUILD_TYPE Debug)set(CMAKE_CUDA_ARCHITECTURES 70 75 80 86)
   set(CMAKE_CUDA_COMPILER "C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.8/bin/nvcc.exe")
   enable_language(CUDA)include_directories(${PROJECT_SOURCE_DIR}/include)
   include_directories(${PROJECT_SOURCE_DIR}/plugin)# include and link dirs of cuda and tensorrt, you need adapt them if yours are different
   if(CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")message("embed_platform on")include_directories(/usr/local/cuda/targets/aarch64-linux/include)link_directories(/usr/local/cuda/targets/aarch64-linux/lib)
   else()message("embed_platform off")# cudafind_package(CUDA REQUIRED)include_directories(${CUDA_INCLUDE_DIRS})# tensorrtset(TRT_DIR "C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\TensorRT-8.6.0.12")  set(TRT_INCLUDE_DIRS ${TRT_DIR}\\include) set(TRT_LIB_DIRS ${TRT_DIR}\\lib) include_directories(${TRT_INCLUDE_DIRS})link_directories(${TRT_LIB_DIRS})# opencvset(OpenCV_DIR "D:\\Program Files\\opencv\\build") set(OpenCV_INCLUDE_DIRS ${OpenCV_DIR}\\include) set(OpenCV_LIB_DIRS ${OpenCV_DIR}\\x64\\vc16\\lib) set(OpenCV_Debug_LIBS "opencv_world4110d.lib") set(OpenCV_Release_LIBS "opencv_world4110.lib") include_directories(${OpenCV_INCLUDE_DIRS})link_directories(${OpenCV_LIB_DIRS})# direntset(Dirent_INCLUDE_DIRS "D:\\Program Files\\dirent\\include")include_directories(${Dirent_INCLUDE_DIRS})
   endif()add_library(myplugins SHARED ${PROJECT_SOURCE_DIR}/plugin/yololayer.cu)
   target_link_libraries(myplugins nvinfer cudart)file(GLOB_RECURSE SRCS ${PROJECT_SOURCE_DIR}/src/*.cpp ${PROJECT_SOURCE_DIR}/src/*.cu)add_executable(yolo11_det ${PROJECT_SOURCE_DIR}/yolo11_det.cpp ${SRCS})
   target_link_libraries(yolo11_det nvinfer)
   target_link_libraries(yolo11_det cudart)
   target_link_libraries(yolo11_det myplugins)
   target_link_libraries(yolo11_det ${OpenCV_Debug_LIBS})
   target_link_libraries(yolo11_det ${OpenCV_Release_LIBS})add_executable(yolo11_cls ${PROJECT_SOURCE_DIR}/yolo11_cls.cpp ${SRCS})
   target_link_libraries(yolo11_cls nvinfer)
   target_link_libraries(yolo11_cls cudart)
   target_link_libraries(yolo11_cls myplugins)
   target_link_libraries(yolo11_cls ${OpenCV_Debug_LIBS})
   target_link_libraries(yolo11_cls ${OpenCV_Release_LIBS})add_executable(yolo11_seg ${PROJECT_SOURCE_DIR}/yolo11_seg.cpp ${SRCS})
   target_link_libraries(yolo11_seg nvinfer)
   target_link_libraries(yolo11_seg cudart)
   target_link_libraries(yolo11_seg myplugins)
   target_link_libraries(yolo11_seg ${OpenCV_Debug_LIBS})
   target_link_libraries(yolo11_seg ${OpenCV_Release_LIBS})add_executable(yolo11_pose ${PROJECT_SOURCE_DIR}/yolo11_pose.cpp ${SRCS})
   target_link_libraries(yolo11_pose nvinfer)
   target_link_libraries(yolo11_pose cudart)
   target_link_libraries(yolo11_pose myplugins)
   target_link_libraries(yolo11_pose ${OpenCV_Debug_LIBS})
   target_link_libraries(yolo11_pose ${OpenCV_Release_LIBS})add_executable(yolo11_obb ${PROJECT_SOURCE_DIR}/yolo11_obb.cpp ${SRCS})
   target_link_libraries(yolo11_obb nvinfer)
   target_link_libraries(yolo11_obb cudart)
   target_link_libraries(yolo11_obb myplugins)
   target_link_libraries(yolo11_obb ${OpenCV_Debug_LIBS})
   target_link_libraries(yolo11_obb ${OpenCV_Release_LIBS})
   

4. 構建項目

   mkdir build
   cd build
   cmake ..
   

   5. vs打開項目，生成解決方案
   
   

5. 裝換.wts為.engine
   
   
   
   轉換前，這里需要根據自己的模型，修改對應的配置，配置文件在以下位置
   
   
   

   

   -s ..\yolo11n.wts yolo11n.engine n
   

6. 利用轉換好的.engine進行推理

   -d yolo11n.engine D:\code\yolov5-6.1\data\images g
   

   
   
   

三. windows10下qt調用tensorrtx加速的yolov11進行檢測

1. 拷貝文件
   
2. 修改Qt項目中的cmakeList.txt文件如下：

   cmake_minimum_required(VERSION 3.5)project(yolov11Test LANGUAGES CXX)add_definitions(-std=c++11)
   add_definitions(-DAPI_EXPORTS)
   add_compile_definitions(NOMINMAX)set(CMAKE_CXX_STANDARD 17)
   set(CMAKE_CXX_STANDARD_REQUIRED ON)set(CMAKE_CUDA_ARCHITECTURES 70 75 80 86)
   set(CMAKE_CUDA_COMPILER "C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.8/bin/nvcc.exe")
   enable_language(CUDA)include_directories(${PROJECT_SOURCE_DIR}/include)
   include_directories(${PROJECT_SOURCE_DIR}/plugin)# cuda
   find_package(CUDA REQUIRED)
   include_directories(${CUDA_INCLUDE_DIRS})# tensorrt
   set(TRT_DIR "C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\TensorRT-8.6.0.12")
   set(TRT_INCLUDE_DIRS ${TRT_DIR}\\include)
   set(TRT_LIB_DIRS ${TRT_DIR}\\lib)
   include_directories(${TRT_INCLUDE_DIRS})
   link_directories(${TRT_LIB_DIRS})# opencv
   set(OpenCV_DIR "D:\\Program Files\\opencv\\build")
   set(OpenCV_INCLUDE_DIRS ${OpenCV_DIR}\\include)
   set(OpenCV_LIB_DIRS ${OpenCV_DIR}\\x64\\vc16\\lib)
   set(OpenCV_Debug_LIBS "opencv_world4110d.lib")
   set(OpenCV_Release_LIBS "opencv_world4110.lib")
   include_directories(${OpenCV_INCLUDE_DIRS})
   link_directories(${OpenCV_LIB_DIRS})# dirent
   set(Dirent_INCLUDE_DIRS "D:\\Program Files\\dirent\\include")
   include_directories(${Dirent_INCLUDE_DIRS})add_library(myplugins SHARED ${PROJECT_SOURCE_DIR}/plugin/yololayer.cu)
   target_link_libraries(myplugins nvinfer cudart)file(GLOB_RECURSE SRCS ${PROJECT_SOURCE_DIR}/src/*.cpp ${PROJECT_SOURCE_DIR}/src/*.cu)add_executable(yolov11Test main.cpp ${SRCS})target_link_libraries(yolov11Test nvinfer)
   target_link_libraries(yolov11Test cudart)
   target_link_libraries(yolov11Test myplugins)
   target_link_libraries(yolov11Test ${OpenCV_Debug_LIBS})
   target_link_libraries(yolov11Test ${OpenCV_Release_LIBS})install(TARGETS yolov11TestLIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
   )

3. main函數代碼如下：

   #include <fstream>
   #include <iostream>
   #include <opencv2/opencv.hpp>
   #include "cuda_utils.h"
   #include "logging.h"
   #include "model.h"
   #include "postprocess.h"
   #include "preprocess.h"
   #include "utils.h"Logger gLogger;
   using namespace nvinfer1;
   const int kOutputSize = kMaxNumOutputBbox * sizeof(Detection) / sizeof(float) + 1;void deserialize_engine(std::string& engine_name, IRuntime** runtime, ICudaEngine** engine, IExecutionContext** context)
   {std::ifstream file(engine_name, std::ios::binary);if (!file.good()){std::cerr << "read " << engine_name << " error!" << std::endl;assert(false);}size_t size = 0;file.seekg(0, file.end);size = file.tellg();file.seekg(0, file.beg);char* serialized_engine = new char[size];assert(serialized_engine);file.read(serialized_engine, size);file.close();*runtime = createInferRuntime(gLogger);assert(*runtime);*engine = (*runtime)->deserializeCudaEngine(serialized_engine, size);assert(*engine);*context = (*engine)->createExecutionContext();assert(*context);delete[] serialized_engine;
   }void prepare_buffer(ICudaEngine* engine, float** input_buffer_device, float** output_buffer_device, float** output_buffer_host, float** decode_ptr_host, float** decode_ptr_device, std::string cuda_post_process)
   {assert(engine->getNbBindings() == 2);// In order to bind the buffers, we need to know the names of the input and output tensors.// Note that indices are guaranteed to be less than IEngine::getNbBindings()const int inputIndex = engine->getBindingIndex(kInputTensorName);const int outputIndex = engine->getBindingIndex(kOutputTensorName);assert(inputIndex == 0);assert(outputIndex == 1);// Create GPU buffers on deviceCUDA_CHECK(cudaMalloc((void**)input_buffer_device, kBatchSize * 3 * kInputH * kInputW * sizeof(float)));CUDA_CHECK(cudaMalloc((void**)output_buffer_device, kBatchSize * kOutputSize * sizeof(float)));if (cuda_post_process == "c") {*output_buffer_host = new float[kBatchSize * kOutputSize];} else if (cuda_post_process == "g") {if (kBatchSize > 1) {std::cerr << "Do not yet support GPU post processing for multiple batches" << std::endl;exit(0);}// Allocate memory for decode_ptr_host and copy to device*decode_ptr_host = new float[1 + kMaxNumOutputBbox * bbox_element];CUDA_CHECK(cudaMalloc((void**)decode_ptr_device, sizeof(float) * (1 + kMaxNumOutputBbox * bbox_element)));}
   }void infer(IExecutionContext& context, cudaStream_t& stream, void** buffers, float* output, int batchsize,float* decode_ptr_host, float* decode_ptr_device, int model_bboxes, std::string cuda_post_process) {// infer on the batch asynchronously, and DMA output back to hostauto start = std::chrono::system_clock::now();context.enqueueV2(buffers, stream, nullptr);if (cuda_post_process == "c") {CUDA_CHECK(cudaMemcpyAsync(output, buffers[1], batchsize * kOutputSize * sizeof(float), cudaMemcpyDeviceToHost,stream));auto end = std::chrono::system_clock::now();std::cout << "inference time: " << std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count()<< "ms" << std::endl;} else if (cuda_post_process == "g") {CUDA_CHECK(cudaMemsetAsync(decode_ptr_device, 0, sizeof(float) * (1 + kMaxNumOutputBbox * bbox_element), stream));cuda_decode((float*)buffers[1], model_bboxes, kConfThresh, decode_ptr_device, kMaxNumOutputBbox, stream);cuda_nms(decode_ptr_device, kNmsThresh, kMaxNumOutputBbox, stream);  //cuda nmsCUDA_CHECK(cudaMemcpyAsync(decode_ptr_host, decode_ptr_device,sizeof(float) * (1 + kMaxNumOutputBbox * bbox_element), cudaMemcpyDeviceToHost,stream));auto end = std::chrono::system_clock::now();std::cout << "inference and gpu postprocess time: "<< std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count() << "ms" << std::endl;}CUDA_CHECK(cudaStreamSynchronize(stream));
   }int main(int argc, char** argv) {// yolo11_det -s ../models/yolo11n.wts ../models/yolo11n.fp32.trt n// yolo11_det -d ../models/yolo11n.fp32.trt ../images ccudaSetDevice(kGpuId);std::string engine_name= "D:\\code\\tensorrtx\\yolo11\\build\\yolo11n.engine"; //轉換好的模型文件路徑std::string img_dir= "D:\\code\\yolov5-6.1\\data\\images\\"; //要預測的圖像文件夾所在路徑std::string cuda_post_process = "g";int model_bboxes;float gd = 0, gw = 0;int max_channels = 0;// 反序列化模型文件 Deserialize the engine from fileIRuntime* runtime = nullptr;ICudaEngine* engine = nullptr;IExecutionContext* context = nullptr;deserialize_engine(engine_name, &runtime, &engine, &context);cudaStream_t stream;CUDA_CHECK(cudaStreamCreate(&stream));cuda_preprocess_init(kMaxInputImageSize);auto out_dims = engine->getBindingDimensions(1);model_bboxes = out_dims.d[0];// 準備cpu和gpu緩存 Prepare cpu and gpu buffersfloat* device_buffers[2];float* output_buffer_host = nullptr;float* decode_ptr_host = nullptr;float* decode_ptr_device = nullptr;// 從文件夾中讀取圖像 Read images from directorystd::vector<std::string> file_names;if (read_files_in_dir(img_dir.c_str(), file_names) < 0) {std::cerr << "read_files_in_dir failed." << std::endl;return -1;}prepare_buffer(engine, &device_buffers[0], &device_buffers[1], &output_buffer_host, &decode_ptr_host,&decode_ptr_device, cuda_post_process);// 批預測batch predictfor (size_t i = 0; i < file_names.size(); i += kBatchSize){// 通過opencv讀取一批圖像Get a batch of imagesstd::vector<cv::Mat> img_batch;std::vector<std::string> img_name_batch;for (size_t j = i; j < i + kBatchSize && j < file_names.size(); j++){cv::Mat img = cv::imread(img_dir + "/" + file_names[j]);img_batch.push_back(img);img_name_batch.push_back(file_names[j]);}// Preprocesscuda_batch_preprocess(img_batch, device_buffers[0], kInputW, kInputH, stream);// 進行推理Run inferenceinfer(*context, stream, (void**)device_buffers, output_buffer_host, kBatchSize, decode_ptr_host,decode_ptr_device, model_bboxes, cuda_post_process);// 保存output_buffer_host的前100個值，一行一個//        std::ofstream out("../models/output.txt");//        for (int j = 0; j < 100; j++) {//            out << output_buffer_host[j] << std::endl;//        }//        out.close();std::vector<std::vector<Detection>> res_batch;if (cuda_post_process == "c"){// NMS非極大值抑制batch_nms(res_batch, output_buffer_host, img_batch.size(), kOutputSize, kConfThresh, kNmsThresh);} else if (cuda_post_process == "g"){//GPU非極大值抑制Process gpu decode and nms resultsbatch_process(res_batch, decode_ptr_host, img_batch.size(), bbox_element, img_batch);}// 繪制結果Draw bounding boxesdraw_bbox(img_batch, res_batch);//顯示圖像for (size_t j = 0; j < img_batch.size(); j++){cv::imshow("results", img_batch[j]);cv::waitKey(0);}// 保存圖像Save imagesfor (size_t j = 0; j < img_batch.size(); j++){cv::imwrite("_" + img_name_batch[j], img_batch[j]);}}// Release stream and bufferscudaStreamDestroy(stream);CUDA_CHECK(cudaFree(device_buffers[0]));CUDA_CHECK(cudaFree(device_buffers[1]));CUDA_CHECK(cudaFree(decode_ptr_device));delete[] decode_ptr_host;delete[] output_buffer_host;cuda_preprocess_destroy();// Destroy the enginedelete context;delete engine;delete runtime;// Print histogram of the output distribution//std::cout << "\nOutput:\n\n";//for (unsigned int i = 0; i < kOutputSize; i++)//{//    std::cout << prob[i] << ", ";//    if (i % 10 == 0) std::cout << std::endl;//}//std::cout << std::endl;return 0;
   }