之前分享了利用FastDet&tensorrt多线程推理的代码,本想着继续学习yolo&tensorrt多线程的代码,但是现在shouxieai直接开源的该项目,而且还包含yolov8实例分割的代码。因此本文主要是对项目代码进行梳理,一方面加深自己对多线程、cuda编程的理解,另一方面希望给有需要的同学提供帮助。本文主要对项目整体的框架进行说明,后续的文章将会对代码细节更加详细的注释。
yolo&tensorrt项目:https://github.com/shouxieai/infer
fastdet&ncnn 项目:https://blog.csdn.net/weixin_42108183/article/details/129331588
首先下载项目
CMakeLists.txt
cmake_minimum_required(VERSION 2.6)
# set (CMAKE_C_COMPILER "/usr/bin/gcc")
# set (CMAKE_CXX_COMPILER "/usr/bin/g++")
project(demo_trt)
option(CUDA_USE_STATIC_CUDA_RUNTIME OFF)
set(CMAKE_CXX_STANDARD 17)
set(CMAKE_BUILD_TYPE Debug)
set(EXECUTABLE_OUTPUT_PATH ${PROJECT_SOURCE_DIR}/workspace)
set(HAS_PYTHON OFF)
# 修改为本机地址
set(TENSORRT_DIR "/home/rex/TensorRT-8.4.0.6")
find_package(CUDA REQUIRED)
find_package(OpenCV)
include_directories(${PROJECT_SOURCE_DIR}/src${PROJECT_SOURCE_DIR}/src/trt${PROJECT_SOURCE_DIR}/src/trt/common${OpenCV_INCLUDE_DIRS}${CUDA_TOOLKIT_ROOT_DIR}/include${TENSORRT_DIR}/include${CUDNN_DIR}/include
)
link_directories(${TENSORRT_DIR}/lib${CUDA_TOOLKIT_ROOT_DIR}/lib64${CUDNN_DIR}/lib
)set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -Wall -O0 -Wfatal-errors -pthread -w -g")
set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -std=c++11 -O0 -Xcompiler -fPIC -g -w ${CUDA_GEN_CODE}")
file(GLOB_RECURSE cpp_srcs ${PROJECT_SOURCE_DIR}/src/*.cpp)
file(GLOB_RECURSE cuda_srcs ${PROJECT_SOURCE_DIR}/src/*.cu)
cuda_add_library(plugin_list SHARED ${cuda_srcs})
target_link_libraries(plugin_list nvinfer nvinfer_plugin)
target_link_libraries(plugin_list cuda cublas cudart cudnn)
target_link_libraries(plugin_list pthread)
target_link_libraries(plugin_list ${OpenCV_LIBS})
add_executable(demo_infer ${cpp_srcs})
target_link_libraries(demo_infer plugin_list)
# 生成engine
sh /workspace/build.sh
# 修改文件中的路径为绝对路径
mkdirt build && cd build
cmake ..
make -j
./../workspace/demo_infer
src
// 单batch推理
void single_inference()
{ // 推荐写绝对路径cv::Mat image = cv::imread("/home/rex/Desktop/infer-lastest/workspace/inference/car.jpg");auto yolo = yolo::load("/home/rex/Desktop/infer-lastest/workspace/yolov8n-seg.b1.transd.engine", yolo::Type::V8Seg);if (yolo == nullptr)return;auto objs = yolo->forward(cvimg(image));int i = 0;cv::Mat image_mask = image.clone();int img_h = image_mask.rows;int img_w = image_mask.cols;std::vector color;srand(time(0));for (int i = 0; i < 80; i++){int b = rand() % 256;int g = rand() % 256;int r = rand() % 256;color.push_back(cv::Scalar(b, g, r));}for (auto &obj : objs){ // 实例分割if (obj.seg){cv::Mat mask = cv::Mat(obj.seg->height, obj.seg->width, CV_8UC1, obj.seg->data);cv::resize(mask, mask, cv::Size((obj.right - obj.left), (obj.bottom - obj.top)), cv::INTER_NEAREST);inRange(mask, 127, 255, mask);cv::imwrite(cv::format("%d_mask.jpg", i), mask);i++;cv::Rect rect(obj.left, obj.top, (obj.right - obj.left), (obj.bottom - obj.top));cv::Mat c = image_mask(rect);c.setTo(color[obj.class_label], mask);}uint8_t b, g, r;tie(b, g, r) = yolo::random_color(obj.class_label);cv::rectangle(image, cv::Point(obj.left, obj.top),cv::Point(obj.right, obj.bottom), cv::Scalar(b, g, r), 5);auto name = cocolabels[obj.class_label];auto caption = cv::format("%s %.2f", name, obj.confidence);int width = cv::getTextSize(caption, 0, 1, 2, nullptr).width + 10;cv::rectangle(image, cv::Point(obj.left - 3, obj.top - 33),cv::Point(obj.left + width, obj.top), cv::Scalar(b, g, r),-1);cv::putText(image, caption, cv::Point(obj.left, obj.top - 5), 0, 1,cv::Scalar::all(0), 2, 16);}// 实例分割addWeighted(image, 0.5, image_mask, 0.5, 0, image);printf("Save result to Result.jpg, %d objects\n", objs.size());cv::imwrite("Result.jpg", image);
}
本文将简单的将infer项目运行起来了,初步理解该项目的运行方式,后面的文章将会对代码细节进行更加详细的学习!