视频输入c++ 调用 libtorch推理

qianbo_insist2024-04-30 13:29

1、支持GPU情况

libtorch 支持GPU情况比较奇怪，目前2.3 版本需要在链接器里面加上以下命令，否则不会支持gpu

-INCLUDE:?ignore_this_library_placeholder@@YAHXZ

2 探测是否支持

加一个函数看你是否支持torch，不然不清楚，看到支持gpu才行

c 复制代码 
void IsSupportCuda()
{
    if (torch::cuda::is_available())
    {
        std::cout << "支持GPU" << std::endl;
    }
    else
    {
        std::cout << "不支持GPU" << std::endl;
    }
    torch::Tensor tensor = torch::rand({ 5,3 });
    torch::Device device1(torch::kCUDA);
    tensor.to(device1);
    std::cout << tensor <<"--"<< tensor.options() << std::endl;
};

int main() {
    IsSupportCuda();
    return 0;
}

转化

使用命令转，如下图所示

c 复制代码 
yolo export model=yolov8s.pt imgsz=640 format=torchscript

成功以后在目录下面生成文件yolov8s.torchscript

c++ 调用

c 复制代码 
int main() {
    //IsSupportCuda();

    //return 0;
    // Device
    torch::Device device(torch::cuda::is_available() ? torch::kCUDA : torch::kCPU);

    // Note that in this example the classes are hard-coded
    std::vector<std::string> classes{ "person", "bicycle", "car", "motorcycle", "airplane", "bus", "train", "truck", "boat", "traffic light", "fire hydrant",
                                      "stop sign", "parking meter", "bench", "bird", "cat", "dog", "horse", "sheep", "cow", "elephant", "bear", "zebra",
                                      "giraffe", "backpack", "umbrella", "handbag", "tie", "suitcase", "frisbee", "skis", "snowboard", "sports ball", "kite",
                                      "baseball bat", "baseball glove", "skateboard", "surfboard", "tennis racket", "bottle", "wine glass", "cup", "fork", "knife",
                                      "spoon", "bowl", "banana", "apple", "sandwich", "orange", "broccoli", "carrot", "hot dog", "pizza", "donut", "cake", "chair",
                                      "couch", "potted plant", "bed", "dining table", "toilet", "tv", "laptop", "mouse", "remote", "keyboard", "cell phone",
                                      "microwave", "oven", "toaster", "sink", "refrigerator", "book", "clock", "vase", "scissors", "teddy bear", "hair drier", "toothbrush" };

    try {
        // Load the model (e.g. yolov8s.torchscript)
        std::string model_path = "./yolov8s.torchscript";
        torch::jit::script::Module yolo_model;
        yolo_model = torch::jit::load(model_path);
        yolo_model.eval();
        yolo_model.to(device, torch::kFloat32);

        // Load image and preprocess
        cv::Mat image = cv::imread("d:/bus.jpg");
        cv::Mat input_image;
        letterbox(image, input_image, { 640, 640 });

        torch::Tensor image_tensor = torch::from_blob(input_image.data, { input_image.rows, input_image.cols, 3 }, torch::kByte).to(device);
        //auto image_tensor_float = image_tensor.to(torch::kFloat32);
        //image_tensor_float /= 255.0;
        image_tensor = image_tensor.toType(torch::kFloat32).div(255);
        image_tensor = image_tensor.permute({ 2, 0, 1 });
        image_tensor = image_tensor.unsqueeze(0);
        std::vector<torch::jit::IValue> inputs{ image_tensor };

        // Inference
        torch::Tensor output = yolo_model.forward(inputs).toTensor().cpu();

        // NMS
        auto keep = non_max_suppression(output)[0];
        auto boxes = keep.index({ Slice(), Slice(None, 4) });
        keep.index_put_({ Slice(), Slice(None, 4) }, scale_boxes({ input_image.rows, input_image.cols }, boxes, { image.rows, image.cols }));

        // Show the results
        for (int i = 0; i < keep.size(0); i++) {
            int x1 = keep[i][0].item().toFloat();
            int y1 = keep[i][1].item().toFloat();
            int x2 = keep[i][2].item().toFloat();
            int y2 = keep[i][3].item().toFloat();
            float conf = keep[i][4].item().toFloat();
            int cls = keep[i][5].item().toInt();
            std::cout << "Rect: [" << x1 << "," << y1 << "," << x2 << "," << y2 << "]  Conf: " << conf << "  Class: " << classes[cls] << std::endl;
        }
        getchar();
    }
    catch (const c10::Error& e) {
        std::cout << e.msg() << std::endl;
    }

    return 0;
}

解码后视频帧调用

先使用opencv,同时使用硬件加速，使用硬件解码，新版本的使用方法已经不一样了，以下先举个例子，指定使用ffmpeg

c 复制代码 
int main() {
    // 创建一个 VideoCapture 对象，并指定使用 FFmpeg 作为后端  
    cv::ocl::setUseOpenCL(true);
    if (!cv::ocl::haveOpenCL()) {
        std::cerr << "OpenCL is not available.\n";
        return -1;
    }
    else {
        std::cout << cv::ocl::Device().getDefault().name() << std::endl;
    }

   cv::VideoCapture cap1("rtsp://127.0.0.1/99-640.mkv", cv::CAP_FFMPEG, {
            cv::CAP_PROP_HW_ACCELERATION,(int)cv::VIDEO_ACCELERATION_D3D11,
            cv::CAP_PROP_HW_DEVICE, 0
   });

   // cv::VideoCapture cap2("d:/8k.mp4", cv::CAP_FFMPEG);


    // 检查是否成功打开视频文件  
    if (!cap1.isOpened()) {
        std::cerr << "Error opening video file" << std::endl;
        return -1;
    }
   
    // 检查是否支持硬件加速
    double hw1 = cap1.get(cv::CAP_PROP_HW_ACCELERATION);
   // double hw2 = cap2.get(cv::CAP_PROP_HW_ACCELERATION);
    if (hw1 >= cv::VIDEO_ACCELERATION_ANY ) {
        // 支持硬件加速，尝试启用
        //cap.set(cv::CAP_PROP_HW_ACCELERATION, cv::VIDEO_ACCELERATION_ANY);
        std::cout << "Hardware acceleration enabled" << std::endl;
    }
    else {
        std::cout << "Hardware acceleration not supported or not available" << std::endl;
    }
    // 设置硬件加速（如果支持的话）  
    // 注意：不是所有的平台和驱动程序都支持硬件加速  
   // cap.set(cv::CAP_PROP_HW_ACCELERATION, cv::VIDEO_ACCELERATION_ANY);  
    cv::UMat uFrame,Frame; // GPU 上的 UMat 对象，用于直接接收解码后的数据  
    // 读取并处理视频帧  
    while (true) {
        // 尝试直接从 VideoCapture 读取帧到 UMat  
        cv::UMat m;

        bool ret = cap1.read(m);
        if (!ret) {
            std::cout << "End of video" << std::endl;
            break;
        }
        
        int w = m.cols;
        int h = m.rows;
        //m.copyTo(uFrame);
        
        //uFrame = m.getUMat(cv::ACCESS_READ);

        cv::UMat m1,m2,m3,m4,m5;
        //视频矫正
        rectify_umat(m, w, h, default_K0, default_D0, m1);
        //cv::Mat m1, m2, m3,m4;
        //放大增强
        cv::resize(m1, m2, cv::Size(w*2, h*2),0,0,cv::INTER_CUBIC);

        letterbox(m2, input_image, { 640, 640 });

        torch::Tensor image_tensor = torch::from_blob(input_image.data, { input_image.rows, input_image.cols, 3 }, torch::kByte).to(device);
        //auto image_tensor_float = image_tensor.to(torch::kFloat32);
        //image_tensor_float /= 255.0;
        image_tensor = image_tensor.toType(torch::kFloat32).div(255);
        image_tensor = image_tensor.permute({ 2, 0, 1 });
        image_tensor = image_tensor.unsqueeze(0);
        std::vector<torch::jit::IValue> inputs{ image_tensor };

        // Inference
        torch::Tensor output = yolo_model.forward(inputs).toTensor().cpu();

        // NMS
        auto keep = non_max_suppression(output)[0];
        auto boxes = keep.index({ Slice(), Slice(None, 4) });
        keep.index_put_({ Slice(), Slice(None, 4) }, scale_boxes({ input_image.rows, input_image.cols }, boxes, { image.rows, image.cols }));

        // Show the results
        for (int i = 0; i < keep.size(0); i++) {
            int x1 = keep[i][0].item().toFloat();
            int y1 = keep[i][1].item().toFloat();
            int x2 = keep[i][2].item().toFloat();
            int y2 = keep[i][3].item().toFloat();
            float conf = keep[i][4].item().toFloat();
            int cls = keep[i][5].item().toInt();
            std::cout << "Rect: [" << x1 << "," << y1 << "," << x2 << "," << y2 << "]  Conf: " << conf << "  Class: " << classes[cls] << std::endl;
        }

        func_3_umat(m2, m3);
        func_1_umat(m3, m4);
        func_0_umat(m4, m5);
     
        cv::imshow("m", m);
        cv::imshow("m2", m2);
        cv::imshow("res", m5);
        if (cv::waitKey(10) == 'q') {
            break;
        }
    }

    // 释放资源  
    cap1.release();
    //cap2.release();
    cv::destroyAllWindows();

    return 0;
}

改进

未完待续。。。

相关推荐
木子.李3471 小时前
排序算法总结（C++）
c++·算法·排序算法
freyazzr2 小时前
C++八股 | Day2 | atom/函数指针/指针函数/struct、Class/静态局部变量、局部变量、全局变量/强制类型转换
c++
fpcc3 小时前
跟我学c++中级篇——理解类型推导和C++不同版本的支持
开发语言·c++
终焉代码3 小时前
STL解析——list的使用
开发语言·c++
DevangLic4 小时前
【 *p取出内容 &a得到地址】
c++
鑫鑫向栄4 小时前
[蓝桥杯]修改数组
数据结构·c++·算法·蓝桥杯·动态规划
鑫鑫向栄4 小时前
[蓝桥杯]带分数
数据结构·c++·算法·职场和发展·蓝桥杯
m0_552200825 小时前
《UE5_C++多人TPS完整教程》学习笔记37 ——《P38 变量复制（Variable Replication）》
c++·游戏·ue5
小wanga5 小时前
【递归、搜索与回溯】专题三 穷举vs暴搜vs回溯vs剪枝
c++·算法·机器学习·剪枝
Code_流苏6 小时前
C++课设：简易日历程序（支持传统节假日 + 二十四节气 + 个人纪念日管理）
开发语言·c++·stl容器·课设·期末大作业·日历程序·面向对象设计
热门推荐
01YOLOv8入门 | 重要性能衡量指标、训练结果评价及分析及影响mAP的因素【发论文关注的指标】02KGG转MP3工具|非KGM文件|解密音频03从零安装 LLaMA-Factory 微调 Qwen 大模型成功及所有的坑04【无人机】无人机通信模块，无人机图数传模块的介绍，数传，图传，图传数传一体电台，05【SpeedAI科研小助手】2分钟极速解决知网维普重复率、AIGC率过高，一键全文降！文件格式不变，公式都保留的！06DeepSeek各版本说明与优缺点分析07神经网络架构KAN确实具有一些独特的特点及底层原理和应用场景08VMware虚拟机安装Win7专业版保姆级教程（附镜像包）09组基轨迹建模 GBTM的介绍与实现（Stata 或 R）10柬埔寨语翻译通App，一款真正实现高棉语翻译、语音识别和中柬双语无障碍交流的应用程序。