• yolov10--C#接口

    一、前言

         本章主要讲解yolov10的C#接口，主要是使用微软开发的openvinocsharp工具加载yolov10模型，并做推理。

    二、yolov10模型转换

         这里为了演示，使用官方yolov10m模型（其他大小的模型同理）做演示，可从下方下载，当然也可以是自己训练好的模型

    https://github.com/THU-MIG/yolov10/releases/download/v1.1/yolov10m.pt

          该原始模型，需要被转换为openvinocsharp所支持的模型格式，为此需要建立一个yolov10的python环境，使用conda创建，requirements.txt 为 yolov10官方代码下的所需包

    1. conda create -n yolov10 python=3.9
    2. conda activate yolov10
    3. pip install -r requirements.txt
    4. pip install -e .

    然后安装OpenVINO™环境，输入以下指令

    pip install openvino==2024.1.0

    在该创建好的虚拟环境，cd 至下载好的yolov10m.pt 所在目录，执行

    1. yolo export model=yolov10m.pt format=onnx opset=11 simplify
    2. ovc yolov10m.onnx

    这样，pt文件的模型将转换为onnx文件，再换换为 openvino 所需的 bin 和 xml 文件格式

    三、C#端openvinosharp相关包安装

             首先需使用VS2022构建项目，其次将openvinosharp相关包安装上

    1. "Microsoft.NET.Sdk">
    2.  
    3.   
    4.     Exe
    5.     net6.0
    6.     enable
    7.     enable
    8.   
    9.  
    10.   
    11.     "OpenCvSharp4" Version="4.9.0.20240103" />
    12.     "OpenCvSharp4.Extensions" Version="4.9.0.20240103" />
    13.     "OpenCvSharp4.runtime.win" Version="4.9.0.20240103" />
    14.     "OpenVINO.CSharp.API" Version="2024.1.0.1" />
    15.     "OpenVINO.CSharp.API.Extensions.OpenCvSharp" Version="1.0.4" />
    16.     "OpenVINO.runtime.win" Version="2024.1.0.1" />
    17.   
    18.

    也就是  OpenCvSharp4、OpenCvSharp4.Extensions、OpenCvSharp4.runtime.win、OpenVINO.CSharp.API、OpenVINO.CSharp.API.Extensions.OpenCvSharp、OpenVINO.runtime.win，这6个包给装上

    这部分参考自下面博客

    【OpenVINO™】在C#中使用 OpenVINO™ 部署 YOLOv10 模型实现目标_yolov10 openvino-CSDN博客

    四、C#加载yolo10推理代码

    这样就可以创建C# winform项目，愉快地加载前面转换好的模型文件做前向推理了

    1. // See https://aka.ms/new-console-template for more information
    2. //Console.WriteLine("Hello, World!");
    3.  
    4. using System.Reflection;
    5. using System.Runtime.InteropServices;
    6. using System;
    7. using OpenVinoSharp;
    8. using OpenVinoSharp.Extensions.utility;
    9. using OpenVinoSharp.Extensions;
    10. using OpenCvSharp;
    11. using OpenCvSharp.Dnn;
    12. using OpenVinoSharp.preprocess;
    13.  
    14. namespace yolov10_det_opencvsharp
    15. {
    16.     internal class Program
    17.     {
    18.         static void Main(string[] args)
    19.         {
    20.             string model_path = "./model_demo/yolov10m.xml";
    21.             string image_path = "./model_demo/cat.png";
    22.             string device = "AUTO";  //CPU GPU AUTO，可选AUTO模式
    23.  
    24.       
    25.  
    26.             // -------- Get OpenVINO runtime version --------
    27.  
    28.             OpenVinoSharp.Version version = Ov.get_openvino_version();
    29.  
    30.             Slog.INFO("---- OpenVINO INFO----");
    31.             Slog.INFO("Description : " + version.description);
    32.             Slog.INFO("Build number: " + version.buildNumber);
    33.  
    34.             Slog.INFO("Predict model files: " + model_path);
    35.             Slog.INFO("Predict image  files: " + image_path);
    36.             Slog.INFO("Inference device: " + device);
    37.             Slog.INFO("Start yolov10 model inference.");
    38.  
    39.             //yolov10_det(model_path, image_path, device);
    40.             yolov10_det_process(model_path, image_path , device);
    41.         }
    42.         static void yolov10_det(string model_path, string image_path, string device)
    43.         {
    44.             DateTime start = DateTime.Now;
    45.             // -------- Step 1. Initialize OpenVINO Runtime Core --------
    46.             Core core = new Core();
    47.             DateTime end = DateTime.Now;
    48.             Slog.INFO("1. Initialize OpenVINO Runtime Core success, time spend: " + (end - start).TotalMilliseconds + "ms.");
    49.             // -------- Step 2. Read inference model --------
    50.             start = DateTime.Now;
    51.             Model model = core.read_model(model_path);
    52.             end = DateTime.Now;
    53.             Slog.INFO("2. Read inference model success, time spend: " + (end - start).TotalMilliseconds + "ms.");
    54.             OvExtensions.printf_model_info(model);
    55.             // -------- Step 3. Loading a model to the device --------
    56.             start = DateTime.Now;
    57.             CompiledModel compiled_model = core.compile_model(model, device);
    58.             end = DateTime.Now;
    59.             Slog.INFO("3. Loading a model to the device success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    60.             // -------- Step 4. Create an infer request --------
    61.             start = DateTime.Now;
    62.             InferRequest infer_request = compiled_model.create_infer_request();
    63.             end = DateTime.Now;
    64.             Slog.INFO("4. Create an infer request success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    65.             // -------- Step 5. Process input images --------
    66.             start = DateTime.Now;
    67.             Mat image = new Mat(image_path); // Read image by opencvsharp
    68.             int max_image_length = image.Cols > image.Rows ? image.Cols : image.Rows;
    69.             Mat max_image = Mat.Zeros(new OpenCvSharp.Size(max_image_length, max_image_length), MatType.CV_8UC3);
    70.             Rect roi = new Rect(0, 0, image.Cols, image.Rows);
    71.             image.CopyTo(new Mat(max_image, roi));
    72.             float factor = (float)(max_image_length / 640.0);
    73.             end = DateTime.Now;
    74.             Slog.INFO("5. Process input images success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    75.             // -------- Step 6. Set up input data --------
    76.             start = DateTime.Now;
    77.             Tensor input_tensor = infer_request.get_input_tensor();
    78.             Shape input_shape = input_tensor.get_shape();
    79.             Mat input_mat = CvDnn.BlobFromImage(max_image, 1.0 / 255.0, new OpenCvSharp.Size(input_shape[2], input_shape[3]), 0, true, false);
    80.             float[] input_data = new float[input_shape[1] * input_shape[2] * input_shape[3]];
    81.             Marshal.Copy(input_mat.Ptr(0), input_data, 0, input_data.Length);
    82.             input_tensor.set_data<float>(input_data);
    83.  
    84.             end = DateTime.Now;
    85.             Slog.INFO("6. Set up input data success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    86.             // -------- Step 7. Do inference synchronously --------
    87.             infer_request.infer();
    88.             start = DateTime.Now;
    89.             infer_request.infer();
    90.             end = DateTime.Now;
    91.             Slog.INFO("7. Do inference synchronously success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    92.             // -------- Step 8. Get infer result data --------
    93.             start = DateTime.Now;
    94.             Tensor output_tensor = infer_request.get_output_tensor();
    95.             int output_length = (int)output_tensor.get_size();
    96.             float[] output_data = output_tensor.get_data<float>(output_length);
    97.             end = DateTime.Now;
    98.             Slog.INFO("8. Get infer result data success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    99.  
    100.              -------- Step 9. Process reault  --------
    101.             start = DateTime.Now;
    102.  
    103.             List position_boxes = new List();
    104.             List<int> class_ids = new List<int>();
    105.             List<float> confidences = new List<float>();
    106.  
    107.             // Preprocessing output results
    108.             for (int i = 0; i < output_data.Length / 6; i++)
    109.             {
    110.                 int s = 6 * i;
    111.                 if ((float)output_data[s + 4] > 0.5)
    112.                 {
    113.                     float cx = output_data[s + 0];
    114.                     float cy = output_data[s + 1];
    115.                     float dx = output_data[s + 2];
    116.                     float dy = output_data[s + 3];
    117.                     int x = (int)((cx) * factor);
    118.                     int y = (int)((cy) * factor);
    119.                     int width = (int)((dx - cx) * factor);
    120.                     int height = (int)((dy - cy) * factor);
    121.                     Rect box = new Rect();
    122.                     box.X = x;
    123.                     box.Y = y;
    124.                     box.Width = width;
    125.                     box.Height = height;
    126.  
    127.                     position_boxes.Add(box);
    128.                     class_ids.Add((int)output_data[s + 5]);
    129.                     confidences.Add((float)output_data[s + 4]);
    130.                 }
    131.             }
    132.  
    133.  
    134.             end = DateTime.Now;
    135.             Slog.INFO("9. Process reault  success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    136.             for (int i = 0; i < class_ids.Count; i++)
    137.             {
    138.                 int index = i;
    139.                 Cv2.Rectangle(image, position_boxes[index], new Scalar(0, 0, 255), 2, LineTypes.Link8);
    140.                 Cv2.Rectangle(image, new OpenCvSharp.Point(position_boxes[index].TopLeft.X, position_boxes[index].TopLeft.Y + 30),
    141.                     new OpenCvSharp.Point(position_boxes[index].BottomRight.X, position_boxes[index].TopLeft.Y), new Scalar(0, 255, 255), -1);
    142.                 Cv2.PutText(image, class_ids[index] + "-" + confidences[index].ToString("0.00"),
    143.                     new OpenCvSharp.Point(position_boxes[index].X, position_boxes[index].Y + 25),
    144.                     HersheyFonts.HersheySimplex, 0.8, new Scalar(0, 0, 0), 2);
    145.             }
    146.             string output_path = Path.Combine(Path.GetDirectoryName(Path.GetFullPath(image_path)),
    147.                 Path.GetFileNameWithoutExtension(image_path) + "_result.jpg");
    148.             Cv2.ImWrite(output_path, image);
    149.             Slog.INFO("The result save to " + output_path);
    150.             Cv2.ImShow("Result", image);
    151.             Cv2.WaitKey(0);
    152.         }
    153.         static void yolov10_det_process(string model_path, string image_path, string device)
    154.         {
    155.             DateTime start = DateTime.Now;
    156.             // -------- Step 1. Initialize OpenVINO Runtime Core --------
    157.             Core core = new Core();
    158.             DateTime end = DateTime.Now;
    159.             Slog.INFO("1. Initialize OpenVINO Runtime Core success, time spend: " + (end - start).TotalMilliseconds + "ms.");
    160.             // -------- Step 2. Read inference model --------
    161.             start = DateTime.Now;
    162.             Model model = core.read_model(model_path);
    163.             end = DateTime.Now;
    164.             Slog.INFO("2. Read inference model success, time spend: " + (end - start).TotalMilliseconds + "ms.");
    165.             OvExtensions.printf_model_info(model);
    166.             PrePostProcessor processor = new PrePostProcessor(model);
    167.             Tensor input_tensor_pro = new Tensor(new OvType(ElementType.U8), new Shape(1, 640, 640, 3));   //注意这个地方要改和模型窗口一致，模型是640，这里也要640
    168.             InputInfo input_info = processor.input(0);
    169.             InputTensorInfo input_tensor_info = input_info.tensor();
    170.             input_tensor_info.set_from(input_tensor_pro).set_layout(new Layout("NHWC")).set_color_format(ColorFormat.BGR);
    171.  
    172.             PreProcessSteps process_steps = input_info.preprocess();
    173.             process_steps.convert_color(ColorFormat.RGB).resize(ResizeAlgorithm.RESIZE_LINEAR)
    174.                 .convert_element_type(new OvType(ElementType.F32)).scale(255.0f).convert_layout(new Layout("NCHW"));
    175.             Model new_model = processor.build();
    176.             // -------- Step 3. Loading a model to the device --------
    177.             start = DateTime.Now;
    178.             CompiledModel compiled_model = core.compile_model(new_model, device);
    179.             end = DateTime.Now;
    180.             Slog.INFO("3. Loading a model to the device success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    181.             // -------- Step 4. Create an infer request --------
    182.             start = DateTime.Now;
    183.             InferRequest infer_request = compiled_model.create_infer_request();
    184.             end = DateTime.Now;
    185.             Slog.INFO("4. Create an infer request success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    186.             // -------- Step 5. Process input images --------
    187.             start = DateTime.Now;
    188.             Mat image = new Mat(image_path); // Read image by opencvsharp
    189.             int max_image_length = image.Cols > image.Rows ? image.Cols : image.Rows;
    190.             Mat max_image = Mat.Zeros(new OpenCvSharp.Size(max_image_length, max_image_length), MatType.CV_8UC3);
    191.             Rect roi = new Rect(0, 0, image.Cols, image.Rows);
    192.             image.CopyTo(new Mat(max_image, roi));
    193.             Cv2.Resize(max_image, max_image, new OpenCvSharp.Size(640, 640));   //注意这个地方要改和模型窗口一致，模型是640，这里也要640
    194.             float factor = (float)(max_image_length / 640.0);
    195.             end = DateTime.Now;
    196.             Slog.INFO("5. Process input images success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    197.             // -------- Step 6. Set up input data --------
    198.             start = DateTime.Now;
    199.             Tensor input_tensor = infer_request.get_input_tensor();
    200.             Shape input_shape = input_tensor.get_shape();
    201.             byte[] input_data = new byte[input_shape[1] * input_shape[2] * input_shape[3]];
    202.             //max_image.GetArray(out input_data);
    203.             Marshal.Copy(max_image.Ptr(0), input_data, 0, input_data.Length);
    204.             IntPtr destination = input_tensor.data();
    205.             Marshal.Copy(input_data, 0, destination, input_data.Length);
    206.  
    207.             end = DateTime.Now;
    208.             Slog.INFO("6. Set up input data success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    209.             // -------- Step 7. Do inference synchronously --------
    210.             infer_request.infer();
    211.             start = DateTime.Now;
    212.             infer_request.infer();
    213.   
    214.             end = DateTime.Now;
    215.             Slog.INFO("7. Do inference synchronously success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    216.  
    217.             // -------- Step 8. Get infer result data --------
    218.             start = DateTime.Now;
    219.             Tensor output_tensor = infer_request.get_output_tensor();
    220.             int output_length = (int)output_tensor.get_size();
    221.             float[] output_data = output_tensor.get_data<float>(output_length);
    222.             end = DateTime.Now;
    223.             Slog.INFO("8. Get infer result data success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    224.  
    225.              -------- Step 9. Process reault  --------
    226.             start = DateTime.Now;
    227.  
    228.             List position_boxes = new List();
    229.             List<int> class_ids = new List<int>();
    230.             List<float> confidences = new List<float>();
    231.  
    232.             // Preprocessing output results
    233.             for (int i = 0; i < output_data.Length / 6; i++)
    234.             {
    235.                 int s = 6 * i;
    236.                 if ((float)output_data[s + 4] > 0.2)
    237.                 {
    238.                     float cx = output_data[s + 0];
    239.                     float cy = output_data[s + 1];
    240.                     float dx = output_data[s + 2];
    241.                     float dy = output_data[s + 3];
    242.                     int x = (int)((cx) * factor);
    243.                     int y = (int)((cy) * factor);
    244.                     int width = (int)((dx - cx) * factor);
    245.                     int height = (int)((dy - cy) * factor);
    246.                     Rect box = new Rect();
    247.                     box.X = x;
    248.                     box.Y = y;
    249.                     box.Width = width;
    250.                     box.Height = height;
    251.  
    252.                     position_boxes.Add(box);
    253.                     class_ids.Add((int)output_data[s + 5]);
    254.                     confidences.Add((float)output_data[s + 4]);
    255.                 }
    256.             }
    257.  
    258.  
    259.             end = DateTime.Now;
    260.             Slog.INFO("9. Process reault  success, time spend:" + (end - start).TotalMilliseconds + "ms.");
    261.             for (int i = 0; i < class_ids.Count; i++)
    262.             {
    263.                 int index = i;
    264.                 Cv2.Rectangle(image, position_boxes[index], new Scalar(0, 0, 255), 2, LineTypes.Link8);
    265.                 Cv2.Rectangle(image, new OpenCvSharp.Point(position_boxes[index].TopLeft.X, position_boxes[index].TopLeft.Y + 30),
    266.                     new OpenCvSharp.Point(position_boxes[index].BottomRight.X, position_boxes[index].TopLeft.Y), new Scalar(0, 255, 255), -1);
    267.                 Cv2.PutText(image, class_ids[index] + "-" + confidences[index].ToString("0.00"),
    268.                     new OpenCvSharp.Point(position_boxes[index].X, position_boxes[index].Y + 25),
    269.                     HersheyFonts.HersheySimplex, 0.8, new Scalar(0, 0, 0), 2);
    270.             }
    271.             string output_path = Path.Combine(Path.GetDirectoryName(Path.GetFullPath(image_path)),
    272.                 Path.GetFileNameWithoutExtension(image_path) + "_result.jpg");
    273.             Cv2.ImWrite(output_path, image);
    274.             Slog.INFO("The result save to " + output_path);
    275.             Cv2.ImShow("Result", image);
    276.             Cv2.WaitKey(0);
    277.         }
    278.     }
    279. }
    280.  
    281.

    五、输出结果

            运行代码，可以得到统计的代码加载、预处理、推理的运行时间，并且得到识别结果，类别号、置信度、以及位置

           有显卡的话，可将模型AUTO改为GPU，运行时间会更快些。。。

  • 相关阅读:
    面向高速公路车辆切入场景的自动驾驶测试用例生成方法
    Java锁常见面试题
    OpenCV 4.0.0学习笔记 (一) 图像与视频的读写
    C语言题收录（四）
    RabbitMQ 消息队列学习 (四)
    《MongoDB入门教程》第15篇 文档更新之$inc操作符
    【MicroPython ESP32】ssd1306 0.96“OLED+气象数据中文显示
    01【什么是设计模式】
    springMVC01,【第一个springMVC例子(注解版)：HelloWorld】
    围剿小程序
  • 原文地址：https://blog.csdn.net/AP1005834/article/details/139779710