C#中使用YOLO的常用方式的詳細指南

更新時間：2025年05月15日 09:18:12 作者：LeonDL168

在C#中使用YOLO（You Only Look Once）目標檢測算法,通常有幾種不同的實現方式,例如OpenCV,ONNX Runtime等,下面我們來看看具體的操作方法吧

方法一：使用ONNX Runtime部署YOLOv5模型

這種方法的優(yōu)勢在于實現簡單，且性能優(yōu)化較好。

具體步驟

1.準備YOLOv5模型

首先，從YOLOv5官方倉庫（https://github.com/ultralytics/yolov5 ）下載預訓練模型。

然后，將PyTorch模型轉換為ONNX格式：

python export.py --weights yolov5s.pt --include onnx

2.創(chuàng)建C#項目并安裝NuGet包

新建一個.NET Core控制臺應用程序。

通過NuGet包管理器安裝以下組件：

Microsoft.ML.OnnxRuntime
System.Drawing.Common（用于圖像處理）

實現代碼

下面是一個完整的示例代碼

using System;
using System.Collections.Generic;
using System.Drawing;
using System.Drawing.Imaging;
using System.IO;
using System.Linq;
using Microsoft.ML.OnnxRuntime;
using Microsoft.ML.OnnxRuntime.Tensors;

namespace YoloV5OnnxRuntime
{
    class Program
    {
        // 檢測類別名稱
        private static readonly string[] classNames = new[] {
            "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"
        };

        static void Main(string[] args)
        {
            // 模型路徑和輸入圖像路徑
            string modelPath = "yolov5s.onnx";
            string imagePath = "test.jpg";
            
            // 加載模型
            using var session = new InferenceSession(modelPath);
            
            // 加載并預處理圖像
            var (inputTensor, originalWidth, originalHeight) = LoadImage(imagePath);
            
            // 準備輸入參數
            var inputs = new List<NamedOnnxValue> { NamedOnnxValue.CreateFromTensor("images", inputTensor) };
            
            // 運行模型推理
            using var results = session.Run(inputs);
            
            // 處理輸出結果
            var outputTensor = results.First().AsTensor<float>();
            var predictions = ProcessOutput(outputTensor, originalWidth, originalHeight);
            
            // 繪制檢測結果
            DrawPredictions(imagePath, predictions, "output.jpg");
            
            Console.WriteLine("檢測完成，結果已保存至 output.jpg");
        }

        // 加載并預處理圖像
        private static (DenseTensor<float> tensor, int width, int height) LoadImage(string imagePath)
        {
            using var image = Image.FromFile(imagePath);
            int originalWidth = image.Width;
            int originalHeight = image.Height;
            
            // 調整圖像大小為模型輸入尺寸（通常為640x640）
            var resizedImage = ResizeImage(image, 640, 640);
            
            // 圖像轉張量
            var tensor = new DenseTensor<float>(new[] { 1, 3, 640, 640 });
            var mean = new[] { 0.0f, 0.0f, 0.0f }; // 均值
            var std = new[] { 1.0f, 1.0f, 1.0f };  // 標準差
            
            using (var bitmap = new Bitmap(resizedImage))
            {
                for (int y = 0; y < bitmap.Height; y++)
                {
                    for (int x = 0; x < bitmap.Width; x++)
                    {
                        var pixel = bitmap.GetPixel(x, y);
                        
                        // 注意：圖像通道順序為BGR（OpenCV默認），而模型可能需要RGB
                        tensor[0, 0, y, x] = (pixel.R / 255.0f - mean[0]) / std[0];
                        tensor[0, 1, y, x] = (pixel.G / 255.0f - mean[1]) / std[1];
                        tensor[0, 2, y, x] = (pixel.B / 255.0f - mean[2]) / std[2];
                    }
                }
            }
            
            return (tensor, originalWidth, originalHeight);
        }

        // 調整圖像大小
        private static Image ResizeImage(Image image, int width, int height)
        {
            var destRect = new Rectangle(0, 0, width, height);
            var destImage = new Bitmap(width, height);
            
            destImage.SetResolution(image.HorizontalResolution, image.VerticalResolution);
            
            using (var graphics = Graphics.FromImage(destImage))
            {
                graphics.CompositingMode = System.Drawing.Drawing2D.CompositingMode.SourceCopy;
                graphics.CompositingQuality = System.Drawing.Drawing2D.CompositingQuality.HighQuality;
                graphics.InterpolationMode = System.Drawing.Drawing2D.InterpolationMode.HighQualityBicubic;
                graphics.SmoothingMode = System.Drawing.Drawing2D.SmoothingMode.HighQuality;
                graphics.PixelOffsetMode = System.Drawing.Drawing2D.PixelOffsetMode.HighQuality;
                
                using (var wrapMode = new System.Drawing.Imaging.ImageAttributes())
                {
                    wrapMode.SetWrapMode(System.Drawing.Drawing2D.WrapMode.TileFlipXY);
                    graphics.DrawImage(image, destRect, 0, 0, image.Width, image.Height, GraphicsUnit.Pixel, wrapMode);
                }
            }
            
            return destImage;
        }

        // 處理模型輸出
        private static List<Prediction> ProcessOutput(Tensor<float> output, int originalWidth, int originalHeight)
        {
            const float confidenceThreshold = 0.5f;
            const float nmsThreshold = 0.4f;
            
            var predictions = new List<Prediction>();
            
            // 解析模型輸出
            for (int i = 0; i < output.Length / 85; i++)
            {
                float confidence = output[0, i, 4];
                
                if (confidence >= confidenceThreshold)
                {
                    // 找到概率最高的類別
                    int classId = 0;
                    float maxClassProbs = 0;
                    
                    for (int c = 0; c < 80; c++)
                    {
                        float classProbs = output[0, i, 5 + c];
                        if (classProbs > maxClassProbs)
                        {
                            maxClassProbs = classProbs;
                            classId = c;
                        }
                    }
                    
                    float score = confidence * maxClassProbs;
                    
                    if (score >= confidenceThreshold)
                    {
                        // 獲取邊界框坐標
                        float cx = output[0, i, 0];
                        float cy = output[0, i, 1];
                        float w = output[0, i, 2];
                        float h = output[0, i, 3];
                        
                        // 轉換為左上角和右下角坐標
                        float x1 = (cx - w / 2) / 640.0f * originalWidth;
                        float y1 = (cy - h / 2) / 640.0f * originalHeight;
                        float x2 = (cx + w / 2) / 640.0f * originalWidth;
                        float y2 = (cy + h / 2) / 640.0f * originalHeight;
                        
                        predictions.Add(new Prediction
                        {
                            ClassId = classId,
                            Score = score,
                            BBox = new RectangleF(x1, y1, x2 - x1, y2 - y1)
                        });
                    }
                }
            }
            
            // 非極大值抑制
            return NonMaxSuppression(predictions, nmsThreshold);
        }

        // 非極大值抑制
        private static List<Prediction> NonMaxSuppression(List<Prediction> predictions, float threshold)
        {
            var result = new List<Prediction>();
            
            // 按置信度降序排序
            predictions = predictions.OrderByDescending(p => p.Score).ToList();
            
            while (predictions.Count > 0)
            {
                var best = predictions[0];
                result.Add(best);
                predictions.RemoveAt(0);
                
                // 移除重疊度高的邊界框
                predictions = predictions.Where(p => IoU(best.BBox, p.BBox) < threshold).ToList();
            }
            
            return result;
        }

        // 計算交并比
        private static float IoU(RectangleF a, RectangleF b)
        {
            float areaA = a.Width * a.Height;
            if (areaA <= 0) return 0;
            
            float areaB = b.Width * b.Height;
            if (areaB <= 0) return 0;
            
            float minX = Math.Max(a.Left, b.Left);
            float minY = Math.Max(a.Top, b.Top);
            float maxX = Math.Min(a.Right, b.Right);
            float maxY = Math.Min(a.Bottom, b.Bottom);
            
            float intersectionWidth = maxX - minX;
            float intersectionHeight = maxY - minY;
            
            if (intersectionWidth <= 0 || intersectionHeight <= 0)
                return 0;
                
            float intersectionArea = intersectionWidth * intersectionHeight;
            return intersectionArea / (areaA + areaB - intersectionArea);
        }

        // 繪制預測結果
        private static void DrawPredictions(string inputImagePath, List<Prediction> predictions, string outputImagePath)
        {
            using var image = Image.FromFile(inputImagePath);
            using var graphics = Graphics.FromImage(image);
            
            // 設置繪圖質量
            graphics.SmoothingMode = System.Drawing.Drawing2D.SmoothingMode.AntiAlias;
            graphics.InterpolationMode = System.Drawing.Drawing2D.InterpolationMode.HighQualityBicubic;
            graphics.PixelOffsetMode = System.Drawing.Drawing2D.PixelOffsetMode.HighQuality;
            
            // 繪制邊界框和標簽
            foreach (var prediction in predictions)
            {
                var bbox = prediction.BBox;
                var label = $"{classNames[prediction.ClassId]}: {prediction.Score:F2}";
                
                // 繪制邊界框
                using var pen = new Pen(Color.FromArgb(255, 255, 0, 0), 2);
                graphics.DrawRectangle(pen, bbox.X, bbox.Y, bbox.Width, bbox.Height);
                
                // 繪制標簽背景
                using var font = new Font("Arial", 10, FontStyle.Bold);
                using var brush = new SolidBrush(Color.FromArgb(255, 255, 0, 0));
                using var textBrush = new SolidBrush(Color.White);
                
                var textSize = graphics.MeasureString(label, font);
                var textBackground = new RectangleF(bbox.X, bbox.Y - textSize.Height, textSize.Width, textSize.Height);
                
                graphics.FillRectangle(brush, textBackground);
                graphics.DrawString(label, font, textBrush, bbox.X, bbox.Y - textSize.Height);
            }
            
            // 保存結果圖像
            image.Save(outputImagePath, ImageFormat.Jpeg);
        }
    }

    // 預測結果類
    public class Prediction
    {
        public int ClassId { get; set; }
        public float Score { get; set; }
        public RectangleF BBox { get; set; }
    }
}

方法二：使用ML.NET部署YOLOv5模型

ML.NET是微軟的跨平臺機器學習框架，也可用于部署YOLO模型。

具體步驟

1.準備工作與方法一相同

下載并轉換YOLOv5模型。

創(chuàng)建C#項目并安裝必要的NuGet包：

Microsoft.ML
Microsoft.ML.OnnxTransformer

實現代碼

下面是使用ML.NET的示例代碼：

using System;
using System.Collections.Generic;
using System.Drawing;
using System.Drawing.Imaging;
using System.IO;
using System.Linq;
using Microsoft.ML;
using Microsoft.ML.Data;
using Microsoft.ML.OnnxTransformer;
using Microsoft.ML.Trainers;

namespace YoloV5MLNet
{
    class Program
    {
        // 檢測類別名稱（同上）
        private static readonly string[] classNames = new[] { ... }; // 同上，省略
        
        static void Main(string[] args)
        {
            string modelPath = "yolov5s.onnx";
            string imagePath = "test.jpg";
            
            // 創(chuàng)建MLContext
            var mlContext = new MLContext();
            
            // 定義模型輸入輸出架構
            var data = new List<ImageInputData> { new ImageInputData { ImagePath = imagePath } };
            var dataView = mlContext.Data.LoadFromEnumerable(data);
            
            // 定義數據轉換管道
            var pipeline = mlContext.Transforms.LoadImages(outputColumnName: "image", imageFolder: "", inputColumnName: nameof(ImageInputData.ImagePath))
                .Append(mlContext.Transforms.ResizeImages(outputColumnName: "image", imageWidth: 640, imageHeight: 640, inputColumnName: "image"))
                .Append(mlContext.Transforms.ExtractPixels(outputColumnName: "images", inputColumnName: "image", interleavePixelColors: true, offsetImage: 0))
                .Append(mlContext.Transforms.ApplyOnnxModel(
                    shapeDictionary: new Dictionary<string, int[]>()
                    {
                        { "images", new[] { 1, 3, 640, 640 } },
                        { "output", new[] { 1, 25200, 85 } }
                    },
                    inputColumnNames: new[] { "images" },
                    outputColumnNames: new[] { "output" },
                    modelFile: modelPath));
            
            // 訓練管道（這里只是為了創(chuàng)建預測引擎）
            var model = pipeline.Fit(dataView);
            
            // 創(chuàng)建預測引擎
            var predictor = mlContext.Model.CreatePredictionEngine<ImageInputData, ImagePrediction>(model);
            
            // 進行預測
            var prediction = predictor.Predict(new ImageInputData { ImagePath = imagePath });
            
            // 處理輸出結果
            var originalImage = Image.FromFile(imagePath);
            var predictions = ProcessOutput(prediction.Output, originalImage.Width, originalImage.Height);
            
            // 繪制檢測結果
            DrawPredictions(imagePath, predictions, "output.jpg");
            
            Console.WriteLine("檢測完成，結果已保存至 output.jpg");
        }

        // 圖像處理和結果解析方法（與ONNX Runtime版本相同）
        private static List<Prediction> ProcessOutput(float[] output, int originalWidth, int originalHeight)
        {
            // 與ONNX Runtime版本中的ProcessOutput方法相同
            // ...
        }

        // 非極大值抑制和IoU計算方法（同上）
        private static List<Prediction> NonMaxSuppression(List<Prediction> predictions, float threshold)
        {
            // ...
        }
        
        private static float IoU(RectangleF a, RectangleF b)
        {
            // ...
        }

        // 繪制預測結果（同上）
        private static void DrawPredictions(string inputImagePath, List<Prediction> predictions, string outputImagePath)
        {
            // ...
        }
    }

    // 數據模型類
    public class ImageInputData
    {
        [LoadColumn(0)]
        public string ImagePath { get; set; }
    }

    public class ImagePrediction
    {
        [ColumnName("output")]
        public float[] Output { get; set; }
    }

    // 預測結果類（同上）
    public class Prediction
    {
        public int ClassId { get; set; }
        public float Score { get; set; }
        public RectangleF BBox { get; set; }
    }
}