【测试通过环境】

win10 x64
vs2019
cuda11.7+cudnn8.8.0
TensorRT-8.6.1.6
opencvsharp==4.9.0
.NET Framework4.7.2

NVIDIA GeForce RTX 2070 Super

版本和上述环境版本不一样的需要重新编译TensorRtExtern.dll，TensorRtExtern源码地址：TensorRT-CSharp-API/src/TensorRtExtern at TensorRtSharp2.0 · guojin-yan/TensorRT-CSharp-API · GitHub

Windows版 CUDA安装参考：Windows版 CUDA安装_win cuda安装-CSDN博客

【特别注意】

tensorrt依赖不同硬件需要自己从onnx转换tensorrt，转换就是调用api实现，比如

TensorRtSharp.Custom.Nvinfer.OnnxToEngine(@"yolov8s-seg.onnx",1024);

【视频演示和解说】

【部分实现源码】

using System;
using System.Collections.Generic;
using System.ComponentModel;
using System.Data;
using System.Drawing;
using System.Linq;
using System.Text;
using System.Threading.Tasks;
using System.Windows.Forms;
using FIRC;
using OpenCvSharp;
using TrtCommon;
using TensorRtSharp;
using TensorRtSharp.Custom;
using System.Diagnostics;namespace WindowsFormsApp1
{public partial class Form1 : Form{public Form1(){InitializeComponent();}private void button1_Click(object sender, EventArgs e){var detector = new Yolov8Seg("yolov8s-seg.engine");Mat image1 = Cv2.ImRead(@"person.jpg");var Results = detector.Predict(new List<Mat> { image1 });Mat re_image1 = Visualize.DrawSegResult(Results[0], image1);Cv2.NamedWindow("result",WindowFlags.KeepRatio);Cv2.ImShow("result", re_image1);Cv2.WaitKey(0);}private void button2_Click(object sender, EventArgs e){TensorRtSharp.Custom.Nvinfer.OnnxToEngine(@"yolov8s-seg.onnx",1024);}private void button3_Click(object sender, EventArgs e){var detector = new Yolov8Seg("yolov8s-seg.engine");VideoCapture capture = new VideoCapture(0);if (!capture.IsOpened()){Console.WriteLine("video not open!");return;}Mat frame = new Mat();var sw = new Stopwatch();int fps = 0;while (true){capture.Read(frame);if (frame.Empty()){Console.WriteLine("data is empty!");break;}sw.Start();var results = detector.Predict(new List<Mat> { frame });Mat resultImg = Visualize.DrawSegResult(results[0], frame);sw.Stop();fps = Convert.ToInt32(1 / sw.Elapsed.TotalSeconds);sw.Reset();Cv2.PutText(resultImg, "FPS=" + fps, new OpenCvSharp.Point(30, 30), HersheyFonts.HersheyComplex, 1.0, new Scalar(255, 0, 0), 3);//显示结果Cv2.ImShow("Result", resultImg);int key = Cv2.WaitKey(10);if (key == 27)break;}capture.Release();}}
}

【演示源码下载地址】

注意源码提供上面对应环境的dll，只需要安装上面一样cuda+cudnn和tensorrt版本即可正常运行。如果您不安装一样版本不能正常运行。此时需要重新编译TensorRtExtern.dll，此外由于tensorrt依赖硬件不一样电脑可能无法共用tensorrt模型，所以必须要重新转换onnx模型到engine才可以运行。