C# OpenCvSharp DNN 部署yolov5不规则四边形目标检测

天天代码码天天2023-12-16 15:55

目录

效果

模型信息

项目

代码

下载

C# OpenCvSharp DNN 部署yolov5不规则四边形目标检测

效果

模型信息

Inputs

name：images

tensor：Float[1, 3, 1024, 1024]

Outputs

name：output

tensor：Float[1, 64512, 11]

项目

代码

using OpenCvSharp;

using OpenCvSharp.Dnn;

using System;

using System.Collections.Generic;

using System.Drawing;

using System.IO;

using System.Linq;

using System.Linq.Expressions;

using System.Numerics;

using System.Reflection;

using System.Windows.Forms;

namespace OpenCvSharp_DNN_Demo

{

public partial class frmMain : Form

{

public frmMain()

{

InitializeComponent();

}

string fileFilter = "*.*|*.bmp;*.jpg;*.jpeg;*.tiff;*.tiff;*.png";

string image_path = "";

DateTime dt1 = DateTime.Now;

DateTime dt2 = DateTime.Now;

float confThreshold;

float nmsThreshold;

float objThreshold;

float[,] anchors = new float[3, 6] {

{31, 30, 28, 49, 50, 31},

{46, 45, 58, 58, 74, 74},

{94, 94, 115, 115, 151, 151}

};

float[] stride = new float[3] { 8.0f, 16.0f, 32.0f };

string modelpath;

int inpHeight;

int inpWidth;

List<string> class_names;

int num_class;

Net opencv_net;

Mat BN_image;

Mat image;

Mat result_image;

private void button1_Click(object sender, EventArgs e)

{

OpenFileDialog ofd = new OpenFileDialog();

ofd.Filter = fileFilter;

if (ofd.ShowDialog() != DialogResult.OK) return;

pictureBox1.Image = null;

pictureBox2.Image = null;

textBox1.Text = "";

image_path = ofd.FileName;

pictureBox1.Image = new Bitmap(image_path);

image = new Mat(image_path);

}

private void Form1_Load(object sender, EventArgs e)

{

confThreshold = 0.5f;

nmsThreshold = 0.5f;

objThreshold = 0.5f;

modelpath = "model/best.onnx";

inpHeight = 1024;

inpWidth = 1024;

opencv_net = CvDnn.ReadNetFromOnnx(modelpath);

class_names = new List<string>();

StreamReader sr = new StreamReader("model/coco.names");

string line;

while ((line = sr.ReadLine()) != null)

{

class_names.Add(line);

}

num_class = class_names.Count();

image_path = "test_img/1.png";

pictureBox1.Image = new Bitmap(image_path);

}

float sigmoid(float x)

{

return (float)(1.0 / (1 + Math.Exp(-x)));

}

Mat ResizeImage(Mat srcimg, out int newh, out int neww, out int top, out int left)

{

int srch = srcimg.Rows, srcw = srcimg.Cols;

top = 0;

left = 0;

newh = inpHeight;

neww = inpWidth;

Mat dstimg = new Mat();

if (srch != srcw)

{

float hw_scale = (float)srch / srcw;

if (hw_scale > 1)

{

newh = inpHeight;

neww = (int)(inpWidth / hw_scale);

Cv2.Resize(srcimg, dstimg, new OpenCvSharp.Size(neww, newh), 0, 0, InterpolationFlags.Area);

left = (int)((inpWidth - neww) * 0.5);

Cv2.CopyMakeBorder(dstimg, dstimg, 0, 0, left, inpWidth - neww - left, BorderTypes.Constant);

}

else

{

newh = (int)(inpHeight * hw_scale);

neww = inpWidth;

Cv2.Resize(srcimg, dstimg, new OpenCvSharp.Size(neww, newh), 0, 0, InterpolationFlags.Area);

top = (int)((inpHeight - newh) * 0.5);

Cv2.CopyMakeBorder(dstimg, dstimg, top, inpHeight - newh - top, 0, 0, BorderTypes.Constant);

}

}

else

{

Cv2.Resize(srcimg, dstimg, new OpenCvSharp.Size(neww, newh));

}

return dstimg;

}

float IoU(BoxInfo polya, BoxInfo polyb, int max_w, int max_h)

{

List<List<OpenCvSharp.Point>> poly_array0 = new List<List<OpenCvSharp.Point>>();

List<List<OpenCvSharp.Point>> poly_array1 = new List<List<OpenCvSharp.Point>>();

poly_array0.Add(polya.pts);

poly_array1.Add(polyb.pts);

Mat _poly0 = Mat.Zeros(max_h, max_w, MatType.CV_8UC1);

Mat _poly1 = Mat.Zeros(max_h, max_w, MatType.CV_8UC1);

Mat _result = new Mat();

List<List<OpenCvSharp.Point>> _pts0 = new List<List<OpenCvSharp.Point>>();

List<int> _npts0 = new List<int>();

foreach (var item in poly_array0)

{

if (item.Count < 3)//invalid poly

return -1f;

_pts0.Add(item);

_npts0.Add(item.Count);

}

List<List<OpenCvSharp.Point>> _pts1 = new List<List<OpenCvSharp.Point>>();

List<int> _npts1 = new List<int>();

foreach (var item in poly_array1)

{

if (item.Count < 3)//invalid poly

return -1f;

_pts1.Add(item);

_npts1.Add(item.Count);

}

Cv2.FillPoly(_poly0, _pts0, new Scalar(1));

Cv2.FillPoly(_poly1, _pts1, new Scalar(1));

Cv2.BitwiseAnd(_poly0, _poly1, _result);

int _area0 = Cv2.CountNonZero(_poly0);

int _area1 = Cv2.CountNonZero(_poly1);

int _intersection_area = Cv2.CountNonZero(_result);

float _iou = (float)_intersection_area / (float)(_area0 + _area1 - _intersection_area);

return _iou;

}

void nms(List<BoxInfo> input_boxes, int max_w, int max_h)

{

input_boxes.Sort((a, b) => { return a.score > b.score ? -1 : 1; });

bool[] isSuppressed = new bool[input_boxes.Count];

for (int i = 0; i < input_boxes.Count(); ++i)

{

if (isSuppressed[i]) { continue; }

for (int j = i + 1; j < input_boxes.Count(); ++j)

{

if (isSuppressed[j]) { continue; }

float ovr = IoU(input_boxes[i], input_boxes[j], max_w, max_h);

if (ovr >= nmsThreshold)

{

isSuppressed[j] = true;

}

}

}

for (int i = isSuppressed.Length - 1; i >= 0; i--)

{

if (isSuppressed[i])

{

input_boxes.RemoveAt(i);

}

}

}

private unsafe void button2_Click(object sender, EventArgs e)

{

if (image_path == "")

{

return;

}

textBox1.Text = "检测中，请稍等......";

pictureBox2.Image = null;

Application.DoEvents();

image = new Mat(image_path);

int newh = 0, neww = 0, padh = 0, padw = 0;

Mat dstimg = ResizeImage(image, out newh, out neww, out padh, out padw);

BN_image = CvDnn.BlobFromImage(dstimg, 1 / 255.0, new OpenCvSharp.Size(inpWidth, inpHeight), new Scalar(0, 0, 0), true, false);

//配置图片输入数据

opencv_net.SetInput(BN_image);

//模型推理，读取推理结果

Mat[] outs = new Mat[3] { new Mat(), new Mat(), new Mat() };

string[] outBlobNames = opencv_net.GetUnconnectedOutLayersNames().ToArray();

dt1 = DateTime.Now;

opencv_net.Forward(outs, outBlobNames);

dt2 = DateTime.Now;

int num_proposal = outs[0].Size(1);

int nout = outs[0].Size(2);

if (outs[0].Dims > 2)

{

outs[0] = outs[0].Reshape(0, num_proposal);

}

float ratioh = 1.0f * image.Rows / newh, ratiow = 1.0f * image.Cols / neww;

float* pdata = (float*)outs[0].Data;

List<BoxInfo> generate_boxes = new List<BoxInfo>();

int row_ind = 0;

for (int n = 0; n < 3; n++)

{

int num_grid_x = (int)(inpWidth / stride[n]);

int num_grid_y = (int)(inpHeight / stride[n]);

for (int q = 0; q < 3; q++) //anchor

{

float anchor_w = anchors[n, q * 2];

float anchor_h = anchors[n, q * 2 + 1];

for (int i = 0; i < num_grid_y; i++)

{

for (int j = 0; j < num_grid_x; j++)

{

float box_score = sigmoid(pdata[8]);

if (box_score > objThreshold)

{

Mat scores = outs[0].Row(row_ind).ColRange(9, 9 + num_class);

double minVal, max_class_socre;

OpenCvSharp.Point minLoc, classIdPoint;

// Get the value and location of the maximum score

Cv2.MinMaxLoc(scores, out minVal, out max_class_socre, out minLoc, out classIdPoint);

int class_idx = classIdPoint.X;

max_class_socre = sigmoid((float)max_class_socre) * box_score;

if (max_class_socre > confThreshold)

{

List<OpenCvSharp.Point> pts = new List<OpenCvSharp.Point>();

for (int k = 0; k < 8; k += 2)

{

float x = (pdata[k] + j) * stride[n]; //x

float y = (pdata[k + 1] + i) * stride[n]; //y

x = (x - padw) * ratiow;

y = (y - padh) * ratioh;

pts.Add(new OpenCvSharp.Point(x, y));

}

Rect r = Cv2.BoundingRect(pts);

generate_boxes.Add(new BoxInfo(pts, (float)max_class_socre, class_idx));

}

}

row_ind++;

pdata += nout;

}

}

}

}

nms(generate_boxes, image.Cols, image.Rows);

result_image = image.Clone();

for (int ii = 0; ii < generate_boxes.Count; ++ii)

{

int idx = generate_boxes[ii].label;

for (int jj = 0; jj < 4; jj++)

{

Cv2.Line(result_image, generate_boxes[ii].pts[jj], generate_boxes[ii].pts[(jj + 1) % 4], new Scalar(0, 0, 255), 2);

}

string label = class_names[idx] + ":" + generate_boxes[ii].score.ToString("0.00");

int xmin = (int)generate_boxes[ii].pts[0].X;

int ymin = (int)generate_boxes[ii].pts[0].Y - 10;

Cv2.PutText(result_image, label, new OpenCvSharp.Point(xmin, ymin - 5), HersheyFonts.HersheySimplex, 0.75, new Scalar(0, 0, 255), 1);

}

pictureBox2.Image = new Bitmap(result_image.ToMemoryStream());

textBox1.Text = "推理耗时:" + (dt2 - dt1).TotalMilliseconds + "ms";

}

private void pictureBox2_DoubleClick(object sender, EventArgs e)

{

Common.ShowNormalImg(pictureBox2.Image);

}

private void pictureBox1_DoubleClick(object sender, EventArgs e)

{

Common.ShowNormalImg(pictureBox1.Image);

}

}

}

using OpenCvSharp;
using OpenCvSharp.Dnn;
using System;
using System.Collections.Generic;
using System.Drawing;
using System.IO;
using System.Linq;
using System.Linq.Expressions;
using System.Numerics;
using System.Reflection;
using System.Windows.Forms;

namespace OpenCvSharp_DNN_Demo
{
    public partial class frmMain : Form
    {
        public frmMain()
        {
            InitializeComponent();
        }

        string fileFilter = "*.*|*.bmp;*.jpg;*.jpeg;*.tiff;*.tiff;*.png";
        string image_path = "";

        DateTime dt1 = DateTime.Now;
        DateTime dt2 = DateTime.Now;

        float confThreshold;
        float nmsThreshold;
        float objThreshold;

        float[,] anchors = new float[3, 6] {
                                           {31, 30, 28, 49, 50, 31},
                                           {46, 45, 58, 58, 74, 74},
                                           {94, 94, 115, 115, 151, 151}
                                           };

        float[] stride = new float[3] { 8.0f, 16.0f, 32.0f };

        string modelpath;

        int inpHeight;
        int inpWidth;

        List<string> class_names;
        int num_class;

        Net opencv_net;
        Mat BN_image;

        Mat image;
        Mat result_image;

        private void button1_Click(object sender, EventArgs e)
        {
            OpenFileDialog ofd = new OpenFileDialog();
            ofd.Filter = fileFilter;
            if (ofd.ShowDialog() != DialogResult.OK) return;

            pictureBox1.Image = null;
            pictureBox2.Image = null;
            textBox1.Text = "";

            image_path = ofd.FileName;
            pictureBox1.Image = new Bitmap(image_path);
            image = new Mat(image_path);
        }

        private void Form1_Load(object sender, EventArgs e)
        {
            confThreshold = 0.5f;
            nmsThreshold = 0.5f;
            objThreshold = 0.5f;

            modelpath = "model/best.onnx";

            inpHeight = 1024;
            inpWidth = 1024;

            opencv_net = CvDnn.ReadNetFromOnnx(modelpath);

            class_names = new List<string>();
            StreamReader sr = new StreamReader("model/coco.names");
            string line;
            while ((line = sr.ReadLine()) != null)
            {
                class_names.Add(line);
            }
            num_class = class_names.Count();

            image_path = "test_img/1.png";
            pictureBox1.Image = new Bitmap(image_path);

        }

        float sigmoid(float x)
        {
            return (float)(1.0 / (1 + Math.Exp(-x)));
        }

        Mat ResizeImage(Mat srcimg, out int newh, out int neww, out int top, out int left)
        {
            int srch = srcimg.Rows, srcw = srcimg.Cols;
            top = 0;
            left = 0;
            newh = inpHeight;
            neww = inpWidth;
            Mat dstimg = new Mat();
            if (srch != srcw)
            {
                float hw_scale = (float)srch / srcw;
                if (hw_scale > 1)
                {
                    newh = inpHeight;
                    neww = (int)(inpWidth / hw_scale);
                    Cv2.Resize(srcimg, dstimg, new OpenCvSharp.Size(neww, newh), 0, 0, InterpolationFlags.Area);
                    left = (int)((inpWidth - neww) * 0.5);
                    Cv2.CopyMakeBorder(dstimg, dstimg, 0, 0, left, inpWidth - neww - left, BorderTypes.Constant);
                }
                else
                {
                    newh = (int)(inpHeight * hw_scale);
                    neww = inpWidth;
                    Cv2.Resize(srcimg, dstimg, new OpenCvSharp.Size(neww, newh), 0, 0, InterpolationFlags.Area);
                    top = (int)((inpHeight - newh) * 0.5);
                    Cv2.CopyMakeBorder(dstimg, dstimg, top, inpHeight - newh - top, 0, 0, BorderTypes.Constant);
                }
            }
            else
            {
                Cv2.Resize(srcimg, dstimg, new OpenCvSharp.Size(neww, newh));
            }
            return dstimg;
        }

        float IoU(BoxInfo polya, BoxInfo polyb, int max_w, int max_h)
        {
            List<List<OpenCvSharp.Point>> poly_array0 = new List<List<OpenCvSharp.Point>>();
            List<List<OpenCvSharp.Point>> poly_array1 = new List<List<OpenCvSharp.Point>>();
            poly_array0.Add(polya.pts);
            poly_array1.Add(polyb.pts);

            Mat _poly0 = Mat.Zeros(max_h, max_w, MatType.CV_8UC1);
            Mat _poly1 = Mat.Zeros(max_h, max_w, MatType.CV_8UC1);
            Mat _result = new Mat();

            List<List<OpenCvSharp.Point>> _pts0 = new List<List<OpenCvSharp.Point>>();
            List<int> _npts0 = new List<int>();

            foreach (var item in poly_array0)
            {
                if (item.Count < 3)//invalid poly
                    return -1f;

                _pts0.Add(item);
                _npts0.Add(item.Count);

            }

            List<List<OpenCvSharp.Point>> _pts1 = new List<List<OpenCvSharp.Point>>();
            List<int> _npts1 = new List<int>();

            foreach (var item in poly_array1)
            {
                if (item.Count < 3)//invalid poly
                    return -1f;

                _pts1.Add(item);
                _npts1.Add(item.Count);

            }

            Cv2.FillPoly(_poly0, _pts0, new Scalar(1));
            Cv2.FillPoly(_poly1, _pts1, new Scalar(1));

            Cv2.BitwiseAnd(_poly0, _poly1, _result);

            int _area0 = Cv2.CountNonZero(_poly0);
            int _area1 = Cv2.CountNonZero(_poly1);
            int _intersection_area = Cv2.CountNonZero(_result);
            float _iou = (float)_intersection_area / (float)(_area0 + _area1 - _intersection_area);
            return _iou;
        }

        void nms(List<BoxInfo> input_boxes, int max_w, int max_h)
        {
            input_boxes.Sort((a, b) => { return a.score > b.score ? -1 : 1; });

            bool[] isSuppressed = new bool[input_boxes.Count];

            for (int i = 0; i < input_boxes.Count(); ++i)
            {
                if (isSuppressed[i]) { continue; }
                for (int j = i + 1; j < input_boxes.Count(); ++j)
                {
                    if (isSuppressed[j]) { continue; }
                    float ovr = IoU(input_boxes[i], input_boxes[j], max_w, max_h);
                    if (ovr >= nmsThreshold)
                    {
                        isSuppressed[j] = true;
                    }
                }
            }

            for (int i = isSuppressed.Length - 1; i >= 0; i--)
            {
                if (isSuppressed[i])
                {
                    input_boxes.RemoveAt(i);
                }
            }

        }

        private unsafe void button2_Click(object sender, EventArgs e)
        {
            if (image_path == "")
            {
                return;
            }
            textBox1.Text = "检测中，请稍等......";
            pictureBox2.Image = null;
            Application.DoEvents();

            image = new Mat(image_path);

            int newh = 0, neww = 0, padh = 0, padw = 0;
            Mat dstimg = ResizeImage(image, out newh, out neww, out padh, out padw);

            BN_image = CvDnn.BlobFromImage(dstimg, 1 / 255.0, new OpenCvSharp.Size(inpWidth, inpHeight), new Scalar(0, 0, 0), true, false);

            //配置图片输入数据
            opencv_net.SetInput(BN_image);

            //模型推理，读取推理结果
            Mat[] outs = new Mat[3] { new Mat(), new Mat(), new Mat() };
            string[] outBlobNames = opencv_net.GetUnconnectedOutLayersNames().ToArray();

            dt1 = DateTime.Now;

            opencv_net.Forward(outs, outBlobNames);

            dt2 = DateTime.Now;

            int num_proposal = outs[0].Size(1);
            int nout = outs[0].Size(2);

            if (outs[0].Dims > 2)
            {
                outs[0] = outs[0].Reshape(0, num_proposal);
            }

            float ratioh = 1.0f * image.Rows / newh, ratiow = 1.0f * image.Cols / neww;

            float* pdata = (float*)outs[0].Data;

            List<BoxInfo> generate_boxes = new List<BoxInfo>();

            int row_ind = 0;

            for (int n = 0; n < 3; n++)
            {

                int num_grid_x = (int)(inpWidth / stride[n]);
                int num_grid_y = (int)(inpHeight / stride[n]);

                for (int q = 0; q < 3; q++)    //anchor
                {
                    float anchor_w = anchors[n, q * 2];
                    float anchor_h = anchors[n, q * 2 + 1];
                    for (int i = 0; i < num_grid_y; i++)
                    {
                        for (int j = 0; j < num_grid_x; j++)
                        {
                            float box_score = sigmoid(pdata[8]);
                            if (box_score > objThreshold)
                            {

                                Mat scores = outs[0].Row(row_ind).ColRange(9, 9 + num_class);
                                double minVal, max_class_socre;
                                OpenCvSharp.Point minLoc, classIdPoint;
                                // Get the value and location of the maximum score
                                Cv2.MinMaxLoc(scores, out minVal, out max_class_socre, out minLoc, out classIdPoint);

                                int class_idx = classIdPoint.X;
                                max_class_socre = sigmoid((float)max_class_socre) * box_score;
                                if (max_class_socre > confThreshold)
                                {
                                    List<OpenCvSharp.Point> pts = new List<OpenCvSharp.Point>();
                                    for (int k = 0; k < 8; k += 2)
                                    {
                                        float x = (pdata[k] + j) * stride[n];  //x
                                        float y = (pdata[k + 1] + i) * stride[n];   //y
                                        x = (x - padw) * ratiow;
                                        y = (y - padh) * ratioh;
                                        pts.Add(new OpenCvSharp.Point(x, y));
                                    }

                                    Rect r = Cv2.BoundingRect(pts);

                                    generate_boxes.Add(new BoxInfo(pts, (float)max_class_socre, class_idx));
                                }
                            }
                            row_ind++;
                            pdata += nout;
                        }
                    }

                }

            }

            nms(generate_boxes, image.Cols, image.Rows);

            result_image = image.Clone();

            for (int ii = 0; ii < generate_boxes.Count; ++ii)
            {
                int idx = generate_boxes[ii].label;

                for (int jj = 0; jj < 4; jj++)
                {
                    Cv2.Line(result_image, generate_boxes[ii].pts[jj], generate_boxes[ii].pts[(jj + 1) % 4], new Scalar(0, 0, 255), 2);
                }

                string label = class_names[idx] + ":" + generate_boxes[ii].score.ToString("0.00");

                int xmin = (int)generate_boxes[ii].pts[0].X;
                int ymin = (int)generate_boxes[ii].pts[0].Y - 10;

                Cv2.PutText(result_image, label, new OpenCvSharp.Point(xmin, ymin - 5), HersheyFonts.HersheySimplex, 0.75, new Scalar(0, 0, 255), 1);
            }

            pictureBox2.Image = new Bitmap(result_image.ToMemoryStream());
            textBox1.Text = "推理耗时:" + (dt2 - dt1).TotalMilliseconds + "ms";
        }

        private void pictureBox2_DoubleClick(object sender, EventArgs e)
        {
            Common.ShowNormalImg(pictureBox2.Image);
        }

        private void pictureBox1_DoubleClick(object sender, EventArgs e)
        {
            Common.ShowNormalImg(pictureBox1.Image);
        }
    }
}

下载

源码下载

相关推荐
plmm烟酒僧几秒前
Windows下QT调用MinGW编译的OpenCV
开发语言·windows·qt·opencv
Debroon9 分钟前
RuleAlign 规则对齐框架：将医生的诊断规则形式化并注入模型，无需额外人工标注的自动对齐方法
人工智能
羊小猪~~16 分钟前
神经网络基础--什么是正向传播？？什么是方向传播？？
人工智能·pytorch·python·深度学习·神经网络·算法·机器学习
AI小杨17 分钟前
【车道线检测】一、传统车道线检测：基于霍夫变换的车道线检测史诗级详细教程
人工智能·opencv·计算机视觉·霍夫变换·车道线检测
晨曦_子画22 分钟前
编程语言之战：AI 之后的 Kotlin 与 Java
android·java·开发语言·人工智能·kotlin
道可云23 分钟前
道可云人工智能&元宇宙每日资讯｜2024国际虚拟现实创新大会将在青岛举办
大数据·人工智能·3d·机器人·ar·vr
人工智能培训咨询叶梓32 分钟前
探索开放资源上指令微调语言模型的现状
人工智能·语言模型·自然语言处理·性能优化·调优·大模型微调·指令微调
zzZ_CMing33 分钟前
大语言模型训练的全过程：预训练、微调、RLHF
人工智能·自然语言处理·aigc
newxtc34 分钟前
【旷视科技-注册/登录安全分析报告】
人工智能·科技·安全·ddddocr
成都古河云35 分钟前
智慧场馆：安全、节能与智能化管理的未来
大数据·运维·人工智能·安全·智慧城市
热门推荐
01【HarmonyOS】HUAWEI DevEco Studio 下载地址汇总02(欧拉)openEuler系统添加网卡文件配置流程、(欧拉)openEuler系统手动配置ipv6地址流程、(欧拉)openEuler系统网络管理说明03组基轨迹建模 GBTM的介绍与实现（Stata 或 R）04【AIGC】重塑未来的科技巨轮05全面解析：构建基于深度学习的安全帽检测系统（UI界面+YOLO代码+数据集）06【经验分享】Ubuntu22.04安装微信(linux官方版)07基于YOLOv10深度学习的CT扫描图像肾结石智能检测系统【python源码+Pyqt5界面+数据集+训练代码】深度学习实战、目标检测08Ubuntu 20.04使用Livox mid 360 测试 FAST_LIO09RAG 实践- Ollama+RagFlow 部署本地知识库10红米手机使用google play