介绍
github地址：
https://github.com/cfzd/Ultra-Fast-Lane-Detection-v2
Ultra Fast Deep Lane Detection With Hybrid Anchor Driven Ordinal classification (TPAMI 2022)
效果
模型信息
Inputs
-------------------------
name：input
tensor：Float[1, 3, 320, 1600]
---------------------------------------------------------------
Outputs
-------------------------
name：loc_row
tensor：Float[1, 200, 72, 4]
name：loc_col
tensor：Float[1, 100, 81, 4]
name：exist_row
tensor：Float[1, 2, 72, 4]
name：exist_col
tensor：Float[1, 2, 81, 4]
---------------------------------------------------------------
项目
VS2022
.net framework 4.8
OpenCvSharp 4.8
Microsoft.ML.OnnxRuntime 1.16.2
代码
using Microsoft.ML.OnnxRuntime;
using Microsoft.ML.OnnxRuntime.Tensors;
using OpenCvSharp;
using System;
using System.Collections.Generic;
using System.Drawing;
using System.Linq;
using System.Runtime.InteropServices;
using System.Text;
using System.Windows.Forms;
namespace Onnx_Demo
{
public partial class frmMain : Form
{
public frmMain()
{
InitializeComponent();
}
string fileFilter = "*.*|*.bmp;*.jpg;*.jpeg;*.tiff;*.tiff;*.png";
string image_path = "";
string startupPath;
string model_path;
DateTime dt1 = DateTime.Now;
DateTime dt2 = DateTime.Now;
Mat image;
Mat result_image;
SessionOptions options;
InferenceSession onnx_session;
Tensor<float> input_tensor;
List<NamedOnnxValue> input_ontainer;
IDisposableReadOnlyCollection<DisposableNamedOnnxValue> result_infer;
DisposableNamedOnnxValue[] results_onnxvalue;
StringBuilder sb = new StringBuilder();
int inpHeight = 320;
int inpWidth = 1600;
int num_row;
int num_col;
List<float> row_anchor = new List<float>();
List<float> col_anchor = new List<float>();
float crop_ratio;
string dataset;
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)
{
startupPath = Application.StartupPath + "\\model\\";
model_path = startupPath + "ufldv2_culane_res18_320x1600.onnx";
// 创建输出会话
options = new SessionOptions();
options.LogSeverityLevel = OrtLoggingLevel.ORT_LOGGING_LEVEL_INFO;
options.AppendExecutionProvider_CPU(0);// 设置为CPU上运行
// 创建推理模型类，读取本地模型文件
onnx_session = new InferenceSession(model_path, options);
if (model_path.Contains("culane"))
{
dataset = "culane";
num_row = 72;
num_col = 81;
crop_ratio = 0.6f;
}
else
{
num_row = 56;
num_col = 41;
crop_ratio = 0.8f;
}
// 创建输入容器
input_ontainer = new List<NamedOnnxValue>();
GenerateAnchor();
}
void GenerateAnchor()
{
for (int i = 0; i < num_row; i++)
{
if (dataset == "culane")
{
row_anchor.Add((float)(0.42 + i * (1.0 - 0.42) / (num_row - 1)));
}
else
{
row_anchor.Add((float)((160 + i * (710 - 160) / (num_row - 1)) / 720.0));
}
}
for (int i = 0; i < num_col; i++)
{
col_anchor.Add((float)(0.0 + i * (1.0 - 0.0) / (num_col - 1)));
}
}
private void button2_Click(object sender, EventArgs e)
{
if (image_path == "")
{
return;
}
textBox1.Text = "检测中，请稍等……";
pictureBox2.Image = null;
Application.DoEvents();
//图片
image = new Mat(image_path);
int img_h = image.Rows;
int img_w = image.Cols;
Mat resize_image = new Mat();
Cv2.Resize(image, resize_image, new OpenCvSharp.Size(inpWidth, inpHeight / crop_ratio));
Mat dstimg = Common.Normalize(resize_image,inpHeight);
var sourceData = Common.ExtractMat(dstimg);
int[] dimensions = new int[4] { 1, 3, inpHeight, inpWidth };
input_tensor = new DenseTensor<float>(sourceData, dimensions);
input_ontainer.Add(NamedOnnxValue.CreateFromTensor("input", input_tensor));
dt1 = DateTime.Now;
//运行 Inference 并获取结果
result_infer = onnx_session.Run(input_ontainer);
dt2 = DateTime.Now;
//将输出结果转为DisposableNamedOnnxValue数组
results_onnxvalue = result_infer.ToArray();
var loc_row = results_onnxvalue[0].AsTensor<float>().ToArray();
var loc_col = results_onnxvalue[1].AsTensor<float>().ToArray();
var exist_row = results_onnxvalue[2].AsTensor<float>().ToArray();
var exist_col = results_onnxvalue[3].AsTensor<float>().ToArray();
var loc_row_dims = results_onnxvalue[0].AsTensor<float>().Dimensions.ToArray();
int num_grid_row = loc_row_dims[1];
int num_cls_row = loc_row_dims[2];
int num_lane_row = loc_row_dims[3];
var loc_col_dims = results_onnxvalue[1].AsTensor<float>().Dimensions.ToArray();
int num_grid_col = loc_col_dims[1];
int num_cls_col = loc_col_dims[2];
int num_lane_col = loc_col_dims[3];
int[] exist_row_dims = results_onnxvalue[2].AsTensor<float>().Dimensions.ToArray();
int[] exist_col_dims = results_onnxvalue[3].AsTensor<float>().Dimensions.ToArray();
int[] max_indices_row = Common.argmax_1(loc_row, loc_row_dims);
int[] valid_row = Common.argmax_1(exist_row, exist_row_dims);
int[] max_indices_col = Common.argmax_1(loc_col, loc_col_dims);
int[] valid_col = Common.argmax_1(exist_col, exist_col_dims);
List<List<OpenCvSharp.Point>> line_list = new List<List<OpenCvSharp.Point>>();
List<OpenCvSharp.Point> temp = new List<OpenCvSharp.Point>();
line_list.Add(temp);
line_list.Add(temp);
line_list.Add(temp);
line_list.Add(temp);
int[] item = new int[2] { 1, 2 };
foreach (var i in item)
{
if (Common.sum_valid(valid_row, num_cls_row, num_lane_row, i) > num_cls_row * 0.5)
{
for (int k = 0; k < num_cls_row; k++)
{
int index = k * num_lane_row + i;
if (valid_row[index] != 0)
{
List<float> pred_all_list = new List<float>();
List<int> all_ind_list = new List<int>();
for (int all_ind = Math.Max(0, (int)(max_indices_row[index] - 1)); all_ind <= (Math.Min(num_grid_row - 1, max_indices_row[index]) + 1); all_ind++)
{
pred_all_list.Add(loc_row[all_ind * num_cls_row * num_lane_row + index]);
all_ind_list.Add(all_ind);
}
List<float> pred_all_list_softmax = new List<float>();
float[] pred_all_list_softmax_temp = new float[pred_all_list.Count];
Common.SoftMaxFast(pred_all_list.ToArray(), ref pred_all_list_softmax_temp, pred_all_list.Count);
pred_all_list_softmax = pred_all_list_softmax_temp.ToList();
float out_temp = 0;
for (int l = 0; l < pred_all_list.Count; l++)
{
out_temp += pred_all_list_softmax[l] * all_ind_list[l];
}
float x = (float)((out_temp + 0.5) / (num_grid_row - 1.0));
float y = row_anchor[k];
line_list[i].Add(new OpenCvSharp.Point((int)(x * img_w), (int)(y * img_h)));
}
}
}
}
item = new int[4] { 0, 1, 2, 3 };
foreach (var i in item)
{
if (Common.sum_valid(valid_col, num_cls_col, num_lane_col, i) > num_cls_col / 4)
{
for (int k = 0; k < num_cls_col; k++)
{
int index = k * num_lane_col + i;
if (valid_col[index] != 0)
{
List<float> pred_all_list = new List<float>();
List<int> all_ind_list = new List<int>();
for (int all_ind = Math.Max(0, (int)(max_indices_col[index] - 1)); all_ind <= (Math.Min(num_grid_col - 1, max_indices_col[index]) + 1); all_ind++)
{
pred_all_list.Add(loc_col[all_ind * num_cls_col * num_lane_col + index]);
all_ind_list.Add(all_ind);
}
List<float> pred_all_list_softmax = new List<float>();
float[] pred_all_list_softmax_temp = new float[pred_all_list.Count];
Common.SoftMaxFast(pred_all_list.ToArray(), ref pred_all_list_softmax_temp, pred_all_list.Count);
pred_all_list_softmax = pred_all_list_softmax_temp.ToList();
float out_temp = 0;
for (int l = 0; l < pred_all_list.Count; l++)
{
out_temp += pred_all_list_softmax[l] * all_ind_list[l];
}
float y = (float)((out_temp + 0.5) / (num_grid_col - 1.0));
float x = col_anchor[k];
line_list[i].Add(new OpenCvSharp.Point((int)(x * img_w), (int)(y * img_h)));
}
}
}
}
result_image = image.Clone();
foreach (var line in line_list)
{
foreach (var p in line)
{
Cv2.Circle(result_image, p, 3, new Scalar(0, 255, 0), -1);
}
}
sb.Clear();
sb.AppendLine("推理耗时:" + (dt2 - dt1).TotalMilliseconds + "ms");
sb.AppendLine("------------------------------");
pictureBox2.Image = new Bitmap(result_image.ToMemoryStream());
textBox1.Text = sb.ToString();
}
private void pictureBox2_DoubleClick(object sender, EventArgs e)
{
Common.ShowNormalImg(pictureBox2.Image);
}
private void pictureBox1_DoubleClick(object sender, EventArgs e)
{
Common.ShowNormalImg(pictureBox1.Image);
}
}
}
