效果(u2net.onnx)

效果(u2net_human_seg.onnx)

项目

代码

using Microsoft.ML.OnnxRuntime;
using Microsoft.ML.OnnxRuntime.Tensors;
using OpenCvSharp;
using System;
using System.Collections.Generic;
using System.Drawing;
using System.Drawing.Imaging;
using System.Linq;
using System.Threading.Tasks;
using System.Windows.Forms;namespace U2Net
{public partial class frmMain : Form{public frmMain(){InitializeComponent();}string fileFilter = "*.*|*.bmp;*.jpg;*.jpeg;*.tiff;*.tiff;*.png";string image_path = "";string startupPath;DateTime dt1 = DateTime.Now;DateTime dt2 = DateTime.Now;string model_path;Mat image;Mat result_image;int modelSize = 512;SessionOptions options;InferenceSession onnx_session;Tensor<float> input_tensor;List<NamedOnnxValue> input_ontainer;IDisposableReadOnlyCollection<DisposableNamedOnnxValue> result_infer;DisposableNamedOnnxValue[] results_onnxvalue;Tensor<float> result_tensors;float[] result_array;private void button1_Click(object sender, EventArgs e){OpenFileDialog ofd = new OpenFileDialog();ofd.Filter = fileFilter;if (ofd.ShowDialog() != DialogResult.OK) return;pictureBox1.Image = null;image_path = ofd.FileName;pictureBox1.Image = new Bitmap(image_path);textBox1.Text = "";image = new Mat(image_path);pictureBox2.Image = null;}private void button2_Click(object sender, EventArgs e){if (image_path == ""){return;}textBox1.Text = "";pictureBox2.Image = null;int oldwidth = image.Cols;int oldheight = image.Rows;//缩放图片大小int maxEdge = Math.Max(image.Rows, image.Cols);float ratio = 1.0f * modelSize / maxEdge;int newHeight = (int)(image.Rows * ratio);int newWidth = (int)(image.Cols * ratio);Mat resize_image = image.Resize(new OpenCvSharp.Size(newWidth, newHeight));int width = resize_image.Cols;int height = resize_image.Rows;if (width != modelSize || height != modelSize){resize_image = resize_image.CopyMakeBorder(0, modelSize - newHeight, 0, modelSize - newWidth, BorderTypes.Constant, new Scalar(255, 255, 255));}Cv2.CvtColor(resize_image, resize_image, ColorConversionCodes.BGR2RGB);for (int y = 0; y < resize_image.Height; y++){for (int x = 0; x < resize_image.Width; x++){input_tensor[0, 0, y, x] = (resize_image.At<Vec3b>(y, x)[0] / 255f - 0.485f) / 0.229f;input_tensor[0, 1, y, x] = (resize_image.At<Vec3b>(y, x)[1] / 255f - 0.456f) / 0.224f;input_tensor[0, 2, y, x] = (resize_image.At<Vec3b>(y, x)[2] / 255f - 0.406f) / 0.225f;}}//将 input_tensor 放入一个输入参数的容器，并指定名称input_ontainer.Add(NamedOnnxValue.CreateFromTensor("input_image", 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();//读取第一个节点输出并转为Tensor数据result_tensors = results_onnxvalue[0].AsTensor<float>();result_array = result_tensors.ToArray();//黑白色反转//for (int i = 0; i < result_array.Length; i++)//{//    result_array[i] = 1 - result_array[i];//}float maxVal = result_array.Max();float minVal = result_array.Min();for (int i = 0; i < result_array.Length; i++){result_array[i] = (result_array[i] - minVal) / (maxVal - minVal) * 255;}result_image = new Mat(modelSize, modelSize, MatType.CV_32F, result_array);Cv2.CvtColor(result_image, result_image, ColorConversionCodes.RGB2BGR);//还原图像大小if (width != modelSize || height != modelSize){Rect rect = new Rect(0, 0, width, height);result_image = result_image.Clone(rect);}result_image = result_image.Resize(new OpenCvSharp.Size(oldwidth, oldheight));pictureBox2.Image = new Bitmap(result_image.ToMemoryStream());textBox1.Text = "推理耗时:" + (dt2 - dt1).TotalMilliseconds + "ms";}private void Form1_Load(object sender, EventArgs e){startupPath = Application.StartupPath;//model_path = startupPath + "\\model\\u2net.onnx";//model_path = startupPath + "\\model\\u2netp.onnx";model_path = startupPath + "\\model\\u2net_human_seg.onnx";modelSize = 320;//创建输出会话，用于输出模型读取信息options = new SessionOptions();options.LogSeverityLevel = OrtLoggingLevel.ORT_LOGGING_LEVEL_INFO;//设置为CPU上运行options.AppendExecutionProvider_CPU(0);//创建推理模型类，读取本地模型文件onnx_session = new InferenceSession(model_path, options);//创建输入容器input_ontainer = new List<NamedOnnxValue>();//输入Tensorinput_tensor = new DenseTensor<float>(new[] { 1, 3, modelSize, modelSize });}private void button3_Click(object sender, EventArgs e){if (pictureBox2.Image == null){return;}Bitmap output = new Bitmap(pictureBox2.Image);var sdf = new SaveFileDialog();sdf.Title = "保存";sdf.Filter = "Images (*.bmp)|*.bmp|Images (*.emf)|*.emf|Images (*.exif)|*.exif|Images (*.gif)|*.gif|Images (*.ico)|*.ico|Images (*.jpg)|*.jpg|Images (*.png)|*.png|Images (*.tiff)|*.tiff|Images (*.wmf)|*.wmf";if (sdf.ShowDialog() == DialogResult.OK){switch (sdf.FilterIndex){case 1:{output.Save(sdf.FileName, ImageFormat.Bmp);break;}case 2:{output.Save(sdf.FileName, ImageFormat.Emf);break;}case 3:{output.Save(sdf.FileName, ImageFormat.Exif);break;}case 4:{output.Save(sdf.FileName, ImageFormat.Gif);break;}case 5:{output.Save(sdf.FileName, ImageFormat.Icon);break;}case 6:{output.Save(sdf.FileName, ImageFormat.Jpeg);break;}case 7:{output.Save(sdf.FileName, ImageFormat.Png);break;}case 8:{output.Save(sdf.FileName, ImageFormat.Tiff);break;}case 9:{output.Save(sdf.FileName, ImageFormat.Wmf);break;}}MessageBox.Show("保存成功，位置：" + sdf.FileName);}}}
}

参考

GitHub - xuebinqin/U-2-Net: The code for our newly accepted paper in Pattern Recognition 2020: "U^2-Net: Going Deeper with Nested U-Structure for Salient Object Detection."

下载

源码下载