C# CvDnn部署CoupledTPS实现旋转图像矫正

说明

效果

模型信息

项目

代码

下载

说明

TPAMI2024 - Semi-Supervised Coupled Thin-Plate Spline Model for Rotation Correction and Beyond

github地址：https://github.com/nie-lang/CoupledTPS

代码实现参考：https://github.com/hpc203/CoupledTPS-opencv-dnn

效果

模型信息

feature_extractor.onnx

Model Properties
-------------------------
---------------------------------------------------------------

Inputs
-------------------------
name：input
tensor：Float[1, 3, 384, 512]
---------------------------------------------------------------

Outputs
-------------------------
name：feature
tensor：Float[1, 256, 24, 32]
---------------------------------------------------------------

regressnet.onnx

Model Properties
-------------------------
---------------------------------------------------------------

Inputs
-------------------------
name：feature
tensor：Float[1, 256, 24, 32]
---------------------------------------------------------------

Outputs
-------------------------
name：mesh_motion
tensor：Float[1, 7, 9, 2]
---------------------------------------------------------------

项目

代码

Form1.cs

using OpenCvSharp;
using System;
using System.Drawing;
using System.Drawing.Imaging;
using System.Windows.Forms;

namespace Onnx_Demo
{
public partial class Form1 : Form
{
public Form1()
{
InitializeComponent();
}

string fileFilter = "*.*|*.bmp;*.jpg;*.jpeg;*.tiff;*.tiff;*.png";
string image_path = "";
DateTime dt1 = DateTime.Now;
DateTime dt2 = DateTime.Now;
Mat image;

CoupledTPS_RotationNet rotationNet;
int iter_num = 3;

private void button1_Click(object sender, EventArgs e)
{
OpenFileDialog ofd = new OpenFileDialog();
ofd.InitialDirectory =Application.StartupPath+"\\test_img\\";
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;
}
button2.Enabled = false;
pictureBox2.Image = null;
textBox1.Text = "";
Application.DoEvents();
//读图片
image = new Mat(image_path);
dt1 = DateTime.Now;
Mat result_image = rotationNet.detect(image, iter_num);
dt2 = DateTime.Now;
Cv2.CvtColor(result_image, result_image, ColorConversionCodes.BGR2RGB);
pictureBox2.Image = new Bitmap(result_image.ToMemoryStream());
textBox1.Text = "推理耗时:" + (dt2 - dt1).TotalMilliseconds + "ms";
button2.Enabled = true;
}

private void Form1_Load(object sender, EventArgs e)
{
rotationNet = new CoupledTPS_RotationNet("model/feature_extractor.onnx", "model/regressnet.onnx");
image_path = "test_img/00150_-8.4.jpg";
pictureBox1.Image = new Bitmap(image_path);
image = new Mat(image_path);
}

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

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

SaveFileDialog sdf = new SaveFileDialog();
private void button3_Click(object sender, EventArgs e)
{
if (pictureBox2.Image == null)
{
return;
}
Bitmap output = new Bitmap(pictureBox2.Image);
sdf.Title = "保存";
sdf.Filter = "Images (*.jpg)|*.jpg|Images (*.png)|*.png|Images (*.bmp)|*.bmp|Images (*.emf)|*.emf|Images (*.exif)|*.exif|Images (*.gif)|*.gif|Images (*.ico)|*.ico|Images (*.tiff)|*.tiff|Images (*.wmf)|*.wmf";
if (sdf.ShowDialog() == DialogResult.OK)
{
switch (sdf.FilterIndex)
{
case 1:
{
output.Save(sdf.FileName, ImageFormat.Jpeg);
break;
}
case 2:
{
output.Save(sdf.FileName, ImageFormat.Png);
break;
}
case 3:
{
output.Save(sdf.FileName, ImageFormat.Bmp);
break;
}
case 4:
{
output.Save(sdf.FileName, ImageFormat.Emf);
break;
}
case 5:
{
output.Save(sdf.FileName, ImageFormat.Exif);
break;
}
case 6:
{
output.Save(sdf.FileName, ImageFormat.Gif);
break;
}
case 7:
{
output.Save(sdf.FileName, ImageFormat.Icon);
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);
}
}
}
}

using OpenCvSharp;
using System;
using System.Drawing;
using System.Drawing.Imaging;
using System.Windows.Forms;namespace Onnx_Demo
{public partial class Form1 : Form{public Form1(){InitializeComponent();}string fileFilter = "*.*|*.bmp;*.jpg;*.jpeg;*.tiff;*.tiff;*.png";string image_path = "";DateTime dt1 = DateTime.Now;DateTime dt2 = DateTime.Now;Mat image;CoupledTPS_RotationNet rotationNet;int iter_num = 3;private void button1_Click(object sender, EventArgs e){OpenFileDialog ofd = new OpenFileDialog();ofd.InitialDirectory =Application.StartupPath+"\\test_img\\";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;}button2.Enabled = false;pictureBox2.Image = null;textBox1.Text = "";Application.DoEvents();//读图片image = new Mat(image_path);dt1 = DateTime.Now;Mat result_image = rotationNet.detect(image, iter_num);dt2 = DateTime.Now;Cv2.CvtColor(result_image, result_image, ColorConversionCodes.BGR2RGB);pictureBox2.Image = new Bitmap(result_image.ToMemoryStream());textBox1.Text = "推理耗时:" + (dt2 - dt1).TotalMilliseconds + "ms";button2.Enabled = true;}private void Form1_Load(object sender, EventArgs e){rotationNet = new CoupledTPS_RotationNet("model/feature_extractor.onnx", "model/regressnet.onnx");image_path = "test_img/00150_-8.4.jpg";pictureBox1.Image = new Bitmap(image_path);image = new Mat(image_path);}private void pictureBox1_DoubleClick(object sender, EventArgs e){Common.ShowNormalImg(pictureBox1.Image);}private void pictureBox2_DoubleClick(object sender, EventArgs e){Common.ShowNormalImg(pictureBox2.Image);}SaveFileDialog sdf = new SaveFileDialog();private void button3_Click(object sender, EventArgs e){if (pictureBox2.Image == null){return;}Bitmap output = new Bitmap(pictureBox2.Image);sdf.Title = "保存";sdf.Filter = "Images (*.jpg)|*.jpg|Images (*.png)|*.png|Images (*.bmp)|*.bmp|Images (*.emf)|*.emf|Images (*.exif)|*.exif|Images (*.gif)|*.gif|Images (*.ico)|*.ico|Images (*.tiff)|*.tiff|Images (*.wmf)|*.wmf";if (sdf.ShowDialog() == DialogResult.OK){switch (sdf.FilterIndex){case 1:{output.Save(sdf.FileName, ImageFormat.Jpeg);break;}case 2:{output.Save(sdf.FileName, ImageFormat.Png);break;}case 3:{output.Save(sdf.FileName, ImageFormat.Bmp);break;}case 4:{output.Save(sdf.FileName, ImageFormat.Emf);break;}case 5:{output.Save(sdf.FileName, ImageFormat.Exif);break;}case 6:{output.Save(sdf.FileName, ImageFormat.Gif);break;}case 7:{output.Save(sdf.FileName, ImageFormat.Icon);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);}}}
}

CoupledTPS_RotationNet.cs

using OpenCvSharp;
using OpenCvSharp.Dnn;
using System.Collections.Generic;
using System.Linq;namespace Onnx_Demo
{public class CoupledTPS_RotationNet{int input_height = 384;int input_width = 512;int grid_h = 6;int grid_w = 8;Mat grid = new Mat();Mat W_inv = new Mat();Net feature_extractor;Net regressNet;public CoupledTPS_RotationNet(string modelpatha, string modelpathb){feature_extractor = CvDnn.ReadNet(modelpatha);regressNet = CvDnn.ReadNet(modelpathb);tps2flow.get_norm_rigid_mesh_inv_grid(ref grid, ref W_inv, input_height, input_width, grid_h, grid_w);}unsafe public Mat detect(Mat srcimg, int iter_num){Mat img = new Mat();Cv2.Resize(srcimg, img, new Size(input_width, input_height));img.ConvertTo(img, MatType.CV_32FC3, 1.0 / 127.5d, -1.0d);Mat input_tensor = CvDnn.BlobFromImage(img);feature_extractor.SetInput(input_tensor);Mat[] feature_oris = new Mat[1] { new Mat() };string[] outBlobNames = feature_extractor.GetUnconnectedOutLayersNames().ToArray();feature_extractor.Forward(feature_oris, outBlobNames);Mat feature = feature_oris[0].Clone();int[] shape = { 1, 2, input_height, input_width };Mat flow = Mat.Zeros(MatType.CV_32FC1, shape);List<Mat> flow_list = new List<Mat>();for (int i = 0; i < iter_num; i++){regressNet.SetInput(feature);Mat[] mesh_motions = new Mat[1] { new Mat() };regressNet.Forward(mesh_motions, regressNet.GetUnconnectedOutLayersNames().ToArray());float* offset = (float*)mesh_motions[0].Data;Mat tp = new Mat();tps2flow.get_ori_rigid_mesh_tp(ref tp, offset, input_height, input_width, grid_h, grid_w);Mat T = W_inv * tp;   //_solve_systemT = T.T();    //舍弃batchsizeMat T_g = T * grid;Mat delta_flow = new Mat();tps2flow._transform(T_g, grid, input_height, input_width, ref delta_flow);if (i == 0){flow += delta_flow;}else{Mat warped_flow = new Mat();grid_sample.warp_with_flow(flow, delta_flow, ref warped_flow);flow = delta_flow + warped_flow;}flow_list.Add(flow.Clone());if (i < (iter_num - 1)){int fea_h = feature.Size(2);int fea_w = feature.Size(3);float scale_h = (float)fea_h / flow.Size(2);float scale_w = (float)fea_w / flow.Size(3);Mat down_flow = new Mat();upsample.UpSamplingBilinear(flow, ref down_flow, fea_h, fea_w, true, scale_h, scale_w);for (int h = 0; h < fea_h; h++){for (int w = 0; w < fea_w; w++){float* p_w = (float*)down_flow.Ptr(0, 0, h);float temp_w = p_w[w];temp_w = temp_w * scale_w;p_w[w] = temp_w;float* p_h = (float*)down_flow.Ptr(0, 1, h);float temp_h = p_h[w];temp_h = temp_h * scale_h;p_h[w] = temp_h;}}feature.Release();feature = new Mat();grid_sample.warp_with_flow(feature_oris[0], down_flow, ref feature);}}Mat correction_final = new Mat();grid_sample.warp_with_flow(input_tensor, flow_list[iter_num - 1], ref correction_final);Mat correction_img = grid_sample.convert4dtoimage(correction_final);return correction_img;}}
}