矩视智能AI视觉软件-多模型单线程调用（C++）

ZHONG JINYU

已于 2023-11-22 11:07:49 修改

阅读量148

点赞数

文章标签：人工智能 c++ 开发语言深度学习计算机视觉低代码机器学习

于 2023-10-16 19:04:01 首次发布

本文链接：https://blog.csdn.net/zjy876388/article/details/133865889

版权

SDK在单线程中加载多个模型以进行预测，逐个加载这些模型，并依次调用每个模型的预测接口，并将串行输出结果。

源码

using System;
using System.Collections.Generic;
using OpenCvSharp;
using System.IO;
using NeuroDetection;
using Newtonsoft.Json.Linq;
using System.Linq;

namespace NeuroDetection
{
class Program
{
// Return all image file names in the path <dirPath>.
// Inputs: dirPath: the path of folder containing all test images
// extension: whether the image name has an extension or not
// Modify: None
// Output: A list of strings (image file names)
static List<string> GetFilesName(string dirPath, bool extension = true)
{
List<string> fileNames = new List<string>();
try
{
string[] files = Directory.GetFiles(dirPath);

foreach (string file in files)
{
string fileName = Path.GetFileName(file);

if (extension)
{
fileNames.Add(fileName);
}
else
{
string fileNameWithoutExtension = Path.GetFileNameWithoutExtension(file);
fileNames.Add(fileNameWithoutExtension);
}
}
}
catch (Exception ex)
{
Console.WriteLine("An error occurred while reading the directory: " + ex.Message);
return new List<string>();
}

return fileNames;
}

// Draw a mask on the <image>.
// Inputs: image: the test image
// res: a single detection result
// Modify: image: add a mask into image.
// Output: None
static void DrawMask(ref Mat image, DetectionResult res)
{
// create a mask
Mat mask = res.mask;
int x0 = Math.Max(Convert.ToInt32(Math.Floor(res.box.x0)) - 1, 0);
int y0 = Math.Max(Convert.ToInt32(Math.Floor(res.box.y0)) - 1, 0);
Rect roi = new Rect(x0, y0, res.mat_mask_width, res.mat_mask_height);

// draw the mask on image
Mat[] channels = new Mat[3];
Cv2.Split(image, out channels);
Cv2.BitwiseOr(mask, channels[0][roi], channels[0][roi]);
Cv2.Merge(channels, image);
}

// Draw a box on the <image>.
// Inputs: image: the test image
// res: a single detection result
// Modify: image: add a box into image.
// Output: None
static void DrawBox(ref Mat image, DetectionResult res)
{
Point p1 = new Point(res.box.x0, res.box.y0);
Point p2 = new Point(res.box.x1, res.box.y1);
// white. TODO: change it if you want
Scalar color = new Scalar(0, 255, 0);
Cv2.Rectangle(image, p1, p2, color, 2);
}

// Print information on the <image>.
// Inputs: image: the test image
// res: a single detection result
// Modify: image: add information (labels, label indices, and confidence scores) into image.
// Output: None
static void PrintInfo(ref Mat image, DetectionResult res)
{
string tagLabel = "Label: " + res.label;
string tagLabelIndex = "Label index: " + res.label_index;
string tagScore = "Score: " + res.score;

// TODO: change following values if you need to display texts clearly on your test images.
// size of characters. TODO: change it if you want
double fontScale = 3;
// thickness of characters. TODO: change it if you want
int thickness = 4;
// white. TODO: change it if you want
Scalar color = new Scalar(255, 255, 255);
// font. TODO: change it if you want
HersheyFonts font = HersheyFonts.HersheyPlain;
// the position where to print tagLabel. TODO: change it if you want
Point up = new Point(res.box.x0, res.box.y0);
// the position where to print tagLabelIndex. TODO: change it if you want
Point center = new Point(res.box.x0, (res.box.y0 + res.box.y1) / 2);
// the position where to print tagScore. TODO: change it if you want
Point bottom = new Point(res.box.x0, res.box.y1);
Cv2.PutText(image, tagLabel, up, font, fontScale, color, thickness);
Cv2.PutText(image, tagLabelIndex, center, font, fontScale, color, thickness);
Cv2.PutText(image, tagScore, bottom, font, fontScale, color, thickness);
}

// Display test images (with)
// Inputs: image: the test image
// res: all detection results on <image>
// Modify: Display <image> and all detection results (masks)
// Output: None
static void DisplayResults(ref Mat image, DetectionResult[] res)
{
Cv2.NamedWindow("Image", WindowMode.FreeRatio);
// TODO: you can change the size of the display window
// Cv2.ResizeWindow("Image", 800, 1200);

for (int i = 0; i < res.Length; ++i)
{
// OCR or Detection
if (res[i].mask == null)
{
DrawBox(ref image, res[i]);
}
// Segmentation
else
{
DrawMask(ref image, res[i]);
}

// print labels, label indices, and confidence scores on the image
// TODO: you can comment it to not print info on test images
PrintInfo(ref image, res[i]);
}

Cv2.ImShow("Image", image);
Cv2.WaitKey(0);
Cv2.DestroyWindow("Image");
}

static int predict(SDK sdk, string modelName, string beforePath, List<string> fileNames, int iFile)
{

string iFileName = beforePath + "\\" + fileNames[iFile];
Console.WriteLine("Reading the image: " + iFileName);

Mat image = new Mat(iFileName);
if (image.Empty())
{
Console.WriteLine("Warning: The image is empty. File name: " + iFileName);
return -1;
}

Mat[] mats = new Mat[] { image };
Console.WriteLine("Image size (height, width)： " + image.Height + " " + image.Width);

PredictionResult res = sdk.PredictModel(modelName, mats);

if (res.status != 0)
{
Console.WriteLine("ERROR: Fail to predict. Status code: " + res.status);
return -1;
}

DetectionResult[][] outResults = res.outResults;

for (int i = 0; i < outResults.Length; ++i)
{
if (outResults[i] == null)
{
Console.WriteLine("Waring: this image has no prediction results!");
continue;
}

Console.WriteLine();
for (int j = 0; j < outResults[i].Length; ++j)
{
Console.WriteLine("label: " + outResults[i][j].label);
Console.WriteLine("label_index: " + outResults[i][j].label_index);
Console.WriteLine("confidential score: " + outResults[i][j].score);
Console.WriteLine("position of result(x0,y0,x1,y1): " + "(" + outResults[i][j].box.x0 + ", " + outResults[i][j].box.y0 + ", " + outResults[i][j].box.x1 + ", " + outResults[i][j].box.y1 + ")");
Console.WriteLine("row_index: " + outResults[i][j].row_index);
Console.WriteLine("col_index: " + outResults[i][j].col_index);
Console.WriteLine("mask_width: " + outResults[i][j].mask_width);
Console.WriteLine("mask_height: " + outResults[i][j].mask_height);
Console.WriteLine("mat_mask_width: " + outResults[i][j].mat_mask_width);
Console.WriteLine("mat_mask_height: " + outResults[i][j].mat_mask_height);
Console.WriteLine();
Console.WriteLine();
}

DisplayResults(ref image, outResults[i]);
}
Console.WriteLine();
Console.WriteLine();
return 1;
}

static void Main(string[] args)
{
SDK sdk = new SDK();

string model_path_1 = "C:\\Users\\NeuroBot\\Desktop\\release 6.2\\cpu_paper_model";
string model_path_2 = "C:\\Users\\NeuroBot\\Desktop\\release 6.2\\cpu_kuaisu_model";
string file_path_1 = "C:\\Users\\NeuroBot\\Desktop\\release 6.2\\cpu_paper_pic";
string file_path_2 = "C:\\Users\\NeuroBot\\Desktop\\release 6.2\\cpu_kuaisu_pic";
string modelName_1 = "1";
string modelName_2 = "2";
string device_name_1 = "cpu";
string device_name_2 = "cpu";

int status_1 = sdk.LoadModel(modelName_1, model_path_1, device_name_1);
if (status_1 != 0)
{
Console.WriteLine("ERROR: Fail to load the model. Status code: " + status_1);
return;
}

int status_2 = sdk.LoadModel(modelName_2, model_path_2, device_name_2);
if (status_2 != 0)
{
Console.WriteLine("ERROR: Fail to load the model. Status code: " + status_2);
return;
}

// TODO: change to your test image folder path!
Console.WriteLine("Image folder path: " + file_path_1);

List<string> fileNames_1 = GetFilesName(file_path_1, extension: true);
List<string> fileNames_2 = GetFilesName(file_path_2, extension: true);
int numFiles_1 = fileNames_1.Count;
int numFiles_2 = fileNames_2.Count;
if (numFiles_1 == 0 && numFiles_2 == 0)
{
Console.WriteLine("Error: This image folder is empty!");
return;
}

int i = 0, j = 0, count = 0;
while (i < numFiles_1 && j < numFiles_2)
{
if (count % 2 == 0)
{
predict(sdk, modelName_1, file_path_1, fileNames_1, i);
++i;
}
else
{
predict(sdk, modelName_2, file_path_2, fileNames_2, j);
++j;
}
count++;
}

while (i < numFiles_1)
{
predict(sdk, modelName_1, file_path_1, fileNames_1, i);
++i;
}
while (j < numFiles_2)
{
predict(sdk, modelName_2, file_path_2, fileNames_2, j);
++j;

}
}
}
}