要在C语言中使用OpenVINO™，你需要首先确保你已经安装了OpenVINO工具包，并且已经将相关的库链接到你的项目中。以下是一个简单的示例，展示如何在C语言中加载一个OpenVINO模型，并使用它进行推理。```c#include #include #
摘要：PP-YOLOE是基于PP-YOLOv2的优秀单级无锚模型，超越了各种流行的YOLO模型。PP-YOLOE有一系列型号，命名为smlx，通过宽度乘数和深度乘数进行配置。PP-YOLOE避免使用特殊的运算符，如可变形卷积或矩阵NMS，以
前言
OpenVINO™ C# API 是一个 OpenVINO™ 的 .Net wrapper，应用最新的 OpenVINO™ 库开发，通过 OpenVINO™ C API 实现 .Net 对 OpenVINO™ Runtime 调用，使用习惯与 OpenVINO™ C++ API 一致。OpenVINO™ C# API 由于是基于 OpenVINO™ 开发，所支持的平台与 OpenVINO™ 完全一致，具体信息可以参考 OpenVINO™。通过使用 OpenVINO™ C# API，可以在 .NET、.NET Framework等框架下使用 C# 语言实现深度学习模型在指定平台推理加速。
OpenVINO™ C# API 项目链接为：
https://github.com/guojin-yan/OpenVINO-CSharp-API.git
项目源码链接为：
https://github.com/guojin-yan/OpenVINO-CSharp-API-Samples.git
1. 简介
PP-YOLOE是基于PP-YOLOv2的优秀单级无锚模型，超越了各种流行的YOLO模型。PP-YOLOE有一系列型号，命名为s/m/l/x，通过宽度乘数和深度乘数进行配置。PP-YOLOE避免使用特殊的运算符，如可变形卷积或矩阵NMS，以便友好地部署在各种硬件上。 在本文中，我们将使用OpenVINO™ C# API 部署 PP-YOLOE实现物体检测。
2. 项目环境与依赖
该项目中所需依赖已经支持通过NuGet Package进行安装，在该项目中，需要安装以下NuGet Package：
OpenVINO C# API NuGet Package：
OpenVINO.CSharp.API
OpenVINO.runtime.win
OpenVINO.CSharp.API.Extensions
OpenVINO.CSharp.API.Extensions.OpenCvSharp
OpenCvSharp NuGet Package：
OpenCvSharp4
OpenCvSharp4.Extensions
OpenCvSharp4.runtime.win
3. 项目输出
项目使用的是控制台输出，运行后输出如下所示：
<00:00:00> Sending http request to https://github.com/guojin-yan/OpenVINO-CSharp-API-Samples/releases/download/Model/ppyoloe_plus_crn_l_80e_coco.tar.
<00:00:02> Http Response Accquired.
<00:00:02> Total download length is 199.68 Mb.
<00:00:02> Download Started.
<00:00:02> File created.
<00:02:03> Downloading: [■■■■■■■■■■] 100% <00:02:03 1.81 Mb/s> 199.68 Mb/199.68 Mb downloaded.
<00:02:03> File Downloaded, saved in E:\GitSpace\OpenVINO-CSharp-API-Samples\model_samples\ppyoloe\ppyoloe_opencvsharp\bin\Release\net6.0\model\ppyoloe_plus_crn_l_80e_coco.tar.
<00:00:00> Sending http request to https://github.com/guojin-yan/OpenVINO-CSharp-API-Samples/releases/download/Image/test_det_02.jpg.
<00:00:02> Http Response Accquired.
<00:00:02> Total download length is 0.16 Mb.
<00:00:02> Download Started.
<00:00:02> File created.
<00:00:02> Downloading: [■■■■■■■■■■] 100% <00:00:02 0.06 Mb/s> 0.16 Mb/0.16 Mb downloaded.
<00:00:02> File Downloaded, saved in E:\GitSpace\OpenVINO-CSharp-API-Samples\model_samples\ppyoloe\ppyoloe_opencvsharp\bin\Release\net6.0\model\test_image.jpg.
[ INFO ] Inference device: CPU
[ INFO ] Start RT-DETR model inference.
[ INFO ] 1. Initialize OpenVINO Runtime Core success, time spend: 4.5204ms.
[ INFO ] 2. Read inference model success, time spend: 228.4451ms.
[ INFO ] Inference Model
[ INFO ] Model name: Model0
[ INFO ] Input:
[ INFO ] name: scale_factor
[ INFO ] type: float
[ INFO ] shape: Shape : {?,2}
[ INFO ] name: image
[ INFO ] type: float
[ INFO ] shape: Shape : {?,3,640,640}
[ INFO ] Output:
[ INFO ] name: multiclass_nms3_0.tmp_0
[ INFO ] type: float
[ INFO ] shape: Shape : {?,6}
[ INFO ] name: multiclass_nms3_0.tmp_2
[ INFO ] type: int32_t
[ INFO ] shape: Shape : {?}
[ INFO ] 3. Loading a model to the device success, time spend:501.0716ms.
[ INFO ] 4. Create an infer request success, time spend:0.2663ms.
[ INFO ] 5. Process input images success, time spend:30.1001ms.
[ INFO ] 6. Set up input data success, time spend:2.3631ms.
[ INFO ] 7. Do inference synchronously success, time spend:286.1085ms.
[ INFO ] 8. Get infer result data success, time spend:0.5189ms.
[ INFO ] 9. Process reault success, time spend:0.4425ms.
[ INFO ] The result save to E:\GitSpace\OpenVINO-CSharp-API-Samples\model_samples\ppyoloe\ppyoloe_opencvsharp\bin\Release\net6.0\model\test_image_result.jpg
图像预测结果如下图所示：
4. 代码展示
以下为嘛中所使用的命名空间代码：
using OpenCvSharp.Dnn;
using OpenCvSharp;
using OpenVinoSharp;
using OpenVinoSharp.Extensions;
using OpenVinoSharp.Extensions.utility;
using System.Runtime.InteropServices;
using OpenVinoSharp.preprocess;
using OpenVinoSharp.Extensions.model;
using OpenVinoSharp.Extensions.result;
using OpenVinoSharp.Extensions.process;
namespace ppyoloe_opencvsharp
{
internal class Program
{
....
}
}
下面为定义的模型预测代码：
一般预测流程：
static void ppyoloe_det(string model_path, string image_path, string device)
{
// -------- Step 1. Initialize OpenVINO Runtime Core --------
Core core = new Core();
// -------- Step 2. Read inference model --------
Model model = core.read_model(model_path);
OvExtensions.printf_model_info(model);
// -------- Step 3. Loading a model to the device --------
CompiledModel compiled_model = core.compile_model(model, device);
// -------- Step 4. Create an infer request --------
InferRequest infer_request = compiled_model.create_infer_request();
// -------- Step 5. Process input images --------
Mat image = new Mat(image_path); // Read image by opencvsharp
float[] factor = new float[] { 640.0f / (float)image.Rows, 640.0f / (float)image.Cols };
float[] im_shape = new float[] { 640.0f, 640.0f };
Mat input_mat = CvDnn.BlobFromImage(image, 1.0 / 255.0, new OpenCvSharp.Size(640, 640), 0, true, false);
float[] input_data = new float[640 * 640 * 3];
Marshal.Copy(input_mat.Ptr(0), input_data, 0, input_data.Length);
// -------- Step 6. Set up input data --------
Tensor input_tensor_data = infer_request.get_tensor("image");
input_tensor_data.set_shape(new Shape(1, 3, 640, 640));
input_tensor_data.set_data<float>(input_data);
Tensor input_tensor_factor = infer_request.get_tensor("scale_factor");
input_tensor_factor.set_shape(new Shape(1, 2));
input_tensor_factor.set_data<float>(factor);
// -------- Step 7. Do inference synchronously --------
infer_request.infer();
// -------- Step 8. Get infer result data --------
Tensor output_tensor = infer_request.get_output_tensor(0);
int output_length = (int)output_tensor.get_size();
float[] output_data = output_tensor.get_data<float>(output_length);
// -------- Step 9. Process reault --------
List<Rect> position_boxes = new List<Rect>();
List<int> class_ids = new List<int>();
List<float> confidences = new List<float>();
for (int i = 0; i < 300; ++i)
{
if (output_data[6 * i + 1] > 0.5)
{
class_ids.Add((int)output_data[6 * i]);
confidences.Add(output_data[6 * i + 1]);
position_boxes.Add(new Rect((int)output_data[6 * i + 2], (int)output_data[6 * i + 3],
(int)(output_data[6 * i + 4] - output_data[6 * i + 2]),
(int)(output_data[6 * i + 5] - output_data[6 * i + 3])));
}
}
for (int index = 0; index < class_ids.Count; index++)
{
Cv2.Rectangle(image, position_boxes[index], new Scalar(0, 0, 255), 2, LineTypes.Link8);
Cv2.Rectangle(image, new OpenCvSharp.Point(position_boxes[index].TopLeft.X, position_boxes[index].TopLeft.Y + 30),
new OpenCvSharp.Point(position_boxes[index].BottomRight.X, position_boxes[index].TopLeft.Y), new Scalar(0, 255, 255), -1);
Cv2.PutText(image, class_ids[index] + "-" + confidences[index].ToString("0.00"),
new OpenCvSharp.Point(position_boxes[index].X, position_boxes[index].Y + 25),
HersheyFonts.HersheySimplex, 0.8, new Scalar(0, 0, 0), 2);
}
string output_path = Path.Combine(Path.GetDirectoryName(Path.GetFullPath(image_path)),
Path.GetFileNameWithoutExtension(image_path) + "_result.jpg");
Cv2.ImWrite(output_path, image);
Slog.INFO("The result save to " + output_path);
Cv2.ImShow("Result", image);
Cv2.WaitKey(0);
}
编译预处理步骤到模型方式推理模型：
static void ppyoloe_det_with_process(string model_path, string image_path, string device)
{
// -------- Step 1. Initialize OpenVINO Runtime Core --------
Core core = new Core();
// -------- Step 2. Read inference model --------
Model model = core.read_model(model_path);
OvExtensions.printf_model_info(model);
PrePostProcessor processor = new PrePostProcessor(model);
Tensor input_tensor_pro = new Tensor(new OvType(ElementType.U8), new Shape(1, 640, 640, 3));
InputInfo input_info = processor.input("image");
InputTensorInfo input_tensor_info = input_info.tensor();
input_tensor_info.set_from(input_tensor_pro).set_layout(new Layout("NHWC")).set_color_format(ColorFormat.BGR);
PreProcessSteps process_steps = input_info.preprocess();
process_steps.convert_color(ColorFormat.RGB).resize(ResizeAlgorithm.RESIZE_LINEAR)
.convert_element_type(new OvType(ElementType.F32)).scale(255.0f).convert_layout(new Layout("NCHW"));
Model new_model = processor.build();
// -------- Step 3. Loading a model to the device --------
CompiledModel compiled_model = core.compile_model(new_model, device);
// -------- Step 4. Create an infer request --------
InferRequest infer_request = compiled_model.create_infer_request();
// -------- Step 5. Process input images --------
Mat image = new Mat(image_path); // Read image by opencvsharp
Mat input_image = new Mat();
Cv2.Resize(image, input_image, new OpenCvSharp.Size(640, 640));
float[] factor = new float[] { 640.0f / (float)image.Rows, 640.0f / (float)image.Cols };
float[] im_shape = new float[] { 640.0f, 640.0f };
// -------- Step 6. Set up input data --------
Tensor input_tensor_data = infer_request.get_tensor("image");
byte[] input_data = new byte[3 * 640 * 640];
Marshal.Copy(input_image.Ptr(0), input_data, 0, input_data.Length);
IntPtr destination = input_tensor_data.data();
Marshal.Copy(input_data, 0, destination, input_data.Length);
Tensor input_tensor_factor = infer_request.get_tensor("scale_factor");
input_tensor_factor.set_shape(new Shape(1, 2));
input_tensor_factor.set_data<float>(factor);
// -------- Step 7. Do inference synchronously --------
infer_request.infer();
// -------- Step 8. Get infer result data --------
Tensor output_tensor = infer_request.get_output_tensor(0);
int output_length = (int)output_tensor.get_size();
float[] output_data = output_tensor.get_data<float>(output_length);
// -------- Step 9. Process reault --------
List<Rect> position_boxes = new List<Rect>();
List<int> class_ids = new List<int>();
List<float> confidences = new List<float>();
for (int i = 0; i < 300; ++i)
{
if (output_data[6 * i + 1] > 0.5)
{
class_ids.Add((int)output_data[6 * i]);
confidences.Add(output_data[6 * i + 1]);
position_boxes.Add(new Rect((int)output_data[6 * i + 2], (int)output_data[6 * i + 3],
(int)(output_data[6 * i + 4] - output_data[6 * i + 2]),
(int)(output_data[6 * i + 5] - output_data[6 * i + 3])));
}
}
for (int index = 0; index < class_ids.Count; index++)
{
Cv2.Rectangle(image, position_boxes[index], new Scalar(0, 0, 255), 2, LineTypes.Link8);
Cv2.Rectangle(image, new OpenCvSharp.Point(position_boxes[index].TopLeft.X, position_boxes[index].TopLeft.Y + 30),
new OpenCvSharp.Point(position_boxes[index].BottomRight.X, position_boxes[index].TopLeft.Y), new Scalar(0, 255, 255), -1);
Cv2.PutText(image, class_ids[index] + "-" + confidences[index].ToString("0.00"),
new OpenCvSharp.Point(position_boxes[index].X, position_boxes[index].Y + 25),
HersheyFonts.HersheySimplex, 0.8, new Scalar(0, 0, 0), 2);
}
string output_path = Path.Combine(Path.GetDirectoryName(Path.GetFullPath(image_path)),
Path.GetFileNameWithoutExtension(image_path) + "_result.jpg");
Cv2.ImWrite(output_path, image);
Slog.INFO("The result save to " + output_path);
Cv2.ImShow("Result", image);
Cv2.WaitKey(0);
}
使用封装的方法：
static void ppyoloe_det_using_extensions(string model_path, string image_path, string device)
{
PPYoloeConfig config = new PPYoloeConfig();
config.set_model(model_path);
PPYoloeDet det = new PPYoloeDet(config);
Mat image = Cv2.ImRead(image_path);
DetResult result = det.predict(image);
Mat result_im = Visualize.draw_det_result(result, image);
Cv2.ImShow("Result", result_im);
Cv2.WaitKey(0);
}
下面为程序运行的主函数代码，该代码会下载转换好的预测模型，并调用预测方法进行预测：
static void Main(string[] args)
{
string model_path = "";
string image_path = "";
string device = "CPU";
if (args.Length == 0)
{
if (!Directory.Exists("./model"))
{
Directory.CreateDirectory("./model");
}
if (!File.Exists("./model/model.pdiparams")
&& !File.Exists("./model/model.pdmodel"))
{
if (!File.Exists("./model/ppyoloe_plus_crn_l_80e_coco.tar"))
{
_ = Download.download_file_async("https://github.com/guojin-yan/OpenVINO-CSharp-API-Samples/releases/download/Model/ppyoloe_plus_crn_l_80e_coco.tar",
"./model/ppyoloe_plus_crn_l_80e_coco.tar").Result;
}
Download.unzip("./model/ppyoloe_plus_crn_l_80e_coco.tar", "./model/");
}
if (!File.Exists("./model/test_image.jpg"))
{
_ = Download.download_file_async("https://github.com/guojin-yan/OpenVINO-CSharp-API-Samples/releases/download/Image/test_det_02.jpg",
"./model/test_image.jpg").Result;
}
model_path = "./model/model.pdmodel";
image_path = "./model/test_image.jpg";
}
else if (args.Length >= 2)
{
model_path = args[0];
image_path = args[1];
device = args[2];
}
else
{
Console.WriteLine("Please enter the correct command parameters, for example:");
Console.WriteLine("> 1. dotnet run");
Console.WriteLine("> 2. dotnet run <model path> <image path> <device name>");
}
// -------- Get OpenVINO runtime version --------
OpenVinoSharp.Version version = Ov.get_openvino_version();
Slog.INFO("---- OpenVINO INFO----");
Slog.INFO("Description : " + version.description);
Slog.INFO("Build number: " + version.buildNumber);
Slog.INFO("Predict model files: " + model_path);
Slog.INFO("Predict image files: " + image_path);
Slog.INFO("Inference device: " + device);
Slog.INFO("Start RT-DETR model inference.");
ppyoloe_det(model_path, image_path, device);
//ppyoloe_det_with_process(model_path, image_path, device);
//ppyoloe_det_using_extensions(model_path, image_path, device);
}
5. 总结
在该项目中，我们结合之前开发的 OpenVINO™ C# API 项目部署PP-YOLOE模型，实现物体检测。
项目完整代码链接为：
https://github.com/guojin-yan/OpenVINO-CSharp-API-Samples/blob/master/model_samples/ppyoloe/ppyoloe_opencvsharp/Program.cs
为了方便EmguCV用户使用需求，同时开发了EmguCV版本，项目链接为：
https://github.com/guojin-yan/OpenVINO-CSharp-API-Samples/blob/master/model_samples/ppyoloe/ppyoloe_emgucv/Program.cs
最后如果各位开发者在使用中有任何问题，欢迎大家与我联系。