一.看成果1.上位机的基建底座-黑白主题多语言2.结果图全屏显示3.训练中心4.运行设置二.重要的事-点赞的朋友早日开上奔驰三.YoloV26–YoloDotNet Source GuideStart HereUse the local clone as the source of truth:E:/HWork/1YOLO/Code/Office/YoloDotNetThis snapshot is fromhttps://github.com/NickSwardh/YoloDotNet.gitat commitc5845b1c51690e9c734453ef7428b54b7f22f636dated2026-02-08.Readreferences/source-map.mdwhen the task needs detailed path routing.WorkflowIdentify whether the task is public API, execution provider, model metadata, preprocessing, postprocessing, a versioned task module, video, tracking, or tests.Start withYoloDotNet/Yolo.csfor the public API.InspectYoloDotNet/Core/ModuleFactory.csto see whichModelVersionandModelTypecombinations are supported.InspectYoloDotNet/Core/YoloCore.csfor preprocessing, buffer pooling, ONNX execution, image-size/gain math, ROI transforms, NMS, and result conversion.For parser changes, inspect the matchingYoloDotNet/Modules/version/task.csfile and the model/result classes underYoloDotNet/Models/.For runtime provider changes, inspectYoloDotNet.ExecutionProvider.*projects andIExecutionProvider.For validation, use thetest/projects and the bundled ONNX/image assets.Change GuidanceKeepYolo,ModuleFactory,YoloCore, module implementations, and tests aligned. A new model version/type is incomplete until the factory can route it and tests cover it.Preserve allocation-conscious patterns:ArrayPool, pinned buffers, spans,SKImage/SKBitmapdisposal, and provider disposal matter in this repo.Treat ONNX metadata and output tensor shapes as contracts. Parser logic belongs in versioned modules; shared math belongs inYoloCoreor extensions.Respect resize mode differences.ImageResize.ProportionalandImageResize.Stretchuse different gain math.Keep CPU/CUDA/DirectML/OpenVINO/CoreML provider projects thin; provider-specific session setup should still satisfyIExecutionProvider.When adding ROI, tracking, or video changes, verify coordinate transforms and frame lifecycle.ValidationUse focused .NET commands:cdE:/HWork/1YOLO/Code/Office/YoloDotNet dotnet build YoloDotNet.sln dotnettesttest/YoloDotNet.Tests/YoloDotNet.Tests.csprojFor parser or performance-sensitive changes, run the matching test class undertest/YoloDotNet.Tests/and consider benchmark smoke tests undertest/YoloDotNet.Benchmarks/.Some execution-provider tests require platform-specific runtimes, GPU drivers, or native dependencies. If unavailable, run CPU-focused tests and state what provider was not exercised.Quick API Referenceusing SkiaSharp;using YoloDotNet;using YoloDotNet.Extensions;using YoloDotNet.ExecutionProvider.Cpu;using YoloDotNet.Models;// Initialize(pick ONE execution provider)using var yolonew Yolo(new YoloOptions{ExecutionProvidernew CpuExecutionProvider(model.onnx), ImageResizeImageResize.Proportional, // or .Stretch});// Object Detection using var imageSKBitmap.Decode(image.jpg);var detectionsyolo.RunObjectDetection(image, confidence:0.25, iou:0.7);image.Draw(detections);image.Save(result.jpg);// Segmentation var segmentsyolo.RunSegmentation(image, confidence:0.25, iou:0.7);image.Draw(segments);// Pose Estimation var posesyolo.RunPoseEstimation(image, confidence:0.25, iou:0.7);image.Draw(poses);// OBB Detection var obbsyolo.RunOBBDetection(image, confidence:0.25, iou:0.7);image.Draw(obbs);// Classification var classesyolo.RunClassification(image, topClasses:5);// Video Processing using var yolonew Yolo(new YoloOptions{...});yolo.RunVideo(input.mp4,output.mp4, confidence:0.25);// Tracking(on video)yolo.RunTracking(input.mp4,output.mp4, confidence:0.25);NuGet Packagesdotnetaddpackage YoloDotNet# Pick ONE execution provider:dotnetaddpackage YoloDotNet.ExecutionProvider.Cpu# Windows/Linux/macOS CPUdotnetaddpackage YoloDotNet.ExecutionProvider.Cuda# NVIDIA GPU TensorRTdotnetaddpackage YoloDotNet.ExecutionProvider.DirectML# Windows GPU (DirectX)dotnetaddpackage YoloDotNet.ExecutionProvider.OpenVino# Inteldotnetaddpackage YoloDotNet.ExecutionProvider.CoreML# macOSONNX Export for YoloDotNet# YOLOv5u through YOLOv12: opset17yoloexportmodelbest.ptformatonnximgsz640opset17simplifyTrue# YOLO26: opset18yoloexportmodelbest.ptformatonnximgsz640opset18simplifyTrueImageResize Modes|Mode|Behavior|When to Use|| — | — | — ||Proportional|Preserves aspect ratio, letterbox pads|Match training behavior, better accuracy||Stretch|Stretches to exact imgsz×imgsz|When input is already square, speed priority|Referencesreferences/source-map.md: source layout, module/provider map, parser routes, tests, and common patch paths.四.完结
YoloV26车牌识别实战
一.看成果1.上位机的基建底座-黑白主题多语言2.结果图全屏显示3.训练中心4.运行设置二.重要的事-点赞的朋友早日开上奔驰三.YoloV26–YoloDotNet Source GuideStart HereUse the local clone as the source of truth:E:/HWork/1YOLO/Code/Office/YoloDotNetThis snapshot is fromhttps://github.com/NickSwardh/YoloDotNet.gitat commitc5845b1c51690e9c734453ef7428b54b7f22f636dated2026-02-08.Readreferences/source-map.mdwhen the task needs detailed path routing.WorkflowIdentify whether the task is public API, execution provider, model metadata, preprocessing, postprocessing, a versioned task module, video, tracking, or tests.Start withYoloDotNet/Yolo.csfor the public API.InspectYoloDotNet/Core/ModuleFactory.csto see whichModelVersionandModelTypecombinations are supported.InspectYoloDotNet/Core/YoloCore.csfor preprocessing, buffer pooling, ONNX execution, image-size/gain math, ROI transforms, NMS, and result conversion.For parser changes, inspect the matchingYoloDotNet/Modules/version/task.csfile and the model/result classes underYoloDotNet/Models/.For runtime provider changes, inspectYoloDotNet.ExecutionProvider.*projects andIExecutionProvider.For validation, use thetest/projects and the bundled ONNX/image assets.Change GuidanceKeepYolo,ModuleFactory,YoloCore, module implementations, and tests aligned. A new model version/type is incomplete until the factory can route it and tests cover it.Preserve allocation-conscious patterns:ArrayPool, pinned buffers, spans,SKImage/SKBitmapdisposal, and provider disposal matter in this repo.Treat ONNX metadata and output tensor shapes as contracts. Parser logic belongs in versioned modules; shared math belongs inYoloCoreor extensions.Respect resize mode differences.ImageResize.ProportionalandImageResize.Stretchuse different gain math.Keep CPU/CUDA/DirectML/OpenVINO/CoreML provider projects thin; provider-specific session setup should still satisfyIExecutionProvider.When adding ROI, tracking, or video changes, verify coordinate transforms and frame lifecycle.ValidationUse focused .NET commands:cdE:/HWork/1YOLO/Code/Office/YoloDotNet dotnet build YoloDotNet.sln dotnettesttest/YoloDotNet.Tests/YoloDotNet.Tests.csprojFor parser or performance-sensitive changes, run the matching test class undertest/YoloDotNet.Tests/and consider benchmark smoke tests undertest/YoloDotNet.Benchmarks/.Some execution-provider tests require platform-specific runtimes, GPU drivers, or native dependencies. If unavailable, run CPU-focused tests and state what provider was not exercised.Quick API Referenceusing SkiaSharp;using YoloDotNet;using YoloDotNet.Extensions;using YoloDotNet.ExecutionProvider.Cpu;using YoloDotNet.Models;// Initialize(pick ONE execution provider)using var yolonew Yolo(new YoloOptions{ExecutionProvidernew CpuExecutionProvider(model.onnx), ImageResizeImageResize.Proportional, // or .Stretch});// Object Detection using var imageSKBitmap.Decode(image.jpg);var detectionsyolo.RunObjectDetection(image, confidence:0.25, iou:0.7);image.Draw(detections);image.Save(result.jpg);// Segmentation var segmentsyolo.RunSegmentation(image, confidence:0.25, iou:0.7);image.Draw(segments);// Pose Estimation var posesyolo.RunPoseEstimation(image, confidence:0.25, iou:0.7);image.Draw(poses);// OBB Detection var obbsyolo.RunOBBDetection(image, confidence:0.25, iou:0.7);image.Draw(obbs);// Classification var classesyolo.RunClassification(image, topClasses:5);// Video Processing using var yolonew Yolo(new YoloOptions{...});yolo.RunVideo(input.mp4,output.mp4, confidence:0.25);// Tracking(on video)yolo.RunTracking(input.mp4,output.mp4, confidence:0.25);NuGet Packagesdotnetaddpackage YoloDotNet# Pick ONE execution provider:dotnetaddpackage YoloDotNet.ExecutionProvider.Cpu# Windows/Linux/macOS CPUdotnetaddpackage YoloDotNet.ExecutionProvider.Cuda# NVIDIA GPU TensorRTdotnetaddpackage YoloDotNet.ExecutionProvider.DirectML# Windows GPU (DirectX)dotnetaddpackage YoloDotNet.ExecutionProvider.OpenVino# Inteldotnetaddpackage YoloDotNet.ExecutionProvider.CoreML# macOSONNX Export for YoloDotNet# YOLOv5u through YOLOv12: opset17yoloexportmodelbest.ptformatonnximgsz640opset17simplifyTrue# YOLO26: opset18yoloexportmodelbest.ptformatonnximgsz640opset18simplifyTrueImageResize Modes|Mode|Behavior|When to Use|| — | — | — ||Proportional|Preserves aspect ratio, letterbox pads|Match training behavior, better accuracy||Stretch|Stretches to exact imgsz×imgsz|When input is already square, speed priority|Referencesreferences/source-map.md: source layout, module/provider map, parser routes, tests, and common patch paths.四.完结
