Framework for machine learning
专为 C# 打造的轻量高效计算机视觉库(.NET 8 / .NET Framework 4.6.1+)—— OpenCvSharp 替代方案。 单一原生 DLL(cvsharp_native.dll,C++17/AVX2/静态 CRT,约 3.4 MB),纯 P/Invoke 无 C++/CLI,可裁剪可 AOT。 PS 级能力:27 种混合模式图层合成、alpha 感知重采样、多色标渐变、软蒙版形态学、网格变形、 镜头模糊(多边形 bokeh/景深)、GrabCut 抠图、16 位 LUT/曲线/直方图/形态学/绘图,全部原生多线程。 几何类型零撞名(PointI/SizeI/RectI),PNG 写出内置 libpng Level 9+自适应、WebP 读写内置 libwebp(有损/无损/alpha/档位可调)、其余图像 IO 走 Windows WIC,原生支持中文路径, 绘图走 GDI 抗锯齿字体(PutText 支持中文)。 算子覆盖:滤波/形态学/阈值(Otsu/自适应/动态阈值/方差阈值)/几何变换/Canny/轮廓/连通域/霍夫/角点/透视变换/缺陷修复, 内置旋转模板匹配 ShapeMatcher(金字塔+多线程+模型缓存)与 CvSharp.Measure 亚像素测量算子族 (卡尺找线/找圆/对边/两线交角点,稳健拟合,典型误差小于 0.05px;宽度 MAD 稳健统计抗毛刺), 区域形状特征 GetRegionProps/SelectRegions 与粘连分割 SplitTouchingRegions、多相机拼接 Stitcher、 骨架化、凹包、任意尺寸 FFT、漫水填充、AVX2 中值滤波、平场校正(单拍 SIMD,比 OpenCV 快约 10 倍)、高反差保留。 颜色相似度:CIE76/CIEDE2000 逐像素色差图、马氏距离 ColorMatcher(训练式颜色匹配,Cognex/Keyence 同类),Lab 转换 AVX2 定点查表(与 OpenCV 同量级)。 纹理与颜色统计:LBP 三变体(Basic/Uniform/旋转不变,AVX2)、HOG 描述子(快 OpenCV 约 5.8 倍)、任意桶数颜色直方图(快约 3.4 倍)、median-cut 主色提取; CornerHarris 重写(SIMD Sobel+滑动窗,快约 2.5 倍)。 随机森林 RandomForestClassifier/RandomForestRegressor(CvSharp.ML):LightGBM 式直方图分裂,训练快经典排序实现约 5 倍, 样本存原生堆无托管数组上限,支持任意 int 标签、OOB 评估、特征重要性、模型序列化与 UserInfo 自定义信息——与 LBP/HOG 组合即完整纹理分类链; 批量概率 PredictProbabilities(Mat) 直出 N×C 概率矩阵,配合 Mat.Reshape 零拷贝样本矩阵即逐像素分类链。 3D 几何(CvSharp.Geometry3D,对标 Open3D):点云容器/KD-tree/体素降采样/法向/曲率/统计与半径离群点/ RANSAC 平面·球·圆柱/欧式聚类/区域生长/ICP 点到点·点到面/FPFH·PPF 特征与粗配准/OBB/深度图与视差/ StereoBM·SGBM/法向图·曲率图滤波/三角网格全套拓扑修复·平滑·简化·体积·惯性张量/PLY·PCD·OBJ 读写; 热路径全 C++ 原生(AVX2+多线程),同机对标 Open3D 0.20 全部算子不慢于它。 OpenCV 主库与 contrib 都没有这层统一数据模型(仅零散 PPF/ICP/rgbd 接口),本库 78 个原生导出函数一次补齐,不依赖 OpenCV/VTK。 OpenCV 对标细节:CopyTo 自动分配目标/带掩码复制、AddWeighted·Min·Max 支持 U8/U16/S16/S32/F32/F64、 Merge/Split 1..512 通道、Laplacian 可选输出深度、LBP 任意半径与逐像素局部直方图、ApplyColorMap Turbo、 DrawContours 抗锯齿/层级/偏移。 检测后处理与标定:NMS 非极大抑制、旋转矩形矫正裁剪 DeskewCrop、亚像素角点 CornerSubPix、旋转中心标定 FitRotationCenter(LMedS 抗粗差)、 棋盘格相机标定(张正友+LM 精化,直接对接 Undistort);ShapeMatcher 支持掩码建模屏蔽可变区域。 Mat 缓冲池:释放缓冲按尺寸复用,临时 Mat 高频分配零页故障(Dft/相位相关提速约 2 倍)。 工业测量扩展:EdgesSubPix 亚像素轮廓(XLD)、ExtractLinesSteger 线条/划痕/光条中心线(含逐点宽度)、 UnwrapAnnulus/WrapAnnulus 圆环极坐标展开回卷、EstimateSimilarity2D/EstimateAffine2D/EstimateHomography RANSAC 鲁棒变换估计(内点掩码+RMS)、CalibrateHandEye 手眼标定(Tsai-Lenz,eye-in-hand/eye-on-base)。 特征与光流:OrbDetect/OrbMatch(GFTT+主方向+旋转导向 BRIEF-256 多尺度 ORB,双向交叉匹配)、 CalcOpticalFlowPyrLK(LK 金字塔稀疏光流,亚像素跟踪,点位并行); 频域与增强:NotchFilter/FindSpectralPeaks(周期纹理陷波消除)、GuidedFilter(引导滤波保边平滑,原生 O(1) 滑动窗)。 条码级二值化与一维信号:HybridThreshold(混合局部阈值,光照不均二维码/喷码主力)、ValleyThreshold(双峰谷值,失败可见)、 Filter1D 一维方向卷积、SampleLine 任意角度亚像素线段采样、BitMat 位打包二值矩阵; 去模糊与质量评估:DeblurWiener 维纳反卷积 + MotionBlurKernel 运动核生成、EstimateBlur 拉普拉斯方差模糊评分; 论文高频通用算子:Filter2D/SepFilter2D 任意卷积(AVX2)、GetGaussianKernel/GetGaborKernel 核生成、BoxFilter、 FastNlMeansDenoising 非局部均值去噪、PyrMeanShiftFiltering 金字塔均值漂移、HoughLines 标准霍夫、FitLine、ConvexityDefects 凸缺陷、 Kmeans(k-means++)、CalcBackProject 反向投影、CompareHist 四度量、CartToPolar/PolarToCart、Exp/Log/Pow/Sqrt、Reduce、MulSpectrums、 WarpPolar 极坐标展开、Accumulate 家族、ApplyColorMap 伪彩、Transpose/Repeat/Hconcat/Vconcat、GetRectSubPix、CornerMinEigenVal、椭圆/十字结构元; 图层底座(CvSharp.Layers):LayerStack/RasterLayer/GroupLayer 图层引擎,PS 兼容 27 种混合模式(原生多线程合成), 不透明度/蒙版/负偏移裁剪、图层组嵌套、Flatten(roi) 局部脏区域重绘、Revision 变更计数——标注与原图分离、类 PS 二次开发底座; 选区与掩码:PathToMask 路径转软边缘掩码(所有选区工具统一出口,NonZero/EvenOdd 填充规则、4× 超采样抗锯齿、羽化半径)、 MaskBoolean 掩码并/交/差(连续值 max/min 语义,原生多线程)、FeatherMask 羽化——产出掩码直接插 RasterLayer.Mask; SnapToEdges 磁性套索(Live-Wire:Sobel 代价图 + Dijkstra 逐段最短路径,锚点间路径自动贴边,闭环+去阶梯平滑)、 FlattenBezier 钢笔出口(三次贝塞尔自适应展平成点列,de Casteljau 二分,px 容差控制密度); 全部带核算子核尺寸自动钳制(核大于图不再崩溃),小图 ROI/金字塔末端免防御。 全码制条码识别 CvSharp.Barcode 已内嵌本包(1.3.0 起):QR/DataMatrix/PDF417/Aztec/GS1 DataBar 全族及 9 种一维码, 候选区定位先行、任意角度一维码、去模糊重试,HALCON 图集实测识别率全面超越 ZXing-C++(详见 docs/性能对比报告.md)。 原生 DLL 已内嵌包中(runtimes/win-x64/native),装包即用。
A high-performance .NET library that extends Emgu.CV (OpenCV wrapper) with zero-copy Span/Memory Mat accessors, async stream copy helpers, ROI and perspective crop utilities, boolean mask operations, pluggable Mat compression (PNG, Deflate, GZip, ZLib, Brotli, Zstd), CMat for memory-efficient compressed image storage, structured contour hierarchy wrappers, and drawing helpers for polygons and Unicode multi-line text with custom fonts, wrapping, fitting, and rotation.
Easy C++ projects configuration with OpenCV SDK
IronOcr.Extensions.AdvancedScan is an Extension Package to IronOCR: https://www.nuget.org/packages/IronOcr This package is not stand-alone. The IronOcr package must also be installed first, and a IronOcr License or Trial Key must be applied. More information: https://ironsoftware.com/csharp/ocr/licensing/ Supported .NET Versions: * .NET 8 * .NET 7 * .NET 6 * .NET 5 * .NET Core 2.0 - 3.1 * .NET Framework 4.6.2 + * .NET Standard 2.0 + Features: * Support for Advanced ML Model to Read and Decode more advanced image templates * License Plate Reading * Passport (MRZ) Reading * Noisy Image and Screenshot Reading * Data Table Reading * Locate Text Regions using Computer Vision
C# wrapper for Realeyes Demographic Estimation Library - face detection and demographic estimation (age, gender, age uncertainty)
Evil DICOM with a twist: OpenCV matrices and computer vision components. Useful for advanced operations with DICOM files.
DotNet Train, evaluation, export library for TensorFlow object detection 2.6.0.
The Basler pylon .NET API provides a convenient object-oriented programming interface to access Basler cameras. This package allows you to build projects that use pylon .NET API features. To run the resulting program, the pylon Camera Software Suite has to be installed.
A library for auto features on the computer (host-based). With the IDS peak AFL, you can e.g. use the autofocus feature of compatible cameras.
By uploading an image or specifying an image URL, Microsoft Computer Vision algorithms can analyze visual content in different ways based on inputs and user choices
First, subscribe on Microsoft Azure. Microsoft gives a 7-day trial Subscription Key https://azure.microsoft.com/en-us/try/cognitive-services/ . We can use that Subscription key for testing purposes. you need to log into the Azure Portal with our Azure credentials. Then we need to create an Azure Computer Vision Subscription Key in the Azure portal. Click on "Create a resource" on the left side menu and it will open an "Azure Marketplace". There, we can see the list of services. Click "AI + Machine Learning" then click on the "Computer Vision" Provision a Computer Vision Subscription Key
YoloDotNet OpenVINO Execution Provider enables optimized inference using Intel® OpenVINO™ on supported Intel CPUs, integrated GPUs, and accelerators. This execution provider integrates ONNX Runtime with Intel OpenVINO to deliver high-performance, low-latency inference on Intel hardware across Windows and Linux. It is ideal for CPU-focused deployments, edge systems, and environments where Intel hardware acceleration is preferred over CUDA-based solutions. The provider is fully modular and designed to work with the execution-provider-agnostic YoloDotNet core library introduced in v4.0. Only one execution provider should be referenced per project.
Optical Character Recognition (OCR) detects text in an image. Extracts characters into a usable character. First, subscribe on Microsoft Azure. Microsoft gives a 7-day trial Subscription Key https://azure.microsoft.com/en-us/try/cognitive-services/ . We can use that Subscription key for testing purposes. you need to log into the Azure Portal with our Azure credentials. Then we need to create an Azure Computer Vision Subscription Key in the Azure portal. var caracters= OCR.GetImageOCR(key,imagePath, endPoint = "https://westus.api.cognitive.microsoft.com/vision/v1.0/ocr")
FrameFlow — cross-platform FFmpeg media pipelines for .NET: decode, encode, play back, capture from cameras, and run inference on frames. UI-agnostic core
First, You need to log into the Azure Portal with our Azure credentials. Then we need to create an Azure Computer Vision Subscription Key in the Azure portal. Then create a resource to get the Key Second, Set the endPoint to "https://westus.api.cognitive.microsoft.com/vision/v1.0/analyze" Finaly to Image information captured using Computer Vision API you should write like this : ImageInfo imageInfo = new ImageInfo(); var imageInformations= imageInfo.MakeAnalysisRequest(subscriptionKey, imagePath, endPoint);
First, subscribe on Microsoft Azure. Microsoft gives a 7-day trial Subscription Key https://azure.microsoft.com/en-us/try/cognitive-services/ . We can use that Subscription key for testing purposes. you need to log into the Azure Portal with our Azure credentials. Then we need to create an Azure Computer Vision Subscription Key in the Azure portal.
FrameFlow — cross-platform FFmpeg media pipelines for .NET: decode, encode, play back, capture from cameras, and run inference on frames. UI-agnostic core
First, subscribe on Microsoft Azure. Microsoft gives a 7-day trial Subscription Key https://azure.microsoft.com/en-us/try/cognitive-services/ . We can use that Subscription key for testing purposes. you need to log into the Azure Portal with our Azure credentials. Then we need to create an Azure Computer Vision Subscription Key in the Azure portal.
First, subscribe on Microsoft Azure. Microsoft gives a 7-day trial Subscription Key https://azure.microsoft.com/en-us/try/cognitive-services/ . We can use that Subscription key for testing purposes. you need to log into the Azure Portal with our Azure credentials. Then we need to create an Azure Computer Vision Subscription Key in the Azure portal.
Librería para procesamiento de documentos de identidad con Azure Computer Vision y ML.NET
High-performance image processing, computer vision primitives, Otsu binarization, adaptive integral thresholding, separable Gaussian blur, Sobel edge detection, and morphology for .NET.
Clean, automatic C# wrapper for OpenCV 5 (Windows x64). Provides 2,600+ API methods with robust IDisposable memory management. Note: Installed footprint is ~112 MB due to bundled native OpenCV binaries.
The C# SDK for EyesOnIt. EyesOnIt is a computer vision product based on a Large Vision Model. The EyesOnIt model was trained on 2 billion images. It detects thousands of types of objects and events based on natural language descriptions. Using this SDK, EyesOnIt can be easily integrated into existing products or projects by those without advanced computer vision or data science skills. EyesOnIt is container based software that can easily be deployed at the edge, on-premises or in cloud environments.
TorchImage seamlessly integrates powerful image processing capabilities into TorchSharp using SixLabors.ImageSharp
SAM3 (Segment Anything 3) concept segmentation (text and box exemplars), interactive click segmentation and segment-everything for .NET via ONNX Runtime. Encode images once, prompt interactively; returns masks, boxes and scores.
YoloDotNet.ExecutionProvider.CoreML enables hardware-accelerated inference on macOS using Apple’s native Core ML framework. This provider integrates ONNX Runtime’s CoreML execution backend, allowing models to run efficiently on Apple Silicon by leveraging system-level ML acceleration. No external runtimes, drivers, or SDKs are required beyond a supported version of macOS. Designed for YoloDotNet’s modular, execution-provider-agnostic architecture, the CoreML provider integrates cleanly with the core library and exposes predictable, explicit inference behavior. It is well-suited for macOS applications that require fast, low-power, native inference using Apple’s ML stack without additional dependencies.
AI-powered component schema generator for Umbraco CMS 17. Turn UI screenshots into editable Umbraco Document Types, Element Types, properties and Block List Data Types. Supports multiple AI providers and vision-capable AI models.
Clean, automatic C# wrapper for OpenCV 5 with GPU/CUDA acceleration (Windows x64). Provides 2,600+ API methods with robust IDisposable memory management. Note: Installed footprint is large due to bundled native OpenCV CUDA binaries.
Clean, automatic C# wrapper for OpenCV 5 with GPU/CUDA acceleration (Linux x64). Provides 2,600+ API methods with robust IDisposable memory management. Note: Installed footprint is large due to bundled native OpenCV CUDA binaries.
YoloDotNet.ExecutionProvider.DirectML enables hardware-accelerated inference on Windows using Microsoft’s DirectML framework. This execution provider integrates ONNX Runtime’s DirectML backend and runs on top of DirectX 12, allowing inference to be accelerated on a wide range of GPUs using the Windows graphics driver stack. DirectML is a Windows-only technology and is supported on Windows 10 and Windows 11 with a DirectX 12–capable GPU. No vendor-specific SDKs or external runtimes are required beyond the standard Windows graphics drivers. This makes the DirectML execution provider a low-friction option for GPU acceleration on Windows systems. Designed for YoloDotNet’s modular, execution-provider-agnostic architecture, the DirectML provider integrates cleanly with the core library and exposes explicit, predictable inference behavior. It is well-suited for Windows applications that require GPU acceleration without locking into a specific hardware vendor or proprietary ML stack.
High-performance ONNX Runtime framework for real-time computer vision pipelines with strongly typed postprocessing.
OpenCV/DLib component of HoloTrack, available as a package for Computer vision-related projects. (NOTE: Not all OpenCV/DLib functionality is included, and are usually abstracted away by us!)
.NET wrapper for the Subvision CV native library. Provides impact detection and sheet coordinate extraction for underwater target shooting sheet analysis. Includes native binaries for Windows x64 and ARM64.
SDK .NET idiomatik untuk DepthAI API V3 (kamera OAK dari Luxonis). Pipeline node graph, stereo depth, neural inference, dan stream reaktif tanpa perlu Python.
A fully managed .NET binding for OpenCV 5, covering image processing, video analysis, camera calibration, object detection, and machine learning. Includes built-in code analysis rules that catch common mistakes when working with image data. This package is the managed API only; a small platform-specific add-on package is also required for the native OpenCV binaries to load.
C# wrapper for Realeyes Liveness Library - face detection and liveness checking
Diagnostics helpers for NeuroModFlowNet.ONNX runtime contexts and model metadata.
Contract-faithful C# port of the Python dynaface-lib facial-analysis pipeline: BlazeFace detection, SPIGA WFLW-98 landmarks + headpose, U²-Net background removal, frontal and lateral analysis, and the full measurement set. Backend-agnostic — supply neural-network inference by implementing IDynafaceInference (see the Dynaface.Onnx package for a ready-made ONNX Runtime backend).
Visualization helpers for rendering NeuroModFlowNet.ONNX inference results.
Template `dotnet new` untuk aplikasi computer vision berbasis OAK: depthai-console, depthai-web, depthai-desktop.
Local PIV facial-image preparation with bounded JPEG 2000 encoding and explicit verification evidence.