MediaPipe.NET graph engine: calculator graphs of nodes connected by timestamped Packet<T> streams, with MediaPipe-style input synchronization, timestamp bound propagation, parallel pipelined scheduling, flow limiting (packet dropping) for live streams, side packets and a .pbtxt-style graph config parser. Created by Gravicode Studios, led by Kang Fadhil.
Selfie segmentation model for MediaPipe.NET (person/background mask). Converted from Google MediaPipe (Apache-2.0) and cross-validated against TFLite. Created by Gravicode Studios, led by Kang Fadhil.
Hand models for MediaPipe.NET: BlazePalm detector, 21-point hand landmark model, gesture embedder and canned gesture classifier. Converted from Google MediaPipe (Apache-2.0) and cross-validated against TFLite. Created by Gravicode Studios, led by Kang Fadhil.
Face models for MediaPipe.NET: BlazeFace short-range detector, Face Mesh V2 (478 landmarks with irises) and face blendshapes. Converted from Google MediaPipe (Apache-2.0) and cross-validated against TFLite. Created by Gravicode Studios, led by Kang Fadhil.
GSI biometric identity SDK binding for .NET MAUI (Android and iOS)
MediaPipe.NET — native .NET port of Google MediaPipe (face detection, face mesh, hand tracking, gestures, pose, holistic, segmentation, interactive segmentation, object detection, image classification and embedding, audio classification, voice activity detection, text classification and embedding, language detection) on ONNX Runtime. This package bundles the CPU runtime (plus CoreML on macOS) and works on Windows, Linux and macOS. Created by Gravicode Studios, led by Kang Fadhil.
Pose models for MediaPipe.NET: BlazePose detector and BlazePose GHUM Lite/Full landmark models (33 landmarks + segmentation). Converted from Google MediaPipe (Apache-2.0) and cross-validated against TFLite. Created by Gravicode Studios, led by Kang Fadhil.
EfficientNet-Lite0 image classification model for MediaPipe.NET (1000 ImageNet classes). Converted from Google MediaPipe (Apache-2.0) and cross-validated against TFLite. Created by Gravicode Studios, led by Kang Fadhil.
MediaPipe.NET task foundations shared by the vision, audio and text tasks: BaseOptions (model sources, execution provider, logging), model loading and classification post-processing. Created by Gravicode Studios, led by Kang Fadhil.
EfficientDet-Lite0 object detection model for MediaPipe.NET (80 COCO classes). Converted from Google MediaPipe (Apache-2.0) and cross-validated against TFLite. Created by Gravicode Studios, led by Kang Fadhil.
MediaPipe.NET with the DirectML (DirectX 12 GPU) ONNX Runtime for Windows — GPU acceleration on NVIDIA, AMD and Intel GPUs. Use instead of the MediaPipeNet package. Created by Gravicode Studios, led by Kang Fadhil.
MediaPipe.NET with the CUDA ONNX Runtime for NVIDIA GPUs on Windows and Linux (requires CUDA 12 and cuDNN 9). Use instead of the MediaPipeNet package. Created by Gravicode Studios, led by Kang Fadhil.
All pretrained models for MediaPipe.NET (face, hand, gesture, pose, holistic, segmentation, object detection, image classification and embedding, face stylization, audio, text). Models: Copyright Google LLC, Apache-2.0. Created by Gravicode Studios, led by Kang Fadhil.
Face detection library using ONNX Runtime and SkiaSharp.
User control for face detection and camera operation for uwp
mediapipenet-cli — run MediaPipe.NET tasks (faces, face mesh, hands, gestures, pose, holistic, segmentation, objects, classification) from the command line, benchmark them, and manage models. Install with: dotnet tool install -g Gravicode.MediaPipeNet.Cli. Created by Gravicode Studios, led by Kang Fadhil.
Mediapipe.NET component of OpenSense. This package uses the CPU runtime, if you want to use other runtimes, please compile from source. The face_detection and face_landmark models are included for convenience.
Text models for MediaPipe.NET: MobileBERT and average-word sentiment classifiers, MobileBERT sentence embedder and the 110-language detector. Converted from Google MediaPipe (Apache-2.0) and cross-validated against TFLite. Created by Gravicode Studios, led by Kang Fadhil.
Image embedding model for MediaPipe.NET: MobileNet V3 small (1024-D feature vectors). Converted from Google MediaPipe (Apache-2.0) and cross-validated against TFLite. Created by Gravicode Studios, led by Kang Fadhil.
MediaPipe.NET visualization utilities: draw landmarks, skeletons, face meshes, bounding boxes and segmentation masks (overlay, background blur/replacement) on SixLabors.ImageSharp images for debugging and demos. Created by Gravicode Studios, led by Kang Fadhil.
.NET for Android (formerly Xamarin.Android) Binding Library - Agora Full Face Detect SDK
MediaPipe.NET video input via OpenCvSharp: WebcamFrameSource and VideoFileFrameSource implementations of IFrameSource with zero-allocation frame reuse. Add an OpenCvSharp4 runtime package for your platform (e.g. OpenCvSharp4.runtime.win). Created by Gravicode Studios, led by Kang Fadhil.
Microsoft.Extensions.DependencyInjection integration for MediaPipe.NET: services.AddMediaPipeNet(...).AddFaceDetector().AddHandLandmarker()... registers thread-safe task singletons for ASP.NET Core and Worker Services, wired to ILogger. Created by Gravicode Studios, led by Kang Fadhil.
Audio models for MediaPipe.NET: YAMNet audio event classifier (521 AudioSet classes, also used for voice activity detection). Converted from Google MediaPipe (Apache-2.0) and cross-validated against TFLite. Created by Gravicode Studios, led by Kang Fadhil.
MediaPipe.NET text tasks: TextClassifier (MobileBERT and average word-embedding sentiment models), TextEmbedder (MobileBERT sentence embeddings) and LanguageDetector (110 languages), with MediaPipe-exact BERT WordPiece, regex and character n-gram tokenizers — native .NET ports of Google MediaPipe Tasks running on ONNX Runtime. Created by Gravicode Studios, led by Kang Fadhil.
MediaPipe.NET audio tasks: AudioClassifier (YAMNet, 521 AudioSet classes) and VoiceActivityDetector, with audio-clip and audio-stream modes, WAV loading and band-limited resampling — native .NET ports of Google MediaPipe Tasks running on ONNX Runtime. Created by Gravicode Studios, led by Kang Fadhil.
SeasonVision is a .NET computer vision library built on ONNX Runtime, providing image classification, face detection, face landmarks, emotion recognition, face attributes, object detection, and instance segmentation helpers.
Reduced-precision variants of the MediaPipe.NET models: FP16 (half size, best on GPUs) and weight-only INT8 (about a quarter of the size), each validated against its float32 model. Selected with InferenceOptions.Precision; models without a variant fall back to float32. Created by Gravicode Studios, led by Kang Fadhil.
spaCy-class NLP for .NET: the model-free core. Doc, Span and Token with character offsets into the original text; tokenizers that match Hugging Face tokenizers byte for byte; sentence segmentation; Matcher and EntityRuler; recognizers for numbers, units, currency, dates and times in ten languages; language identification for 75 languages. No native code, NativeAOT and trim clean. Unaffiliated with other projects named OneNLP.
Android binding library for GsiMaui face detection SDK.
YuNet face detection and SFace face recognition running on OpenVINO through OpenVinoSharp (Guojin Yan). Pure C# preprocessing, alignment and post-processing - no OpenCV dependency. Face detection: opencv_zoo face_detection_yunet_2023mar. Face recognition: opencv_zoo face_recognition_sface_2021dec (128-D features, cosine / L2 matching with official thresholds).
Face stylizer model for MediaPipe.NET: BlazeFaceStylizer color sketch (256x256) and the 192x192 face mesh it aligns faces with. Converted from Google MediaPipe (Apache-2.0) and cross-validated against TFLite. Created by Gravicode Studios, led by Kang Fadhil.
A .NET MAUI library for face detection and document verification via the GSI SDK. Supports Android and iOS.
Real-time computer-vision components for Blazor on MediaPipe.NET (native .NET MediaPipe Tasks on ONNX Runtime): camera viewer and overlays, face detection, 478-point face mesh and blendshapes, attendance and verification, hand tracking and gestures, sign language, pose and exercise analytics, holistic tracking and motion capture, background removal, object detection, image classification, vision copilot, 3D avatars and industry, health, agritech and analytics dashboards.
caveman-ml: command-line manager and runner for the CavemanML models - install / status / verify / update / remove the SynthelionML and PrivacyGuardML models from Hugging Face, compress prompts with the learned compressor, and detect PII.
Managed face detection, 1:1 face verification, and camera liveness checks for .NET 10, with SIMD CPU inference and optional Vulkan acceleration.
OpenCvSharp interop for OpenVinoSharp.SFace. Accepts OpenCvSharp Mat input directly for YuNet face detection and SFace face recognition - no Bitmap detour for camera/realtime pipelines. Requires the main OpenVinoSharp.SFace package.
Face-aware maximum square cropping with an embedded BlazeFace short-range face detector.
Face detection and blur over a VisionImage for Linkway.Vision.Capture hosts: RetinaFace (ResNet-50) over ONNX Runtime, pixels through ImageSharp. The model ships embedded in the assembly.
AI for power systems: LLM grid-operator agent with tool calling (OpenAI, Azure OpenAI, Hugging Face, Ollama, offline), anomaly detection, fault classification, and reinforcement-learning environments for Volt/VAR control.