.NET 10 console app with an interactive menu of small, focused OpenCvSharp samples, grouped by module.

dotnet run --project Samples

You'll get a numbered menu grouped by category (Core, ImgProc, Features2D, Calib3D, Video, Photo, ObjDetect, Dnn, Ml, Stitching). Enter a number to run that sample, c to clear the screen, h for help, 0 to exit.

Samples that would normally open a Window or block on Cv2.WaitKey can instead run without a display by writing their output frames as PNGs under headless-output/<SampleName>/:

dotnet run --project Samples -- --headless

or, equivalently, via an environment variable:

OPENCV_SAMPLES_HEADLESS=1 dotnet run --project Samples

$env:OPENCV_SAMPLES_HEADLESS=1; dotnet run --project Samples

This is how the whole sample suite can be smoke-tested in CI. Samples that inherently need live hardware (e.g. CameraCaptureSample) skip themselves with a warning when run headless.

• Program.cs — discovers every sample class via reflection (ITestBase + [SampleCategory]) and hands off to the console menu.
• Console/ — the menu/runner (ConsoleTestManager, ConsoleTestBase).
• Interfaces/DisplayHelper.cs — the display abstraction each sample uses instead of calling highgui directly, so the same sample code works both windowed and headless.
• Samples/ — the actual sample implementations, one subfolder per module (see below).
• Data/ — bundled images, video clips, cascade/model files, and ML datasets used by the samples. No downloads needed.
• Path.cs — centralizes paths into Data/ (ImagePath, MoviePath, TextPath).

• Calib3D — camera calibration + undistortion from chessboard images; stereo disparity (StereoBM/StereoSGBM).
• Core — DCT/DFT round trips, Mat submatrix/ROI operations, array-based pixel access and performance comparisons, multidimensional scaling, split/merge, solving linear equations.
• Dnn — super-resolution upscaling with a bundled FSRCNN model.
• Features2D — keypoint detectors and matchers: BRISK, FAST, FREAK, KAZE/AKAZE, MSER, SIFT/SURF, SimpleBlobDetector, StarDetector, FLANN matching, best-match bounding box.
• ImgProc — CLAHE, connected components, GrabCut, histograms, Hough lines, morphology, perspective warp, square/rectangle detection, Delaunay triangulation (Subdiv2D), watershed, adaptive binarization.
• Ml — SVM regression/classification.
• ObjDetect — ArUco marker detection, Haar/LBP face detection, HOG people detection, QR code detection.
• Photo — inpainting, watermark removal, seamless (Poisson) cloning.
• Stitching — image stitching from cropped overlapping patches. Known issue: currently not working reliably (see the TODO in Stitching.cs).
• Video — background subtraction, live webcam capture, Kalman filter tracking, CSRT object tracking, Lucas-Kanade optical flow, video read/write.

None beyond the .NET 10 SDK — all data assets are bundled. Video/CameraCaptureSample.cs additionally needs a webcam and is skipped automatically in headless mode.