OpenVisionLab — 2D Inspection Workbench

Build and validate rule-based 2D inspection Recipes on Windows with OpenCvSharp 4 and C#.

English | 한국어

Try it in 3 steps

1. Download the win-x64 ZIP from Releases. It requires the .NET 8 Desktop Runtime (x64).
2. Extract the ZIP into a new folder and run OpenVisionLab.exe.
3. Open a public sample in Sample Catalog, then choose Preview or Run.

The published ZIP is v2.1.0-rc.3; it predates the source workflow shown here. To use that workflow, build from source: clone the repository, build the solution, then press F5.

Current version: v2.2.0-dev.11 · Changelog

What it does

• Teach inspection Tools through PropertyGrid parameters, ROIs and acceptance limits.
• Connect Tools in a Pipeline and save the Steps, parameters and Layer connections as Recipe XML.
• Compare source, intermediate and output Layers, detection drawings, measurements and OK/NG reasons.
• Check the same Recipe against Good and Bad samples with Batch inspection.
• Use LLM/XML assistance optionally; it is not required to teach or run an inspection.

Build and Check a Recipe

1. In Recipe Manager, create Shape Count, then choose Open work with selected inspection and open an image.
2. Open Threshold. Turn Run on change off, set the threshold to 150, and choose Run tool (F8). Compare the original and binary images, then choose Add to pipeline → Add at end.
3. Open Contour and confirm the input is Threshold_1. Turn Use threshold off, draw the inspection ROI around the five shapes, then choose Apply or press Enter.
4. Enable Acceptance and select Detection count. Check Lower and Upper, set both to 5, and press F8. Confirm 5 detections, then press Ctrl+Enter to add the Step.
5. Run the Pipeline with F5. Review Results and compare source, intermediate and output Layers in the one-, two- or four-panel viewer.
6. Save with Ctrl+S. Select another Recipe and return to reload the saved Steps, parameters, ROI and Layer connections.
7. Add Good and Bad images and run Batch inspection with the same Recipe. Inspect each image's measurements and OK/NG result.

With Run on change off, parameter changes leave the result pending until the Tool runs. Loading a Recipe restores its settings without running the inspection. Saving and running are separate actions.

If no enabled Step has acceptance criteria, a run can finish without an OK/NG verdict. Execution errors are reported separately from NG. After input or parameter changes, retained results are marked as previous results; run again to update them. Batch inspection does not retain every intermediate image. Run the selected image to inspect those outputs.

Use the saved Shape Count Recipe on the original and missing-shape images. Assign Good and Bad labels, then run Batch inspection. With the ROI and count limits above, the original returns 5 detections / OK and the missing-shape image returns 2 detections / NG: 1 OK, 1 NG and 0 errors.

Good/Bad labels specify the expected class; OK/NG is the inspection result. Select an image to compare its source, processed image, overlays and acceptance limits. Page Up returns to the previous image.

One Image, Every Inspection Stage

Compare the source, filtered, binary and cleaned images in the four-panel viewer. The public perforated-plate Recipe runs Filter -> Threshold -> Morphology, then feeds the cleaned Layer into Blob and Contour. Both branches return 34 accepted holes in Results. The same area limits exclude the contour around the full-image background.

Open the example Pipeline or select Product_Field_PerforatedPlate_HoleArray in Sample Catalog.

Adjust each Step while comparing its intermediate image, then check the final overlays and measurements in Pipeline Review. Preview and Run execute only when requested.

Tool Setup and Results

The recording shows Matching, Line, Blob and Contour settings and results in the main workbench.

Watch the Tool View MP4

• Matching: inspect the loaded template, matching scores and detections.
• Line: choose Edge, Distance or Intersection mode; this example measures pin spacing.
• Blob: inspect particle detections and count limits.
• Contour: set the inspection ROI and count limits, then check the detected shapes.

Use Run step (F8) for the selected Tool or Run pipeline (F5) for the complete sequence.

Chained Processing

Connect preprocessing and detection Tools in a Pipeline. The recording compares intermediate images and results from two three-Step Pipelines, including the Morphology operation and kernel settings:

• Filter -> Threshold -> Contour
• Threshold -> Morphology -> Contour

Watch the chained-processing MP4

Select a Step in Pipeline Review to inspect its intermediate image. Check the final Step for detected objects and overlays. Run a Tool with F8 or the Pipeline with F5. Tools with Run on change can also run when parameters change.

The next example connects Filter → Threshold → Contour and Threshold → Blob. Compare the input and output images, then inspect the detection count and acceptance limits.

Main Features

Area	Description
Main Workspace	Load images, manage Layers, zoom, pan and inspect pixel values
Layer Docking	Compare source, intermediate and final images in tabs or split views
Tool View	Configure Threshold, Filter, Morphology, Blob, Contour, Matching, FeatureMatching, EdgeBasedMatching, Line/Length, Mean and other Tools
PropertyGrid	Edit Tool parameters through a consistent property interface
Preview / Run	Run a Tool or complete Pipeline; control parameter-triggered execution with Run on change where supported
Pipeline / Recipe	Connect Tools as ordered Steps and save them as XML
Pipeline Review	Inspect Step inputs and outputs, measurements, drawings, logs and OK/NG reasons
Sample Catalog	Browse sample images, baseline Pipelines, expected results and Good/Bad pairs
Good/Bad Pair	Compare normal and defective samples using the same measurements
Learn Mode	Open guidance for Matching, Blob, Contour, Threshold, Mean, FeatureMatching, EdgeBasedMatching and Line
Language	Switch between Korean and English

Pipeline Review also shows connections and branches, retained result images, acceptance ranges and values, and parameters to adjust. Metrics include ResultCount, AreaMax, ScoreMax and LineAngleAvg. See the workspace reference and Pipeline Review image.

Public Samples

Public samples include images, recommended Pipelines, measurements and Good/Bad expectations. Browse the public catalog or product catalog. The root Sample/ folder holds local or third-party data and is not part of the public sample distribution.

Task	Good sample	Bad sample	Measurements
Locate a target with Matching	Public_Matching_DiePad_Good	Public_Matching_DiePad_NoTarget_Bad	ResultCount, ScoreMax
Count particles with Blob	Public_Blob_Particles_Good	Public_Blob_Particles_Sparse_Bad	ResultCount
Count shapes with Contour	Public_Contour_Shapes_Good	Public_Contour_Shapes_Missing_Bad	ResultCount
Segment bright pads with Threshold	Public_Threshold_BandPads_Good	Public_Threshold_BandPads_Missing_Bad	ResultCount
Check brightness with Mean	Public_Mean_Brightness_Good	Public_Mean_Brightness_Dark_Bad	MeanValueAvg
Compare feature-matching scores	Public_Feature_Card_Good	Public_Feature_Card_Wrong_Bad	ScoreMax, ResultCount
Compare edge-based shapes	Public_Edge_Fiducial_Good	Public_Edge_Fiducial_Wrong_Bad	ScoreMax, ResultCount
Measure a distance with Line	Public_Line_Pins_Good	Public_Line_Pins_WidePin_Bad	DistanceMmAvg

Build from source

Windows 10/11 x64; Visual Studio 2022 17.8+ with .NET desktop development. Git must be on PATH; the first NuGet restore needs internet access.

1. Clone https://github.com/Noah8218/OpenVisionLab.git to a short path.
2. Open OpenVisionLab.sln; set OpenVisionLab as the startup project.
3. Choose Debug / Any CPU, build the solution, then press F5.

Full prerequisites, command-line builds and Windows Sandbox.

Documentation

Select Guide for the offline manual in the current UI language. Its workflow sections and 17 Tool chapters include screenshots, numbered callouts, steps to follow and results to check.

User manual · Tutorial · Learn Mode Build · Changelog · Documentation index

Status and limitations

The source is v2.2.0-dev.11; the published ZIP is the older v2.1.0-rc.3 prerelease. LLM/XML assistance is optional and is not part of inspection execution. Validate Recipes with representative production images before using them for inspection. Check UI changes in the build that contains them; recordings do not verify later changes.

Part of OpenVisionLab

Project	What it does
2D Workbench	Rule-based 2D inspection Recipes
3D Studio	3D teaching, Thickness and Warpage
Labeling Studio	Labeling, training and model comparison
Machine Studio	Machine simulation before hardware
Vision SDK	Shared C# inspection library

License

OpenVisionLab is licensed under the Apache License 2.0.

This project includes software developed by 최노아(Noah-Choi).
Copyright (c) 2026 최노아(Noah-Choi).

Commercial use, modification and redistribution are permitted under the license terms. Copies or substantial portions must retain LICENSE, NOTICE, copyright and attribution notices.