VisionLab SDK

An embeddable machine-vision runtime and plugin SDK for industrial inspection equipment.

VisionLab lets machine builders and system integrators ship inspection/measurement machines without maintaining an in-house vision-algorithm team. You embed the runtime into your machine, teach an inspection recipe (ROIs + algorithms + between-step geometry + PASS/NG logic) before shipment, and the machine then runs the recipe offline, per device.

This repository contains the public SDK surface: the plugin ABI, the integration protocol, host-side SDKs, sample plugins/recipes, and user/feature documentation.

Model: the SDK and ABI here are open (Apache-2.0). The runtime engine is proprietary and licensed commercially (per-device / seat). See Licensing.

---

What you can build

• Custom algorithm plugins that run inside the runtime (in-process C ABI) — your model participates in the algorithm registry, the parameter UI, and the recipe system.
• Standalone host applications that drive the runtime over shared-memory IPC (RunRecipe for a whole recipe, or a streaming session for many shots).
• Recipes that combine built-in algorithms with between-step geometric constraints to emit a semantic PASS/NG result (pose / presence / defect / string / measured values).

Capabilities (built into the runtime)

• Sub-pixel shape fitting: circle, line, ellipse, rectangle.
• Template matching (gradient-orientation, rotation-invariant).
• Blob analysis, golden-template diff, scratch / crack detection.
• OCR (DNN / dot-matrix), barcode / QR, DataMatrix.
• Unsupervised anomaly detection (PatchCore-style — build a library from good parts, flag anything off the normal feature manifold) for unknown defects.
• Dataset annotation + YOLOv8 training in-tool (C++/LibTorch, no Python) — annotate bounding boxes, fine-tune/train a detector, test it, and export ONNX for inspection.
• Process-chain recipes: multi-shot capture, cross-field-of-view coordinate unification, measurement/compute nodes, coordinate frames and placement arrays.
• Verification tooling: batch runs, GRR / repeatability reports, SQLite result logging, per-execution debug capture (original + result image + reproducing parameters).

See docs/USER_MANUAL.md, docs/RECIPE_GUIDE.md and docs/ALGO_USAGE_GUIDE.md.

Two integration models

Model	Where	How the plugin runs	Use when
In-process algorithm plugin	shared/plugin_sdk/algo_plugin	C ABI DLL loaded by the runtime	You want your algorithm inside the runtime's UI + recipe engine
Out-of-process host SDK	Master/	Your app launches the runtime and talks to it over shared-memory IPC	You want the runtime as a black-box "vision brain" driven by your own software

1. In-process plugin (C ABI)

// shared/plugin_sdk/algo_plugin/circle_qt_plugin.h
CQ_EXPORT int circle_qt_plugin_register(const char* appVersion, char* outJson, int outBufSize);
CQ_EXPORT int circle_qt_plugin_run(const char* algo, const CQImage* img,
                                   const char* paramsJson, char* outResultJson, int outResultBufSize,
                                   CQImage* outDisplayBgr, char* errBuf, int errBufSize);

See docs/plugin_abi.md and a complete example under shared/plugin_sdk/algo_plugin/examples/.

2. Host SDK (out-of-process, shared memory)

// Master/QtHost/IpcClient.h  — RunRecipe / StartRecipe+PushShot+FinishRecipe
IpcClient::sendRunRecipe(images, recipeJson);

# Master/plugin_host.py
ipc.send_run_recipe([img0, img1], recipe_json)
ipc.send_start_recipe(recipe_json); ipc.send_push_shot(img, 0); ipc.send_finish_recipe()

// Master/WpfHost/PluginIpc.cs
await _host.RunRecipeAsync(images, recipeJson);

Compilable samples are included — build them directly:

• Master/QtHost/sample/ — cmake -S . -B build -DCMAKE_PREFIX_PATH=<qt> && cmake --build build
• Master/WpfHost/sample/ — dotnet build -c Release

See docs/PLUGIN_INTEGRATION_GUIDE.md and docs/RECIPE_IPC_PROTOCOL.md.

Documentation

Document	Contents
docs/plugin_abi.md	In-process plugin ABI specification
docs/PLUGIN_SDK_QUICK_START.md	Write your first plugin
docs/PLUGIN_SDK_API_REFERENCE.md	SDK API reference
docs/PLUGIN_SDK_WIN32_EXAMPLE.md · docs/PLUGIN_SDK_QT_EXAMPLE.md	Worked plugin examples
docs/PLUGIN_HOST_DEV_GUIDE.md	Build a host application
docs/RECIPE_IPC_PROTOCOL.md	IPC protocol + semantic result schema
docs/RECIPE_GUIDE.md	Recipes and geometric constraints
docs/CLI_JSON_SPEC.md	CLI JSON interface
docs/USER_MANUAL.md · docs/ALGO_USAGE_GUIDE.md	Using the tool and the algorithms
docs/TRAINING_GUIDE.md	Dataset annotation + YOLOv8 training / test / ONNX export
docs/VERSUS_HALCON.md	Capability comparison and positioning
tools/dataset/	Video frame extraction, 4K tiling, dataset layout helpers
tools/patchcore/	PatchCore anomaly detection — backbone export, memory bank, heatmaps

Licensing

• SDK / ABI / sample code / docs in this repository: Apache License 2.0 (see LICENSE). Use them freely, including commercially.
• Runtime engine: proprietary, licensed perpetually (node-locked) per device / per seat, with an optional annual maintenance/support subscription. It is not in this repository and is distributed on request, with a license key. Copying it does not grant a license.

Third-party components in this repository

shared/json.hpp is nlohmann/json, © 2013– Niels Lohmann, licensed under the MIT License (full text: https://github.com/nlohmann/json/blob/develop/LICENSE.MIT). The bundled runtime also includes third-party components (Qt under LGPL-3.0, OpenCV under Apache-2.0, and others); their notices and licence texts ship inside the runtime package as THIRD_PARTY_NOTICES.md + licenses/.

Getting the runtime (key on request)

1. Open a license request issue (evaluation or commercial; state your OS and use case).
2. Download the runtime — VisionLab-runtime-win64.zip from the latest release — and unzip it.
3. Download getuuid.exe (a single self-contained exe), run it on the target machine, and send back the fingerprint.
4. You get a license.json bound to that machine — put it next to VisionLab.exe and run. (The runtime will not start without a valid license.)

Details: docs/EVALUATION_AND_LICENSING.md.

Contact

• Repository: https://github.com/binary-pixels/visionlab-sdk
• Issues: https://github.com/binary-pixels/visionlab-sdk/issues
• GitHub: @binary-pixels

Not included

The runtime engine source, the algorithm internals, roadmaps and commercial terms are not part of this repository. Some optional integrations reference commercial third-party libraries (e.g. Basler pylon, HALCON) that are not redistributed here and must be licensed separately by you.