Measurific

Sub-pixel dimensional inspection for backlit, telecentric machine-vision images.

Measurific is a desktop app for macOS, Windows and Linux. It takes a silhouette image from a file or a live camera, extracts sub-pixel contours, recognizes the geometry (lines, arcs, circles, slots, threads and tooth profiles), and measures it: distances, diameters, angles, radii, pitch, thread and profile parameters, with tolerance checks and an annotated report.

Everything runs locally on the CPU (GPU optional). The app ships with its own Python runtime and ONNX models, so it needs no separate install.

(MP4 version)

Download

Installers for each version are on the Releases page:

Platform	File	Notes
macOS (Apple silicon)	Measurific-<version>-mac-arm64.dmg	Drag to Applications.
Windows (x64)	Measurific-<version>-win-x64.exe	Per-user install, no admin rights needed.
Linux (x64)	Measurific-<version>-linux-x86_64.AppImage	chmod +x, then run.

Builds are not yet signed with a paid certificate:

• macOS: the first launch is blocked. Open System Settings → Privacy & Security and click Open Anyway.
• Windows: SmartScreen shows a warning. Click More info → Run anyway.

The app checks GitHub Releases for updates. Uninstalling keeps your settings and calibrations.

Cameras

The bundled camera bridge uses Aravis and works with USB3 Vision and GigE Vision cameras without vendor SDKs. See camera-bridge/README.md.

Repository layout

Path	What it is
app/	Electron + React desktop app and its Python worker. Packaging and release configuration.
camera-bridge/	Native Aravis client (C) that the app runs as a child process.
packages/measurement/	inspection-measurement: CPU-only geometric measurement from contours and a JSON request.
packages/edge-model/	inspection-edge-model (edgesr): learned sub-pixel edge and coarse segmentation models.
packages/pae/	inspection-pae: Partial Area Effect sub-pixel edge detector (MIT fork of subpixel-edges).
packages/contour-model/	inspection-contour-model: learned contour-primitive recognizer, training code, and released checkpoints.
research/	Synthetic data generators and renderers used for training and validation.
docs/	Release process.

How the pieces fit:

image ─▶ edge extraction ─────────▶ contour recognition ─▶ measurement ─▶ report
         (coarse model + PAE,        (contour-model,        (measurement
          or SDF edge model)          ONNX Runtime)          package)

The app bundles packages/contour-model/release-v6/onnx, packages/edge-model/models/sdf_v5-onnx and packages/edge-model/models/coarse_v2-onnx.

Development

Requirements: Node.js 22, uv and Python 3.13. Building the camera bridge also needs CMake, Ninja and pkg-config (see camera-bridge/README.md).

cd app
npm install
uv venv .venv --python 3.13
uv pip install --python .venv/bin/python \
  -e '../packages/edge-model[onnx]' -e ../packages/pae -e '../packages/measurement[test,image]' \
  -e '../packages/contour-model[test,bridge]' -e '../research/synthetic-shapes[test]'
npm run dev

The development environment installs PyTorch for the training code. The packaged app uses ONNX Runtime only.

Run the tests:

cd app && npm test && npx tsc -b
cd app && .venv/bin/python -m pytest -q python
cd packages/measurement && ../../app/.venv/bin/python -m pytest -q

Each package's own tests run the same way from its folder. See CONTRIBUTING.md for details and docs/RELEASING.md for building installers and publishing a release.

License

MIT © Kyle Bartholomew. packages/pae is derived from gravi-toni/subpixel-edges (MIT). Released installers bundle third-party components under their own licenses, including LGPL libraries in the camera bridge. See THIRD_PARTY_NOTICES.md.