License Plate Annotation Tool

An approachable browser tool for turning vehicle footage into clean, useful license-plate datasets—without requiring computer-vision or command-line expertise.

Assembled and packaged by Hazen.ai, this tool brings existing open-source packages together in an easy-to-use browser interface. We released it to make license-plate dataset preparation accessible to people with little technical experience, reducing the time and manual effort needed to extract frames, remove repeated images, and locate plates.

It is powered by FastALPR, using ONNX models for plate detection and OCR.

Start in two steps

Python 3.10 or newer is required.

1. Install the tool:

   pip install license-plate-annotation-tool
   

2. Launch it:

   license-plate-annotation-tool
   

The browser interface opens automatically. Uploads, settings, processing, model downloads, and result downloads are handled inside the app, making it suitable for less-technical users.

What it does

The app provides three tools that can be used independently:

1. Video to frames — upload a video, optionally trim it, and sample it at 5, 10, 15, or 20 frames per second.
2. Reduce similar frames — upload images/ZIPs or reuse frames from step 1, then retain the clearest representatives from visually similar groups.
3. Extract license plates — upload images/ZIPs or reuse an earlier result, detect and crop plates, and name them with OCR text when available.

The Guided workflow runs all three stages from a single video. Every result is provided as a downloadable ZIP containing the output images, manifest.csv, and run_config.json.

Run locally

Python 3.10 or newer is recommended.

For the easiest setup, run:

python launch.py

The launcher creates a private .venv, installs or updates the required components, and opens the web interface. After that, uploads, settings, processing, model downloads, and result downloads are all handled inside the interface.

If the launch command is not available in your terminal, use the equivalent module command:

python -m hazen_license_plate_annotation_tool

Models are downloaded and cached the first time plate extraction is run. CPU processing is supported; long videos and higher frame rates naturally take more time and disk space.

Tests

pip install --editable .
python -m unittest discover -s tests -v

Privacy and storage

Uploads and generated files are kept in a temporary directory for the active browser session. Download any results you want to keep before closing or restarting the app.

Build a release

python -m pip install --upgrade build twine
python -m build
python -m twine check dist/*

The resulting wheel and source archive in dist/ are ready for TestPyPI, PyPI, or the repository's GitHub publishing workflow.