MareArts ANPR

Automatic Number Plate Recognition — plate detection + OCR for 80+ countries, with GPU acceleration, ARM/Windows support, and Vehicle Intelligence (make, model, color, type, side, plate nation, and server-side OCR — cloud or on-device).

The ANPR SDK license covers: Python SDK · REST API Server · MASS · Road Object Detection. The Mobile App has its own license.

New SDK licenses open when 4.0 is released. Existing MareArts ANPR (SDK + Mobile) subscriptions keep working. · SDK page: marearts.com/pages/marearts-anpr-sdk · Try the live demo · Full docs & examples on GitHub

Install

pip install marearts-anpr

Linux / Windows / macOS · Python 3.10–3.14 · CPU by default (CUDA & DirectML optional).

What's new in 4.0

• v17 models with a MAEV3 licence key. Your MAEV3 key is on your My licenses page.
• Models download once, then run offline. The first run downloads each model with its certificate into ~/.marearts/anpr/. After that no internet is needed — copy that folder to an offline machine to use it there.
• v14 / v15 / v16 models and MAEV2 keys are not supported in 4.x. To keep using them, stay on 3.x: pip install "marearts-anpr<4" with your MAEV2 key.
• REST server access control. The server answers only this machine by default. When bound to the network, requests need the API token (ma-anpr server token).

Python SDK

Detect and read license plates in a few lines of Python.

from marearts_anpr import (
    ma_anpr_detector_v17, ma_anpr_ocr_v17, marearts_anpr_from_image_file
)

user_name = "your@email.com"          # the e-mail your licence was issued to
serial_key = "MAEV3:anpr|..."         # your MAEV3 licence key

detector = ma_anpr_detector_v17("640p_fp32", user_name, serial_key)
ocr = ma_anpr_ocr_v17("fp32", "univ", user_name, serial_key)

result = marearts_anpr_from_image_file(detector, ocr, "car.jpg")
print(result)
# {'results': [{'ocr': 'ABC1234', 'ocr_conf': 99, 'ltrb': [...], ...}], ...}

Models: detector 320p_fp32, 640p_fp32 · OCR fp32, int8.

Keeping the key in an environment variable? A MAEV3 key contains |, so quote it — unquoted, the shell reads | as a pipe and cuts the key short:

export MAREARTS_ANPR_SERIAL_KEY='MAEV3:anpr|...'           # Linux / macOS
set "MAREARTS_ANPR_SERIAL_KEY=MAEV3:anpr|..."              # Windows cmd
$env:MAREARTS_ANPR_SERIAL_KEY = 'MAEV3:anpr|...'           # Windows PowerShell

ma-anpr config stores it for you, with no quoting needed.

REST API Server

Production-ready server with a web dashboard — zero code deployment.

ma-anpr config              # one-time credential setup (MAEV3 key)
ma-anpr server start        # http://localhost:8000

curl -X POST http://localhost:8000/api/anpr -F "image=@car.jpg"

Serving other machines (ma-anpr server start --host 0.0.0.0): every request needs the API token — ma-anpr server token prints it and a dashboard sign-in link.

curl -X POST http://SERVER:8000/api/anpr -H "Authorization: Bearer <token>" -F "image=@car.jpg"

20+ endpoints: detect, batch, history, watchlist, alerts, stats, config, model selection, and more.

Docker

docker run -d --gpus all -p 8000:8000 \
  -e MAREARTS_ANPR_USERNAME="your@email.com" \
  -e MAREARTS_ANPR_SERIAL_KEY="your_MAEV3_key" \
  -e MAREARTS_ANPR_SIGNATURE="your_signature" \
  marearts-anpr-server:latest

Includes CUDA acceleration, automatic CPU fallback, web dashboard, and all endpoints.

Vehicle Intelligence (make / model / color / …)

Go beyond plate text — identify the vehicle itself: make, model, color, type, vehicle side, plate nation, and server-side OCR.

from marearts_anpr import ma_anpr_mmc

mmc = ma_anpr_mmc(user_name, serial_key, signature)
info = mmc.enrich(image, anpr_results)   # run ANPR first, then enrich
print(info)

Two backends, identical output fields:

• Cloud — no local model, results in a single API call.
• On-device (offline) — a compact vision-language model (Qwen3-VL-2B) on your GPU/CPU, with no internet and no quota.

ma-anpr mmc-setup           # download on-device model + engine (~1.5 GB)
ma-anpr mmc-backend local   # switch enrichment to on-device (or: cloud)
ma-anpr mmc-uninstall --all # cleanly remove the engine + model

Also available on the server via POST /api/anpr/mmc, with backend switching through GET /api/mmc/status and PUT /api/mmc/backend.

Road Object Detection

Detect persons, vehicles, and 2-wheelers in real time — included in your ANPR license.

from marearts_road_objects import ma_road_object_detector

rod = ma_road_object_detector("640p_fp32", user_name, serial_key, signature)
result = rod.detect("street.jpg")

See MareArts Road Objects.

Supported Regions

80+ countries across 12 regional groups, each with per-country character sets for maximum accuracy.

Pass a 2-letter country code (e.g. de, au, th) for best accuracy, or a group code. Unknown codes fall back to univ.

Mobile App

On-device ANPR for iOS and Android — parking, security, fleet management. On-device AI (~100–160 ms), offline capable, team collaboration, webhooks, cloud sync, and Vehicle Intelligence.

• Download on the App Store
• Get it on Google Play

The app has its own license, ANPR Mobile: marearts.com/products/anpr.

Support

© 2026 MareArts. All rights reserved.