Python SDK · Desktop app (MASS) · REST server

MareArts ANPR SDK

Detection and OCR for 80+ countries, a REST server with a dashboard, Docker, Road Objects, and Vehicle Intelligence.

Prefer an app? MASS is the same engine as a desktop plate recognition app for Windows, macOS and Ubuntu: no code, with plate history, watchlist alerts and the REST API. In beta testing now.

PyPI package marearts-anpr · Python 3.10–3.14 · current release 3.9.1. Docs and examples: github.com/MareArts/MareArts-ANPR.

Install

pip, then GPU if you have one

# CPU — Linux, macOS, Windows, ARM
pip install marearts-anpr

# NVIDIA CUDA
ma-anpr gpu-setup cuda

# Windows AMD / Intel / NVIDIA
ma-anpr gpu-setup directml

ma-anpr config      # username, serial key, signature
ma-anpr validate    # check the license

Credentials go to ~/.marearts/.marearts_env, or set MAREARTS_ANPR_USERNAME, MAREARTS_ANPR_SERIAL_KEY, and MAREARTS_ANPR_SIGNATURE.

Python

Detect and read in a few lines

from marearts_anpr import (
    ma_anpr_detector_v16, ma_anpr_ocr_v16, marearts_anpr_from_image_file
)

detector = ma_anpr_detector_v16("640p_fp32", user_name, serial_key, signature)
ocr = ma_anpr_ocr_v16("fp32", "univ", user_name, serial_key, signature)

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

Same call shape for a file, a PIL image, or a BGR array:

from marearts_anpr import marearts_anpr_from_pil, marearts_anpr_from_cv2

marearts_anpr_from_image_file(detector, ocr, "car.jpg")
marearts_anpr_from_pil(detector, ocr, pil_image)
marearts_anpr_from_cv2(detector, ocr, bgr_array)

You can pass detector only (boxes, no text), OCR only (a cropped plate), or both.

V16 models

RTX 4090, one process, detection + OCR (from the GitHub README, measured):

Full tables, backends, and CLI: python-sdk/README.md.

ma-anpr car.jpg
ma-anpr car.jpg --region kr --backend cuda
ma-anpr *.jpg --json results.json

Regions

Per-country character sets

Pass a 2-letter country code for best accuracy, a group code, or univ. Unknown codes fall back to univ. Switch without reloading the model.

ocr.set_region("kr")
ocr.set_region("eup")    # EU + Ex-USSR + UK
ocr.set_region("na")
ocr.set_region("univ")   # default

ch is Switzerland. China is cn. Country list: Regions in the SDK README.

Vehicle info

Go beyond plate text

Cloud Vehicle Intelligence adds make, model, colour, type, front or rear, plate nation, and a server-side OCR check. Needs internet. Local ANPR still returns if the cloud call fails.

from marearts_anpr import ma_anpr_mmc

mmc = ma_anpr_mmc(user_name, serial_key, signature)
result = marearts_anpr_from_image_file(detector, ocr, "car.jpg", mmc)
if result.get("mmc_error"):
    print(result["mmc_error"])   # plates are still in result["results"]
for r in result["results"]:
    print(r["ocr"], r.get("mmc_make"), r.get("mmc_model"), r.get("mmc_color"))

On-device, no quota: ma-anpr mmc-setup then ma-anpr mmc-backend local, and from marearts_anpr import ma_anpr_mmc_local. Same enrich fields. REST: POST /api/anpr/mmc.

Server

Dashboard and REST, same install

The server is in pip install marearts-anpr. No second package.

ma-anpr config
ma-anpr server start                 # http://127.0.0.1:8000
ma-anpr server start --daemon        # background
curl -X POST http://127.0.0.1:8000/api/anpr -F "image=@car.jpg"
curl -X POST http://127.0.0.1:8000/api/anpr/mmc -F "image=@car.jpg"

Dashboard at port 8000. Swagger at /docs.

ma-anpr server status
ma-anpr server list
ma-anpr server logs --follow
ma-anpr server stop
ma-anpr server detect car.jpg --mmc

server stop checks server_id: marearts-anpr before it kills a process. Full CLI: server/README.md.

Docker

Build the image, then run

Images are built from MareArts-ANPR/docker. GPU needs NVIDIA Container Toolkit. CPU image runs without --gpus.

git clone https://github.com/MareArts/MareArts-ANPR.git
cd MareArts-ANPR/docker

docker build -t marearts-anpr-server:latest .
docker run -d --gpus all --name marearts-anpr-server -p 8000:8000 \
  -e MAREARTS_ANPR_USERNAME="your@email.com" \
  -e MAREARTS_ANPR_SERIAL_KEY="your_serial_key" \
  -e MAREARTS_ANPR_SIGNATURE="your_signature" \
  -v ~/.marearts:/root/.marearts \
  marearts-anpr-server:latest

docker build -t marearts-anpr-server-cpu:latest -f Dockerfile.cpu .
docker run -d --name marearts-anpr-server-cpu -p 8000:8000 \
  -e MAREARTS_ANPR_USERNAME="your@email.com" \
  -e MAREARTS_ANPR_SERIAL_KEY="your_serial_key" \
  -e MAREARTS_ANPR_SIGNATURE="your_signature" \
  -v ~/.marearts:/root/.marearts \
  marearts-anpr-server-cpu:latest

Same dashboard and /api/anpr as the pip server. Guide: docker/README.md.

Road objects

Person, vehicle, two-wheeler

Included in the ANPR license. Separate package marearts-road-objects.

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")
# [{'class': 'car', 'conf': 0.97, 'ltrb': [...]}, {'class': 'person', ...}]

Support

License and next steps

1. The ANPR SDK license (SDK + MASS) is separate from ANPR Mobile and is not issued yet. It opens with the next SDK release.
2. pip install marearts-anpr then ma-anpr config.
3. Call the SDK, or ma-anpr server start.

marearts-anpr 3.9.1 · Python 3.10–3.14