Python SDK · 8 classes · pip

MareArts Road Objects

Detect person, bicycle, motorcycle, car, bus, truck, traffic light, and stop sign. Same license as the ANPR SDK. CPU, CUDA, and DirectML.

PyPI package marearts-road-objects · Python 3.9–3.12 · current release 2.0.6.

Install

pip, then GPU extra if you have one

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

# NVIDIA CUDA
pip install "marearts-road-objects[gpu]"

# Windows AMD / Intel / NVIDIA
pip install "marearts-road-objects[directml]"

ma-robj config
ma-robj validate

Credentials go to ~/.marearts/.marearts_env (same files as ANPR), or set MAREARTS_ANPR_USERNAME, MAREARTS_ANPR_SERIAL_KEY, and MAREARTS_ANPR_SIGNATURE.

Aliases: ma-robj, marearts-robj, marearts-road-objects.

Python

Load once, then call detector

import cv2
from marearts_road_objects import ma_road_object_detector

detector = ma_road_object_detector(
    model_name="small_fp32",   # small_fp32, medium_fp32, large_fp32
    user_name=user_name,
    serial_key=serial_key,
    signature=signature,
    backend="auto",            # auto, cuda, directml, cpu
    conf_thres=0.5,
    iou_thres=0.5,
)

image = cv2.imread("traffic.jpg")
result = detector.detector(image)
print(result)
# {'results': [{'ltrb': [...], 'ltrb_conf': 85, 'class_id': 3, 'class': 'car'}, ...],
#  'ltrb_proc_sec': 0.08}

The call is detector.detector(image) on a BGR array. There is no .detect() and the model names are not ANPR’s 640p_fp32.

{
    'results': [
        {
            'ltrb': [88.1, 421.0, 164.9, 476.2],
            'ltrb_conf': 85,
            'class_id': 3,
            'class': 'car'
        }
    ],
    'ltrb_proc_sec': 0.178
}

Models

small, medium, large

Times from the GitHub README, CUDA on an NVIDIA GeForce RTX.

First run downloads the model. Production: export MAREARTS_ROBJ_SKIP_UPDATE=1 skips the update check. Verbose: export MAREARTS_VERBOSE=1.

Classes

Eight labels

CLI

ma-robj

ma-robj config
ma-robj validate
ma-robj detect traffic.jpg
ma-robj detect traffic.jpg -m medium_fp32 -b cuda
ma-robj gpu-info
ma-robj version

ANPR

One serial

The ANPR SDK license unlocks this package. Plate reading is a separate install: pip install marearts-anpr.

Intro: ANPR SDK · Mobile app.

Support

Start

1. Road Objects uses the ANPR SDK license (SDK + MASS), which is separate from ANPR Mobile and is not issued yet. It opens with the next SDK release.
2. pip install marearts-road-objects then ma-robj config.
3. Try the live demo in the browser.

marearts-road-objects 2.0.6 · Python 3.9–3.12