MareArts Road Objects Detector

🚗 MareArts Road Objects Detection

High-performance road object detection with 8-class support: person, bicycle, motorcycle, car, bus, truck, traffic_light, stop_sign. Built with advanced deep learning for accurate real-time detection.

✨ Features

🎯 8-Class Detection: 👤 person, 🚲 bicycle, 🏍️ motorcycle, 🚗 car, 🚌 bus, 🚚 truck, 🚦 traffic_light, 🛑 stop_sign
⚡ GPU Acceleration: NVIDIA CUDA and DirectML support
🛠️ CLI Interface: Easy command-line tools (ma-robj, marearts-robj)
🌐 Cross-platform: Windows (x64/ARM64), macOS (Intel/Apple Silicon), Linux (x64/ARM64)
🔑 Unified License: Same license works for both MareArts-ANPR and Road Objects
🚀 Auto-download: Models download automatically on first use

🚀 Quick Start

Installation

# Basic installation (CPU)
pip install marearts-road-objects

# With GPU acceleration (recommended)
pip install marearts-road-objects[gpu]          # NVIDIA CUDA
pip install marearts-road-objects[directml]     # Windows DirectML
pip install marearts-road-objects[all-gpu]      # All GPU support

🔑 Get Your License

One license works for both ANPR and Road Objects packages!

Subscribe to MareArts ANPR/LPR Solution →

🐍 Python API

Simple Detection

import cv2
from marearts_road_objects import ma_road_object_detector

# License credentials - Option A: Hardcoded
username = "your-email@domain.com"
serial_key = "your-serial-key"
signature = "your-signature"

# Initialize detector ONCE (model downloads automatically on first use)
detector = ma_road_object_detector(
    model_name="small_fp32",  # Options: small_fp32, medium_fp32, large_fp32
    user_name=username,
    serial_key=serial_key,
    signature=signature,
    backend="auto",      # Options: "auto", "cuda", "directml", "cpu"
    conf_thres=0.5,      # Confidence threshold (0.0-1.0)
    iou_thres=0.5        # NMS IoU threshold (0.0-1.0)
)

# Use detector.detector() for inference (can be called repeatedly in loops)
image = cv2.imread("traffic.jpg")
result = detector.detector(image)

# Result format (JSON)
print(result)

📊 Model Comparison

Choose the right model for your needs:

Model Name	Model Size	Detection Speed	Accuracy
`small_fp32`	102 MB	8ms (125 FPS)	Standard
`medium_fp32`	195 MB	83ms (12 FPS)	Better
`large_fp32`	314 MB	96ms (10 FPS)	Best

Benchmark System:

GPU: NVIDIA GeForce RTX (CUDA)
CPU: Modern x86_64 processor
Backend: CUDA (GPU accelerated)

🛠️ CLI Usage

Configure License

# Interactive setup (recommended)
ma-robj config

# Or set environment variables
export MAREARTS_ANPR_USERNAME="your-email@domain.com"
export MAREARTS_ANPR_SERIAL_KEY="your-serial-key"
export MAREARTS_ANPR_SIGNATURE="your-signature"

# After configuration, source the environment:
source ~/.marearts/.marearts_env

Available Commands

ma-robj config               # Configure license credentials
ma-robj validate             # Validate license
ma-robj detect IMAGE         # Detect objects in image
ma-robj gpu-info             # Check GPU acceleration support
ma-robj version              # Show package version
ma-robj detect traffic.jpg   # Detection Example

Command Aliases: ma-robj, marearts-robj, marearts-road-objects (all work the same)

📊 Output Format

ANPR-compatible JSON format for easy integration:

{
    'results': [
        {
            'ltrb': [88.1, 421.0, 164.9, 476.2],  # Bounding box [left, top, right, bottom]
            'ltrb_conf': 85,                       # Confidence 0-100 (integer)
            'class_id': 3,                         # Class ID (0-7)
            'class': 'car'                         # Class name
        },
        {
            'ltrb': [201.3, 401.7, 265.1, 452.2],
            'ltrb_conf': 84,
            'class_id': 3,
            'class': 'car'
        }
    ],
    'ltrb_proc_sec': 0.178  # Processing time in seconds
}

🎯 Detection Classes

ID	Class	Description
0	person	Pedestrians and people
1	bicycle	Bicycles
2	motorcycle	Motorcycles and scooters
3	car	Passenger cars
4	bus	Buses
5	truck	Trucks and vans
6	traffic_light	Traffic signals
7	stop_sign	Stop signs

🔧 Environment Variables

# Skip model update checks for faster initialization (production)
export MAREARTS_ROBJ_SKIP_UPDATE=1

# Enable verbose logging for debugging
export MAREARTS_VERBOSE=1

🎯 Try Road Object Detector

Web Demo: https://live.marearts.com/?menu=road_objects

API Test (1000 free requests/day) - Just change the model name and image path:

curl -X POST "https://we303v9ck8.execute-api.eu-west-1.amazonaws.com/Prod/marearts_robj" \
  -H "Content-Type: image/jpeg" \
  -H "x-api-key: !23J4K9L2Wory34@G7T1Y8rt-PP83uSSvkV3Z6ioSTR!2" \
  -H "user-id: marearts_robj@public" \
  -H "model_name: small_fp32" \
  --data-binary "@your_image.jpg"

Only change: model_name (small_fp32/medium_fp32/large_fp32) and @your_image.jpg

🆘 Support

Live Test : https://live.marearts.com
License: Get your subscription
Issues: GitHub Issues
Email: hello@marearts.com

🔗 Related Packages

MareArts AI Ecosystem (same license for all):

marearts-anpr - License plate recognition (ANPR/ALPR)
marearts-crystal - Licensing framework
marearts-xcolor - Color space conversions

photo by Andrea Cau on Unsplash

Country/region