MareArts ANPR SDK

Github page: https://github.com/MareArts/MareArts-ANPR

Automatic Number Plate Recognition (ANPR) SDK for multiple regions with GPU acceleration support.

💎 One License, All Access: SDK + Mobile App + Road Objects Detection - Use everywhere with a single license.

🎉 MareArts SDK v3.8.3

V15 OCR - Next Generation Recognition ⭐

🎯 +6.7–7.4% Better Accuracy: V15 OCR achieves 98–99% exact match across all regions
🌎 North America +26%: Massive improvement from 70% → 97% for NA plates
📝 Better Multi-line Handling: Improved recognition of multi-line license plates
💾 INT8 Support: 75% smaller model files for mobile and embedded devices
⚡ 235–280 FPS (GPU): Strong real-time throughput across all model sizes
✅ Default for New Projects: V15 OCR now recommended (V14 still fully supported)

🔄 Easy Upgrade: Simple drop-in replacement - change ma_anpr_ocr_v14 to ma_anpr_ocr_v15

📊 See detailed V14 vs V15 comparison →

🎉 MareArts ANPR Mobile App!

📱 Now available on iOS!

Experience the power of MareArts ANPR directly on your mobile device! Fast, accurate, on-device license plate recognition for parking management, security, and vehicle tracking.

Features

🌍 Multi-Region Support: Korean, Europe+, North America, China, and Universal license plates
🔄 Dynamic Region Switching: Change regions instantly with set_region() without model reload
⚡ GPU Acceleration: CUDA, DirectML support for real-time processing
🎯 High Accuracy: Advanced models with regional vocabulary optimization
📦 Batch Processing: Process multiple plates simultaneously
🐳 Production Ready: Docker API with smart model caching and multi-architecture support

Quick Start

Installation

# CPU Installation
pip install marearts-anpr

# GPU Installation (CUDA, DirectML)
pip install marearts-anpr[gpu]        # NVIDIA CUDA
pip install marearts-anpr[directml]   # Windows GPU (AMD/Intel/NVIDIA)

📦 See complete installation guide

Basic Usage

💡 Model names: See models and benchmarks (e.g., micro_320p_fp32, medium_640p_fp32)

from marearts_anpr import ma_anpr_detector_v14, ma_anpr_ocr_v15, marearts_anpr_from_image_file

# Initialize detector (320p models 2× faster!)
detector = ma_anpr_detector_v14(
    "micro_320p_fp32",  # Fast: micro_320p_fp32 | Accurate: micro_640p_fp32
    user_name,
    serial_key,
    signature,
    backend="cuda",  # cpu, cuda, directml
    conf_thres=0.25,  # Detection confidence threshold (0.0-1.0)
    iou_thres=0.5     # IoU threshold for NMS (0.0-1.0)
)

# Initialize V15 OCR (Latest - Better accuracy, faster)
ocr = ma_anpr_ocr_v15(
    "large_fp32",       # FP32: pico/micro/small/medium/large | INT8: smaller files
    "univ",             # Region: kr, eup, na, cn, univ
    user_name,
    serial_key,
    signature
)

# Process image
result = marearts_anpr_from_image_file(detector, ocr, "image.jpg")
print(result)
# Output: {'results': [{'ocr': 'ABC123', 'ocr_conf': 99, ...}], ...}

💡 V15 OCR is recommended (+6.7–7.4% better accuracy, 98–99% exact match). V14 OCR available for backward compatibility: ma_anpr_ocr_v14 | Compare V14 vs V15 →

Dynamic Region Switching

Switch regions without reinitialization:

ocr.set_region('kr')    # Korean
ocr.set_region('eup')   # Europe+
ocr.set_region('na')    # North America
ocr.set_region('cn')    # China
ocr.set_region('univ')  # Universal (all regions)

🔄 Learn more about dynamic region switching

Multiple Input Formats & CLI

From different image sources:

import cv2
from PIL import Image
from marearts_anpr import marearts_anpr_from_cv2, marearts_anpr_from_pil

result = marearts_anpr_from_cv2(detector, ocr, cv2.imread("image.jpg"))
result = marearts_anpr_from_pil(detector, ocr, Image.open("image.jpg"))
result = marearts_anpr_from_image_file(detector, ocr, "image.jpg")

CLI commands:

ma-anpr image.jpg                    # Process image with V15 OCR
ma-anpr test-api image.jpg           # Free test API (1000/day, uses V15 OCR)
ma-anpr validate                     # Validate license
ma-anpr models                       # List all available models

🔧 See complete usage examples and CLI reference

Model Performance

Detector Performance

Model Name	Detection Rate	Speed (GPU)	Notes
micro_320p_fp32	97.13%	128 FPS (7.8ms)	🏆 Best overall
micro_320p_fp16	97.13%	56 FPS (17.9ms)	🏆 Best mobile (50% smaller)
micro_640p_fp32	98.99%	68 FPS (14.6ms)	Highest detection rate
small_320p_fp32	98.00%	142 FPS (7.0ms)	⚡ Fastest
medium_320p_fp32	98.06%	136 FPS (7.4ms)	High detection
large_320p_fp32	98.40%	131 FPS (7.6ms)	Strong performance
pico_320p_fp32	96.02%	129 FPS (7.8ms)	📱 Smallest + fast

Key Findings:

320p models: 2× faster than 640p with excellent detection rates (96-98%)
640p models: Highest detection rates (98-99%) for difficult cases
FP16 models: 50% smaller size, same detection rate, ~50% slower

V15 OCR Performance (Latest - Recommended)

V15 OCR - Average across all regions | FP32 & INT8 Models Available

Model	Exact Match	Char Accuracy	FPS	Notes
pico_fp32	98.66%	99.74%	235	📱 Edge/Mobile
micro_fp32	99.01%	99.80%	246	Fast + accurate
small_fp32	98.66%	99.75%	280	⚡ Fastest
medium_fp32	99.13%	99.83%	254	🎯 Best accuracy
large_fp32	98.99%	99.81%	241	High accuracy

INT8 models: 75% smaller files with similar accuracy (pico_int8, micro_int8, small_int8, medium_int8, large_int8)

V15 OCR Accuracy by Region (medium_fp32)

Region	Exact Match	Char Accuracy	FPS
🇨🇳 China	99.91%	99.99%	257
🇰🇷 Korean	99.56%	99.91%	251
🌍 Universal	99.29%	99.83%	251
🇪🇺 Europe+	98.67%	99.70%	258
🇺🇸 North America	98.22%	99.74%	256

Supported Regions: Korean (kr), Europe+ (eup), North America (na), China (cn), Universal (univ)

✅ V15 Benefits: +6.7–7.4% accuracy over V14, 235–280 FPS | 💾 INT8 Models: 75% smaller for mobile/embedded

📊 V14 vs V15 detailed comparison | Regional benchmarks

V14 vs V15 OCR Quick Comparison

	V14 OCR	V15 OCR ⭐
Accuracy (Average)	91-92%	98-99% (+6.7–7.4%)
Speed	265-302 FPS	235-280 FPS
INT8 Support	❌ Not available	✅ Yes (smaller files)
Multi-line Plates	Good	✅ Improved
Recommendation	Legacy/Existing projects	✅ Default (new projects)

Bottom Line: V15 is significantly more accurate (+6.7–7.4%) with strong real-time throughput, and supports INT8. Use V15 for new projects. V14 remains available for backward compatibility.

Regional Support

MareArts ANPR supports license plates from multiple regions with specialized vocabulary optimization:

🇰🇷 Korean (kr) - Korean license plates with Hangul characters (best accuracy: 99.56%)
🇪🇺 Europe+ (eup) - EU countries + Albania, Andorra, Bosnia & Herzegovina, Indonesia, and more
🇺🇸 🇨🇦 🇲🇽 North America (na) - USA, Canada, and Mexico license plates
🇨🇳 China (cn) - Chinese license plates with province codes
🌍 Universal (univ) - All regions (default, but choose specific region for best accuracy)

💡 Dynamic Region Switching: Use ocr.set_region('kr') to switch regions instantly without reloading the model, saving ~180 MB per additional region.

🌍 See complete regional support and character sets

🌐 NEW: ANPR Management Server

Professional server with REST API and Web Dashboard

Deploy a complete ANPR management system with real-time monitoring, detection history, and visual analytics.

Management Server Dashboard

Quick Start:

cd management_server
pip install -r requirements.txt
ma-anpr config
python server.py
# Open http://localhost:8000/

Features: REST API, Web Dashboard, Real-time logs, SQLite database, Live model switching

📖 Get Started with Management Server →

Documentation

📦 Installation Guide - Detailed installation options and requirements
🔧 Usage Examples - Python SDK, CLI usage, dynamic region switching, and environment variables
💻 Example Code - Basic, advanced, and batch processing examples
🚀 Model Versions - Available models, benchmarks, and performance metrics
🌍 Regional Support - Supported countries and character sets
🐳 Docker Deployment - Container setup, API server, and multi-architecture builds
🧪 Try ANPR - Test our ANPR without license (1000 requests/day)
❓ FAQ - Licensing, regions, features, and troubleshooting

MareArts Ecosystem

Explore our AI toolkit:

marearts-anpr - Automatic Number Plate Recognition (GitHub)
marearts-road-objects - Road object detection for persons, vehicles, and 2-wheelers (GitHub)
marearts-xcolor - Color extraction and similarity analysis (GitHub)
marearts-mast - Real-time panoramic stitching (GitHub)
marearts-crystal - Encryption and decryption toolkit (PyPI)

Support & Resources

Resource	Link
📧 Contact	hello@marearts.com
🏠 Homepage	https://marearts.com
💳 License Purchase	ANPR Solution
🎮 Live Demo	http://live.marearts.com
📺 Video Examples	YouTube Playlist
📱 iOS App Store	MareArts ANPR App

License

This software requires a valid license key. Visit MareArts ANPR Solution for licensing options.

Country/region

MareArts ANPR SDK

MareArts ANPR SDK

🎉 MareArts SDK v3.8.3

🎉 MareArts ANPR Mobile App!

Features

Quick Start

Installation

Basic Usage

Dynamic Region Switching

Multiple Input Formats & CLI

Model Performance

Detector Performance

V15 OCR Performance (Latest - Recommended)

V15 OCR Accuracy by Region (medium_fp32)

V14 vs V15 OCR Quick Comparison

Regional Support

🌐 NEW: ANPR Management Server

Documentation

MareArts Ecosystem

Support & Resources

License