mindar 0.1.3

High-performance MindAR implementation for real-time image recognition on edge devices. Developed by FANSEE LAB.

MindAR Python

High-performance MindAR implementation for real-time image recognition on edge devices.

Features

• MindAR Compatible: Full compatibility with MindAR .mind file format
• High Performance: Optimized with numba JIT compilation for edge devices
• Real-time Detection: Efficient feature detection and matching
• Modern Architecture: Clean configuration-based API with dataclasses
• Type Safe: Full type hints and proper error handling
• Production Ready: Comprehensive testing and linting (pylint score 7.5+)

Installation

From PyPI (Recommended)

pip install mindar

From Source

git clone https://github.com/FANSEE-LAB/mind-ar.git
cd mind-ar
pip install -e .

Requirements

• Python >= 3.9 (required for numba optimization)
• OpenCV
• NumPy
• msgpack (for .mind file format)
• numba (for performance optimization)

Usage

Basic Detection

import cv2
from mindar import Detector, Matcher, MindARCompiler
from mindar.types import DetectorConfig, MatcherConfig

# Configure detector with new configuration system
detector_config = DetectorConfig(
    method="super_hybrid",
    max_features=1000,
    debug_mode=False
)
detector = Detector(detector_config)

# Configure matcher
matcher_config = MatcherConfig(
    ratio_threshold=0.75,
    min_matches=8,
    debug_mode=False
)
matcher = Matcher(matcher_config)

# Detect features in image
image = cv2.imread("target.jpg", cv2.IMREAD_GRAYSCALE)
result = detector.detect(image)
feature_points = result["feature_points"]

print(f"Detected {len(feature_points)} features")

Compile .mind Files

from mindar.compiler import MindARCompiler

# Initialize compiler with debug mode
compiler = MindARCompiler(debug_mode=True)

# Compile images to .mind file
success = compiler.compile_directory("./images", "./targets.mind")
if success:
    print("✅ Compilation successful")

# Load compiled targets
mind_data = compiler.load_mind_file("./targets.mind")
print(f"Loaded {len(mind_data['dataList'])} targets")

Performance

Optimized for edge devices like Raspberry Pi 4:

• Detection: ~50ms per frame (640x480)
• Matching: ~20ms per target
• Memory: <100MB usage

License

MIT License - Compatible with original MindAR project