YOLOBridge: Intelligent Object Detection Backend for Label Studio

Description

YOLOBridge is a sophisticated machine learning backend that connects YOLO's powerful object detection capabilities with Label Studio's annotation platform. It creates an intelligent feedback loop between annotation and model training, enabling semi-automated labeling workflows and continuous model improvement.

The system provides a complete pipeline for training, versioning, and deploying YOLO-based object detection models directly within the Label Studio ecosystem.

Key Features

• 🧠 Smart Model Registry - Automatic version control with comprehensive tracking of model lineage, metrics, and artifacts
• 🔄 Continuous Learning Loop - Seamless integration between user annotations and model training
• 🎯 State-of-the-Art Detection - Leveraging the latest YOLOv8 architectures for accurate object detection
• 📊 Intelligent Data Management - Automated train/validation splits with consistent hashing for reproducibility
• 🚀 Production-Ready Deployment - Docker support for easy scaling in production environments
• 📝 Extensive Logging - Comprehensive logging system for debugging and performance monitoring
• ⚙️ Flexible Configuration - Easily customizable settings for both inference and training
• 🔍 Real-Time Inference - Fast prediction capabilities for interactive labeling sessions

Requirements

• Python 3.8+
• Label Studio (Now only support 1.10.x)
• Docker (optional, but recommended)
• CUDA-compatible GPU (recommended for training)

Installation

Using Docker (Recommended)

docker-compose up -d

Manual Installation

1. Install requirements:

pip install -r requirements.txt

2. Set environment variables:

export LABEL_STUDIO_HOST=<your-label-studio-host>
export LABEL_STUDIO_API_KEY=<your-api-key>

Project Structure

├── yolo.py              # Main implementation with ML Backend classes
├── yolo_config.py       # Configuration settings for model and training
├── _wsgi.py             # WSGI entry point
├── requirements.txt     # Python dependencies
├── Dockerfile           # Docker configuration
└── docker-compose.yml   # Docker Compose configuration

Configuration

The system can be easily configured through yolo_config.py. Key settings include:

Model Configuration

# YOLO model version selection (yolov8s, yolov8m, yolov8l, yolov8x)
yolo_model_version = 'yolov8x'

# Whether to retrain from scratch or continue from previous version
fine_tune_start_from_original = False

Prediction Configuration

model_predict_config = dict(
    conf=0.3,  # Confidence threshold
    iou=0.1    # IoU threshold
)

Training Configuration

model_train_config = dict(
    data='datasets/data.yaml',
    epochs=1000,
    patience=20,
    plots=True,
    exist_ok=True,
    verbose=True
)

Usage

1. Start Label Studio and create a project with object detection configuration

2. Configure the ML Backend in Label Studio settings:

   • Add ML Backend URL (e.g., http://localhost:9090)
   • Enable "Use for interactive preannotations"
   • Enable "Use for autoannotation"

3. Start annotating and experience the intelligent workflow:

   • The system automatically trains on your annotations
   • Provides predictions for new images
   • Tracks model versions and performance metrics

API Integration

YOLOBridge implements the Label Studio ML Backend API with three core endpoints:

• /predict - For real-time model predictions
• /train - For model training and fine-tuning
• /health - For health checks and system status

Model Registry

The system maintains a comprehensive model registry that tracks:

• Model versions and lineage
• Training configurations
• Performance metrics (mAP, precision, recall)
• Model status (initializing, training, completed, failed)
• Weights file locations

License

MIT

Support

For issues and feature requests, please create an issue in the repository.