This is a full-stack application for real-time object detection using YOLO-World model with React Native frontend and FastAPI backend.

• Real-time object detection using YOLO-World model
• Camera integration for live photo capture
• Gallery image selection support
• Configurable detection classes
• Confidence threshold adjustment

• Interactive image labeling interface
• Drag-and-drop bounding box creation
• Support for existing and new class labels
• YOLO format data storage for training

• Automatic model improvement with labeled data
• Training data management and statistics
• Custom class integration
• Model retraining capabilities

• Frontend: React Native with Expo
• Backend: FastAPI with YOLO-World
• Training: YOLOv8 fine-tuning pipeline

• Node.js 18+
• Python 3.8+
• pnpm (recommended) or npm

make setup

# Start both frontend and backend
make dev

# Or start individually
make server    # Backend only
make frontend  # Frontend only

1. Capture or select an image
2. Click "Manual Labeling" button
3. Draw bounding boxes by dragging on the image
4. Assign labels to each bounding box
5. Submit the labeled data

# Start model fine-tuning
make train-model

# View training statistics
make view-training-stats

# Clean training data
make clean-training-data

• POST /labeling/submit - Submit labeled training data
• GET /training/data/stats - Get training dataset statistics
• POST /training/start - Start model fine-tuning

• GET /model/classes - Get current detection classes
• POST /model/classes - Add new detection classes
• POST /detect - Detect objects in image
• POST /detect/with-confidence - Detect with custom confidence

The application stores training data in YOLO format:

• training_data/images/ - Training images
• training_data/labels/ - YOLO format annotations
• training_data/classes.txt - Class name mappings
• training_data/data.yaml - Training configuration

1. Initial Detection: Try automatic detection with existing classes
2. Add Classes: If objects aren't detected, add new classes
3. Manual Labeling: For persistent detection issues, create manual labels
4. Model Training: Periodically retrain the model with accumulated labels
5. Improved Detection: Enjoy better detection accuracy

make test

Access interactive API docs at: http://localhost:8000/docs