Ultralytics YOLO

The realvirtual.io AI Builder includes built-in integration with Ultralytics YOLO (You Only Look Once) for state-of-the-art object detection capabilities. While YOLO provides fast and accurate real-time object detection for industrial automation scenarios, it represents just one option among many training frameworks that can use AI Builder’s synthetic data.

Overview

YOLO is a popular real-time object detection system that can detect multiple objects in images with high speed and accuracy. While realvirtual.io AI Builder includes built-in YOLO integration for convenience, the synthetic training data generated by AI Builder can be used with various machine learning frameworks including TensorFlow, PyTorch, Detectron2, and other training pipelines.

The built-in YOLO integration allows you to:

• Train custom YOLO models using synthetic training data
• Deploy trained models for real-time object detection
• Test models within digital twin environments
• Export models in ONNX format for cross-platform deployment

For alternative training approaches, the synthetic data can be exported in standard formats for use with your preferred AI framework.

GitHub Repository

The YOLO integration is available as a separate repository that works with realvirtual.io AI Builder:

🔗 YOLO Integration for realvirtual.io AI Builder

Licensing Information

Caution

Important Licensing Notice

This repository includes example pipelines that are licensed under AGPL-3.0.

Ultralytics YOLO is also licensed under AGPL-3.0. If you plan to use this integration for commercial purposes, please review the Ultralytics license conditions carefully.

For commercial licensing options, please contact Ultralytics directly or review their licensing terms at https://ultralytics.com/license.

Key Features

YOLO Model Variants

• YOLOv8 Nano: Fastest inference, smallest model size
• YOLOv8 Small: Balanced speed and accuracy
• YOLOv8 Medium: Higher accuracy with moderate speed
• YOLOv8 Large: Best accuracy, larger model size
• YOLOv8 Extra Large: Maximum accuracy for complex scenarios

Training Capabilities

• Custom Dataset Training: Train on your specific industrial objects
• Transfer Learning: Fine-tune pre-trained models
• Synthetic Data Integration: Use AI Builder generated training data
• Multi-class Detection: Detect multiple object types simultaneously

Export Options

• ONNX Format: Cross-platform deployment
• TensorRT: NVIDIA GPU acceleration
• OpenVINO: Intel hardware optimization
• CoreML: Apple device deployment

Integration Workflow

1. Generate Training Data: Use realvirtual.io AI Builder to create synthetic training datasets
2. Choose Training Approach:
   • Use built-in YOLO integration for immediate training
   • Export data for use with external training frameworks
3. Configure YOLO (if using built-in integration): Set up YOLO training parameters and model selection
4. Train Model: Execute training using the YOLO integration or your preferred framework
5. Test in Digital Twin: Validate model performance in virtual environment
6. Deploy: Export trained model for production use

Installation

The Ultralytics YOLO integration is automatically handled by AI Builder:

Automatic Installation

1. Open AI Builder Scene: Load your AI Builder scene in Unity
2. Navigate to Training Component: Select the AI Training component in the Inspector
3. Install Pipeline: When you see “Ultralytics training pipeline not found. Please install the package.”, click the “Install Ultralytics Pipeline” button
4. Package Manager: The Ultralytics dependency will be automatically installed and will appear in Unity’s Package Manager
5. Ready to Use: Once installed, the YOLO training capabilities will be available

Manual Repository Access

For advanced users or custom implementations, the YOLO integration source is available at:

• Repository: https://github.com/game4automation/io.realvirtual.aibuilder-ultralytics

Getting Started

To use the YOLO integration:

1. Install the pipeline using the automatic installation method above
2. Ensure you comply with the AGPL-3.0 licensing requirements for your use case
3. Configure your training parameters in the AI Training component
4. Start training with your synthetic training data

See Also

• Generate AI Training Data - Creating synthetic datasets
• Testing AI in a Digital Twin - Model validation
• Deploying the AI - Production deployment
• AI Training - General training information