YOLO RB3Gen2 Trainer

Instruction set for running powerful and efficient YOLOv8-based object detection training pipeline on RB3Gen2, optimized for custom dataset training with advanced configuration options.

Some useful links

https://developer.android.com/tools/releases/platform-tools

https://docs.qualcomm.com/bundle/publicresource/topics/80-70014-253/demo_app.html

🚀 Features

• Yolov8 Setup for RB3Gen2
  • Quantization and Compiliation
• YOLOv8 model support with multiple architecture options (nano to extra-large)
• Configurable training parameters through YAML configuration
• Support for both CPU and CUDA training
• Advanced training optimizations including:
  • Cosine learning rate scheduling
  • Mixed precision training
  • Rectangular training for memory efficiency
  • Mosaic augmentation
  • Label smoothing
  • Dropout regularization

📋 Prerequisites

• Google Account
• RB3Gen2

🛠️ Installation

1. Open .ipynb and click the link "Open in Collab"
2. Make sure you're using a T4 GPU Runtime

⚙️ Configuration

The project uses a config.yaml file for all training parameters. Key configuration sections include:

• Model Configuration: Choose YOLOv8 model size and pretrained options
• Dataset Configuration: Set training/validation data paths and number of classes
• Training Configuration: Fine-tune training parameters like epochs, batch size, etc.

Example configuration:

model:
  type: 'yolov8n'  # Options: yolov8n, yolov8s, yolov8m, yolov8l, yolov8x
  pretrained: true

data:
  train: 'data/train'
  val: 'data/val'
  nc: 1  # Number of classes

📁 Project Structure

yolo-rb3gen2-trainer/
├── config.yaml           # Training configuration
├── requirements.txt      # Project dependencies
├── training_images/      # Training dataset directory
└── base_image.jpg       # Base image for reference

🎯 Usage

1. Prepare your dataset in the training_images directory
2. Update the config.yaml file with your specific requirements
3. Open the ipynb in collab and run it then follow the instructions in the Jupyter Notebook

📊 Training Parameters

The training pipeline supports various parameters for YOLOv8 that can be configured in config.yaml:

• epochs: Number of training epochs
• batch_size: Batch size for training
• imgsz: Input image size
• device: Training device (cuda/cpu)
• optimizer: Choice of optimizer (SGD/Adam)
• And many more advanced options

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• Ultralytics YOLOv8
• AI Hub Qualcomm
• PyTorch
• OpenCV
• And all other open-source libraries used in this project