Evolutionary Training Manager for YOLO Models

An evolutionary algorithm-based training system for YOLO object detection models. This project automatically optimises both model architecture and training parameters to achieve the best trade-off between accuracy and inference speed.

Features

• Evolutionary optimisation of YOLO model architecture and hyperparameters
• Hardware-aware training optimised for both NVIDIA (CUDA) and AMD (ROCm) GPUs
• Automatic resource management to prevent VRAM exhaustion
• Comprehensive checkpointing system for resuming training
• Population-based approach with configurable selection, crossover, and mutation
• Built on top of the TrainingAutomation framework for YOLO model training

Requirements

• Python 3.8 or higher
• PyTorch 1.7 or higher
• GPU with CUDA or ROCm support (optional but recommended)

Installation

1. Clone the repository:

   git clone git@github.com:SethBennett2523/EvolutionaryTrainingManager.git
   cd EvolutionaryTrainingManager

2. Install dependencies:

   pip install -r requirements.txt

3. Update the configuration in config.yaml to match your environment.

Usage

Basic usage

python main.py --config config.yaml

Additional options

• --resume: Resume training from the last checkpoint
• --output-dir OUTPUT_DIR: Specify output directory
• --generations N: Set maximum number of generations
• --population N: Set population size
• --device {cuda,rocm,cpu,auto}: Specify device to use
• --debug: Enable debug logging

Configuration

The main configuration file (config.yaml) includes settings for:

• Evolution parameters (population size, mutation rate, etc.)
• Model parameters (base model type)
• Hardware settings (device, memory threshold)
• Hyperparameter ranges for evolution

How It Works

1. The algorithm initialises a population of YOLO models with random hyperparameters and architecture settings
2. Each model is trained on the dataset and evaluated for accuracy (mAP) and speed (inference time)