Add hyp.scratch-med.yaml

[glenn-jocher]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Introducing a new set of hyperparameters tailored for medium-augmentation COCO training from scratch.

📊 Key Changes

• 🆕 Added hyp.scratch-med.yaml, a configuration file with medium augmentation hyperparameters.
• ⚙️ Includes settings for learning rate, momentum, weight decay, warmup strategies, loss gains, IoU threshold, and various data augmentations.

🎯 Purpose & Impact

• 💡 The purpose is to provide a starting point for users wanting to train models on the COCO dataset with an intermediate level of data augmentation.
• 🚀 Potential impact includes improved model accuracy and performance due to better-tuned hyperparameters for this specific training scenario.
• 🌐 This could benefit users by reducing the time and effort required to find the right balance of hyperparameters for their training needs.