Added user option to specify dataset download path

[kalenmike]

Added the option for the user to pass the --sandbox='custom/path/to' argument which ensures YOLOv5 unzips the dataset into this custom folder. The --project path is overwritten to match what is passed into --sandbox this ensures both the dataset and runs are kept in the same folder.

The idea is that this becomes a sandbox option and all training-specific files are kept in the same user-defined folder to ensure that runs do not mix or overwrite each other.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhanced dataset handling and configuration flexibility in YOLOv5 training.

📊 Key Changes

• Commented out the Pillow requirement in requirements.txt.
• Added a --sandbox flag to train.py for alternative file storage, affecting dataset checks and path settings.
• Simplified image and label verification logic in utils/datasets.py, making error processes more transparent.
• Updated check_dataset function in utils/general.py to handle custom dataset root paths via the --sandbox flag and clarified autodownload prompts.

🎯 Purpose & Impact

• 🛠 Removing the strict dependency on the Pillow library could help in environments where this library is problematic or conflicts with other packages.
• 🗃 The new --sandbox option allows all training-related files, including datasets, to be saved in a specified directory, streamlining project organization and improving user flexibility.
• 📈 Simplification of the verification process in dataset utilities enhances code maintainability and could lead to fewer silent failures during dataset loading.
• 🔄 The updates to check_dataset support clearer dataset download feedback and better management of dataset paths, potentially making YOLOv5 more robust in various usage scenarios.

[github-actions]

This pull request has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions YOLOv5 🚀 and Vision AI ⭐.