Fix list paths

[Hatovix]

I noticed earlier today that check_dataset function was only expecting single string from dict['train'] and dict['val'], generating errors while providing a list of paths.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhanced dataset auto-download and improved platform compatibility in general utilities.

📊 Key Changes

• 🛠️ Changed the way the Python platform module is imported and used.
• 🐛 Fixed the check_git_status function to use platform.system() for better OS detection.
• 🧹 Cleaned up the comment wording for the check_file function.
• 💾 Updated the check_dataset function, adding support for MacOS curl and refining the auto-download script logic.

🎯 Purpose & Impact

• 🚀 Ensure that the utility functions are more reliable across different operating systems.
• 📈 Improve automatic dataset downloading, making it easier for users to start projects without manual data setup.
• 🍏 Provide a specific solution for MacOS users to avoid certificate errors during the auto-download process.
• 👥 These updates aim to streamline user experience and lower entry barriers for new users trying to utilize YOLOv5 for their projects.

[github-actions]

Hello @Hatovix, thank you for submitting a PR! To allow your work to be integrated as seamlessly as possible, we advise you to:

• Verify your PR is up-to-date with origin/master. If your PR is behind origin/master update by running the following, replacing 'feature' with the name of your local branch:

git remote add upstream https://github.com/ultralytics/yolov5.git
git fetch upstream
git checkout feature  # <----- replace 'feature' with local branch name
git rebase upstream/master
git push -u origin -f

• Verify all Continuous Integration (CI) checks are passing.
• Reduce changes to the absolute minimum required for your bug fix or feature addition. "It is not daily increase but daily decrease, hack away the unessential. The closer to the source, the less wastage there is." -Bruce Lee

[glenn-jocher]

@Hatovix ah, thank you, good catch! I will look through this.

[glenn-jocher]

@Hatovix maybe we should just flip the if statements so it only checks datasets with download keys.
if 'download' in dict:

I assume only the official datasets would have download keys, and they are all single-source datasets (i.e. train and val strings).

[Hatovix]

You are probably right about the download keys, provided by public datasets.

Another think about this function (check_dataset), in my case of use:

• custom dataset
• list of paths
• no download key

A nonexistant path displays the associated warning, but the training only stop later, after loading previous existing paths from the list, and fails on the first problematic one, (just noticed that Exception('Dataset autodownload unavailable.') probably needs raise statement).

I think it is better to check that everything is ready to be loaded, since loading can be time consuming.

[glenn-jocher]

Just realized if newcomers use the coco128.yaml template for their own dataset, they may leave the download key there with an empty string maybe download: '', or just blank download: , without knowing they can also just delete it.

Perhaps this function may need a bit of a broader rethink.

[glenn-jocher]

At the end of the day the desired outcome is simply that the official datasets autodownload if not found locally, the details of the function aren't too important, but I wanted to create a structure that allows people to implement on their own custom datasets a s well, should they choose to do so.

[glenn-jocher]

This should be more robust than before. I removed checking of the train paths, as some use cases involve running test.py on coco val without actually requiring you to download coco train. Also added an OS check, as I'm getting a certificate error on MacOS locally with torch download.

[glenn-jocher]

CI passing, merging. Thank you for your contributionS @Hatovix.