Fix ap calculation bug #8464

[UnglvKitDe]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Refinement of validation stats recording in YOLOv5's evaluation script.

📊 Key Changes

• Altered the structure of the stats list to include label class indices.

🎯 Purpose & Impact

• 🎩 This change enables more detailed analysis of performance by allowing evaluators to see not only if a prediction was correct, but which class it was correct for.
• 🔍 It can improve debugging and enhance the granularity of model evaluation, potentially impacting users by providing deeper insights into model performance per class.
• 👍 Users making use of the validation outputs for detailed analysis will benefit from more informative statistics.

[UnglvKitDe]

Before:

After:

[glenn-jocher]

@UnglvKitDe got it, thanks for the PR! So with the default settings it appears results are unchanged correct?

Seems good to merge.

[UnglvKitDe]

@UnglvKitDe got it, thanks for the PR! So with the default settings it appears results are unchanged correct?

Seems good to merge.

@glenn-jocher In this case, it does not look like that. In the general case, it leads to a "reduction" of the AP because you have more total positives, but so the AP calculation would be wrong. But it is generally very rare that no prediction is made for an image with a label.

[glenn-jocher]

@UnglvKitDe yes I understand. PR is merged. Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐