Avoid out-of-image class labels

[zldrobit]

Avoid out-of-image class labels in detection.
Without this PR, labels may be out of the image.

In this PR, labels are contained within the image.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improved label positioning for object detection annotations in Ultralytics' YOLOv5.

📊 Key Changes

• Removed redundant font height calculation for drawing text.
• Adjusted label positioning logic to allow labels to be drawn inside or outside of bounding boxes, depending on available space.
• Unified label positioning style across both PIL and OpenCV to enhance consistency.

🎯 Purpose & Impact

• Enhanced Readability: Better placement of labels ensures they do not overlap with other UI elements or go off-screen, enhancing readability.
• Increased Robustness: By checking if the label fits outside the box before deciding its position, the change provides smarter annotation that could minimize occlusion on small objects or images with limited space.
• Visual Consistency Across Backends: Whether using PIL or OpenCV for drawing, labels will now follow the same positioning rules, providing a consistent visual experience for end-users.

[glenn-jocher]

@zldrobit PR is merged. Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐

[glenn-jocher]

@zldrobit just tested, PR works well!