Segment prediction labels normalization fix

[glenn-jocher]

Segment prediction labels normalization fix for #10196

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improvement of segmentation output normalization in YOLOv5's prediction pipeline.

📊 Key Changes

• Added normalize boolean parameter to the scale_segments function.
• Implemented normalization of segment coordinates within the scale_segments function when normalize=True is passed.
• Normalization adjusts the segment dimensions relative to the image width and height, rather than absolute pixel values.

🎯 Purpose & Impact

• Purpose: To provide an option for normalizing segment coordinates, making them relative to the image size (values between 0 and 1) instead of absolute pixel coordinates.
• Impact: This can ease post-processing and is beneficial for tasks that require consistent scale-invariant representation like multi-scale training or when integrating with systems that expect normalized coordinates.
• Potential Benefit to Users: Users now have the flexibility to choose between normalized or non-normalized segment outputs based on their specific use case or downstream application requirements.

[glenn-jocher]

@AyushExel

[AyushExel]

@glenn-jocher okay

[glenn-jocher]

@AyushExel ok all done. Note I had to swap the axes from the first commit. Not sure if we have this in our new work yet or not.