Detect() and Segment() fixes for CoreML and Paddle

[glenn-jocher]

Eliminates .clone() that was causing Paddle export issues and inplace .sigmoid_() that was causing CoreML export issues.

Signed-off-by: Glenn Jocher [email protected]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhancements in YOLOv5's output processing for object detection and segmentation.

📊 Key Changes

• Improved handling of bounding box (xy and wh) predictions and class confidence (conf) computations.
• Added support for a new segment type in the output processing, which now manages boxes and masks together.
• Code refactoring to cleanly separate the processing logic for Detect (boxes only) and Segment (boxes + masks) output types.

🎯 Purpose & Impact

• 🎨 The update provides a clearer separation of detection and segmentation logic, making the codebase easier to understand and maintain.
• 📈 It prepares YOLOv5 for better segmentation capabilities, potentially increasing its use in applications that require both object detection and segmentation, such as advanced computer vision tasks in autonomous driving or medical image analysis.
• 🚀 For users, this means improved functionality and potentially more accurate and versatile models that can handle complex tasks involving not just identifying objects, but also delineating their exact shape and size within an image.

[glenn-jocher]

Resolves #9439