Issue: Handling Multiple Connected Regions from SAM2 for YOLO Segmentation Training

Hello,
I'm encountering a challenge when using Segment Anything 2 (SAM2) generated masks for YOLO segmentation model training. I'd appreciate any insights or suggestions on how to address this issue.
Current Situation:

Using SAM2 to generate object masks
These masks sometimes contain multiple connected regions for a single object
Preparing data for YOLO segmentation model training

Problem:
The YOLO segmentation model training process expects each object annotation to be a single polygon (one connected region) per line. However, the SAM2-generated masks often produce multiple connected regions for a single object.
Impact:
This mismatch leads to incorrect object annotations, especially for objects that are naturally segmented or partially occluded in the image.
Attempted Solution:
I've tried to merge multiple connected regions into a single connected region, but the results have been inconsistent and not sufficiently stable for reliable training data.
Question:
Are there any recommended approaches or best practices for handling this situation? Specifically, I'm looking for methods to:

Effectively combine multiple connected regions into a single, coherent polygon suitable for YOLO segmentation training, or
Adapt the YOLO segmentation training process to handle multiple polygons for a single object annotation.

Any guidance, examples, or references to relevant resources would be greatly appreciated. Thank you in advance for your time and expertise!