Why is "overlap" not an option for instance segmentation models?

As per the title - for object detection, “confidence” and “overlap” are parameters when making predictions; however, for instance segmentation, “overlap” is not available. Why is this the case?

If I’m mistaken, how can I adjust “overlap” for my instance segmentation model?

Thanks!

1 Like

Just to add that when I downloaded the annotations and trained the model locally using YOLOv8, IoU (overlap) is a parameter and affects inference quite a lot, so curious why this wouldn’t be the case on Roboflow Instance Segmentation inference.

Hello there.
We do have this parameter for models - its name is iou_threshold
this is a base class being used: instance_segmentation_base - Roboflow Inference

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.