Converting Object Detection to Instance Segmentation Project

If I labeled a bunch of images in an object detection project with polygons and want to export those annotated images and reupload them to an instance segmentation project, should I use the COCO format? Does it matter which format its in, if I’m just reuploading the dataset back to Roboflow?

According to this blog post: Instance Segmentation Projects The best format to export to is COCO JSON if the goal is to keep the polygons.

I also found that COCO JSON is a format that is used for multiple computer vision problem types (object detection, instance segmentation, semantic segmentation). the actual contents of the JSON file differs based on the problem type.