Hello,

Thanks for the great new model! Just curious if you're open to a PR that allows the current implementation to work with a YOLO-formatted dataset instead of just COCO?

(no worries if not, I can guess why).

Cheers.