Coco rle to polygon

[MacieJayDaaaaa]

Hello,I am trying to train the YOLOv8-SEG model using SA-1b. Yolov8 requires a Polygon type like x1, y1, x2, y2 ············ I use mask = mask_utils.decode(annotation["segmentation"]) How to convert a mask to a polygon type? Using cv2.findContours will result in excessive data

[math-yyj]

Why insist on polygon?

1 In SA-1B dataset, many mask have holes, and polygon fails to work well. 2 A better way is to customize yolov8's dataloader for RLE format input.