Mask RCNN dataset labelling

[skarfie123]

Hi, I'm trying to create a dataset for training Mask RCNN. I'm using CVAT for annotations. I understand I need to export in COCO json format. Do I need to draw bounding boxes too or will they be assumed from the polygons I have drawn?

[skarfie123]

turns out when exporting COCO format from CVAT, it generates the both the bbox and segmentation attributes for Polygon Shapes. so no extra work. Note. Rectangle Shapes only generate bbox, and segmentation will be empty, so detection will complain.