How to export data to COCO format while keeping polygon data?

[StepanVL]

Hi. I have annotated datasets of images in CVAT.
I want to try detection approach rather than segmentation approach. I have two classes of objects:

• target objects are convex polygons that can be converted to circumscribing rectangles without loss of accuracy
• "bad_image" polygons of irregular concave that represent areas that should be excluded from the loss (and metrics) calculation

What is the best way to export data in coco format with rectangles kept as rectangles and polygons kept in two copies: as rectangle and as a polygon? I can write a custom parser, but I don't want to reinvent a tool if it already exists.

[zhiltsov-max]

Hi, CVAT doesn't force annotation representation on export in the COCO format. Rectangles are saved as bbox only, polygons are saved as polygons, and masks are saved as RLE.