To utilise already labelled RGB images with polygon annotations for training SD-maskrcnn

[MadhuriPatil1694]

I have a JSON file(coco format) that contains the image path, name, size, bbox, mask in bitmask format. I also have the corresponding depth image. But I am a little bit confused about how can I convert it into dataset format for this task. Can someone please guide me?

[MadhuriPatil1694]

Should I make a different coco format JSON file for the depth images with the same mask co-ordinates