MaskRCNN create own instance segmentation COCO dataset and convert to TFRecords

I am trying to create my own COCO dataset for a MaskRCNN network. I am following the guide the tlt-getting started guide: https://docs.nvidia.com/metropolis/TLT/tlt-getting-started-guide/#data_io_topic

They provide steps for converting a KITTI format dataset but not a COCO format dataset. What are the appropriate steps to follow to convert my COCO dataset to TFRecords so that i can do transfer learning?

The MaskRCNN network id like to use is this one from the tlt website:

Kind regards
HK