DetectorRS implementation from paperswithcode top benchmark

DetectorRS is the top model with all reported metrics according to Mask AP score in this benchmark:

https://paperswithcode.com/sota/instance-segmentation-on-coco

It has been ported to mmdetection few months ago.

How can this model be used from Detectron2?

All models from detectron2 model zoo perform so-so on my dataset.

Start by reviewing the Papers with Code instance-segmentation benchmark and the DetectorRS implementation in mmdetection, then compare them with the Detectron2 model zoo. Determine what integration or usage path the issue requires and define done as a documented, working way to use DetectorRS from Detectron2 with its reported evaluation metrics.

Written by the indexing model from the issue text.