Applying a custom NMS to the results of Faster R-CNN and R-FCN models

[KeeganFernandesWork]

How can I effectively integrate and use a custom NMS (Non-Maximum Suppression) algorithm in Detectron2 for object detection tasks? Specifically, I'm looking for guidance on implementing the custom NMS logic and incorporating it into the existing Detectron2 pipeline to enhance the precision of object detection results. Any insights or code examples would be greatly appreciated.