Detectron2 converting instances to annotations

[jdc2021]

Hi all, I have trained an image recognition model to detect fruit flies. However, I am now trying to improve the model by re-training it with better data from different images. I would like to use the original model to do the majority of the annotation work so that I can simply verify the annotations and correct false positives and negatives. However, when I run the model on the new images, I can't figure out a way to convert the output from the predictor (when I print it I can see it has the required bounding box coordinates that I'm looking for) to a file that I can input to Roboflow (the annotation tool I'm using, but I'm open to suggestions for better ones if they'll work for what I want to do). I would ideally like to create a COCO json file that I can easily upload to Roboflow to initialise the annotations but I have been struggling for a long time to figure out how to do this. I would very much appreciate any help that the community can provide - thank you.