How to train RTMDet-Rotated with uncertain annotations (ignore_iof_thresh, isrowd) on COCO format dataset?

[Renat2001]

I have a dataset that contains iscrowd annotations. I do not want my model to consider them neither background nor object.
https://mmyolo.readthedocs.io/en/latest/recommended_topics/training_testing_tricks.html#training-tricks
Here is the page dedicated to "Testing tricks" while using MMYOLO. But I need more deep explanation of https://mmyolo.readthedocs.io/en/latest/recommended_topics/training_testing_tricks.html#consider-ignore-scenarios-to-avoid-uncertain-annotations part.
Is there someone who managed to use this feature successfully on COCO format dataset. Thanks!