Training MogaNet + Cascade Mask R-CNN without segmentation and MogaNet + RetinaNet on custom dataset

[MrDreamy]

Hello, I'am impressed by the results of MogaNet and would like to try it's performance on a custom dataset VinBigDataCXR - https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection. This dataset doesn't have instance segmentation masks and thats why I would like to use MogaNet + Cascade Mask R-CNN without the Mask part. Is there a possibility of doing this ? If yes, could you please provide a config example ? Also is there a way to use a custom dataset ? If yes, how ?