How to use a class from MS-COCO and add 2 new classes with Mask RCNN?

Question

I want my model to detect only 3 classes i.e. Person, Tyre and Road. I'm using the pretrained MS COCO weights which are already trained on the 'Person' class.

How do I use this already present 'Person' class and further train my model for 2 new classes i.e 'Tyre' and 'Road'?

I've set NUM_CLASSES = 1 + 3 (background + objects).

My dataset however only has annotations for 'Tyre' and 'Road' and hence these are the only classes I'm adding using self.add_class(source_name, class_id, class_name).