Training a machine learning model called YOLOP

I'm currently using a machine learning model called YOLOP to do some robotics-related tasks. This model detects objects, segments drivable areas and detects roadside lines. For my task, I only need to retrain YOLOP for object detection and drivable area segmentation on the dataset I have already collected. I will provide unlabeled datasets for learning that only contain a cycling path class that the model doesn't segment. moreover, I want to reduce the detection classes of YOLOP to 2. please have a look on the link below for YOLOP before contacting.
https://github.com/hustvl/YOLOP