It is very easy to train on custom dataset. It depends on your custom data type whether it is single class or multi-class. If it is single class then you can just convert your train into MOT format and use MOT training pipeline. If it is multi-class then you can just convert your data into bdd100k data format and use motr_bdd100k training pipeline.

@HassanAliAnsari can you explain How?
I have labeled my data in CVAT and exported it as MOT 1.1 format. Is this format suitable for training? I am missing seqinfo files. It would be very helpful if you could explain it to me.

and It is just single class.