About pretraining with Mapillary Vistas

It is general to pretrain with Mapillary Vistas and fine-tune on Cityscapes. The thing and stuff categories are quite different for these two data sets, is there some special treatment required?
Especially, some things (e.g. traffic lights, poles) in Mapillary Vistas are marked as stuff in Cityscapes, how you deal with such categories?
