how can i train faster R-CNN using a new customised network model?

[MjdMahasneh]

I would like to know, If it is possible to train faster R-CNN using a new customised network model?

for example: i want to create a model that takes multiple inputs (2 images per class object), can i adapt this model to faster R-CNN? if so, what do i need to modify in the implementation to achieve it?