How to prepare input data for rcnn?

I have been working on cnn for past few months. I have run and tested (with 96% Accuracy) a simple 12 layer cnn with a custom face dataset (of 10 subjects, 10 samples each). I now need to convert my cnn to rcnn or find out a way to use my cnn architecture so that it works on videos. I had used a pretrained rcnn demo for creating a road-sign detector which only has 2 labels (signs and background). However now I need to write a rcnn with 10+1(background) labels. I havent yet found a way to do that. The solution must be obvious but as there are no guides on how to prepare variables and data for training a rcnn it has been frustrating. Can someone please help? thank you in advance.