Faster RCNN implementation over a custom made deep net

 I built a custom deep net in MATLAB for a single object classification, I have trained this network using images that contain only one instance of the object/no object.
Now I want to use images with multiple instances of the same object and want localise them using bounding boxes. How can I implement a faster RCNN network over my current trained network (with single object images).

Thank you, Regards, Akhil