Implementation of the multi loss function in Faster R-CNN

I'm interested in the implementation of the Faster R-CNN multi loss function which optimizes over two tasks, the function in question is in the original paper.

I dived into the source code for the Matlab implementation of the techinique and quite frankly I was lost, I couldn't find the relavent parts to the forward loss and backward loss of the function.

If it's possible for anyone who has went through the code with a more experienced eye to point me to where it is and how it's implemented I'd be very happy.