Saliency map generation using c++

[Siddarth Malreddy]

Hey, 

I'm trying to generate a saliency map using the imagenet model. After getting the top classes for classification, I want to backpropogate to find the pixels which are used to estimate the class (like in this paper http://arxiv.org/abs/1312.6034). But I could not find any documentation on how to do that in C++. I know how to do it in python, but I want to implement in c++. Any help is appreciated. thanks in advance.