k-fold Cross-Validation: how to produce a ROC curve

[Carlo Alessi]

hello,

I trained the ILSVRC network and I have got 4 classifiers as I am doing 4-fold cross validation.

Now I would like to do a ROC curve. 

1) what are the steps I should consider to produce the ROC curve? 

2) Is there any shortcut to do that exploiting Caffe, at least partially? 

background information: 

- animal detection, so there will be multiple animals in a single image. 

- I'm using the classification.cpp example, so it just output a vector of scores

since I need TPR and FPR, should I write the code to compute them?