Both functions are great for comparing, for example, survival models. The first especially for computing Harrell's C-index, the second for NRI and IDI.
However, it looks like
rcorr.cens takes survival probabilities and
improveProb takes event probabilites, that is, 1 - survival probabilities, as arguments. Could somebody confirm this? It just feels weird, and I haven't been able to find information on it in the documentation.