no re-ranking?
JonghwanMun opened this issue · 2 comments
Actually, I know that re-ranking gives higher score (for m-RNN, about 3 points gain in terms of BLEU4 score and about 9 points in terms of CIDEr)
I'm wondering that this work does not use re-ranking as other works do? also, no any post-processing?
Hi JonghwanMun,
Our work came out a few months prior to Microsoft's nearest neighbour method and Junhua's consensus re-ranking idea. This is why re-ranking is not part of this project.
However, I have tried out Junhua's re-ranking method (not exactly the same, but similar) using our models and have gotten similar performance improvements with CIDEr.
@ryankiros
Thanks for answering.
Is there an additional performance improvements on BLEU? (which is higher score than in paper - 70.7)