Closed Eurus-Holmes closed 4 years ago
Hi,
Those models are not related at all to the baselines in the paper, so you should just ignore them. You are right in the sense that we still did not provide a simple codebase to train the baselines presented in the paper, sorry for that. What kind of baseline are you specifically interested in?
@ozancaglayan Thanks a lot! I'm specifically interested in Machine Translation baseline, I found that the BLUE of the baseline is the same with multimodal baseline, so I want to know the video feature how to help machine translation?
From https://github.com/lium-lst/nmtpytorch, there are only limited related video_nmt codes,
experimental/coattentive_video_nmt.py
andexperimental/video_nmt.py
, and their organization is not very well, could you provide your own baseline on How2 dataset from original paper?