hainow / mctn Goto Github PK

Sir, your paper "Found in Translation: Learning Robust Joint Representations by Cyclic Translations Between Modalities" surprised me , I never see anyone deal the problem by MCTN.

Pity, I have an issue. I ask your help here heartedly. :D

problem:

During the 'Bimodal Cyclic Translations', we need capture the joint representation.
Then, how many seq2seq model we need? only one? or two?
Can we deal bimodal cyclic translations by a seq2seq model with an encoder and a decoder?

Thanks for u reading.~~and expect u reply.
:D

Hi, is there anywhere I can download the data?

Hi Hainow,
Thanks for sharing the code.
I have a question about the feature extraction,in the paper you say that details on feature extraction are in supplementary materials,but i cant find it on the website.Could you offer it for us?
Cheers,
wlimin

Hi,

Could you please let me know where in the code is beam search decoding implemented?

Thanks,

Hello, I am trying to run your code
However I encountered some problem preprocessing data.
The official CMU_MOSI dataset provided only h5 file or pickle style file, but in your utils/data_loader.py the program need csv file as input (eg. line 50 "file_name = facet_path + video_index + '.FACET_out.csv'")
Where can I get the csv format file ? Or can you provide a link to download the dataset ("data/mosi" in you project) ?