search

neccam / nslt

Neural Sign Language Translation (CVPR'18)
Apache License 2.0
149 stars 41 forks source link

How to train a S2G2T model #9

Open MeteorsHub opened 5 years ago

MeteorsHub commented 5 years ago

Hello! In my opinion, the training sample code in your readme is a S2T model, right? How can I train a S2G2T model in your paper?

akonstantinou commented 5 years ago

The code for the S2G2T is not given, this repo is for the S2T model only. I hope they change their mind and publish the code for that model too.

MeteorsHub commented 5 years ago

Have you asked them for it?

neccam commented 5 years ago

Hi,

In S2G2T, we use the earlier state-of-the-art gloss recognition network by Koller et al (CVPR 2017) as our spatial embedding and tokenization layer and feed the predicted glosses to a text2gloss model. A Text2Gloss model can be trained by using the original tensorflow-nmt code base by Luong et al.

I will talk to Koller et al. to see if they would be willing to release the model or at least the predicted glosses.

saisriteja commented 5 years ago

@MeteorsHub did u get any resource or similar way to do this.?

MeteorsHub commented 5 years ago

No, I didn't

EricWangyz commented 5 years ago

@neccam feed the predicted glosses to a text2gloss model? or feed the predicted glosses to a gloss2text model?

ycmin95 commented 4 years ago

@neccam if it is not convenient for releasing previous work, can you release the translation results of G2T and S2G2T models?