search

neccam / nslt

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

Train nslt with glosses as target #10

Closed iliasprc closed 5 years ago

iliasprc commented 5 years ago

What are the optimal parameters ( number of layers, hidden units ....) to train nslt with glosses as targets ?

neccam commented 5 years ago

Hi Ilias,

I haven't personally tried training recognition with the NSLT pipeline, but one might guess that it would be similar to the ones I have used in the paper for doing translation.

One thing I must point out is that if you are going to evaluate your predictions with the builtin metrics in this code base, you should check Koller et al's evaulation scripts for recognition which comes with the phoenix 2014 dataset. In them there is some preprocessing done on the predicted glosses before calculating the WER.