Second step of training is unclear

ronghanghu / speaker_follower

Code release for Fried et al., Speaker-Follower Models for Vision-and-Language Navigation. in NeurIPS, 2018.

http://ronghanghu.com/speaker_follower/

Other

127 stars 31 forks source link

Second step of training is unclear #13

Open mees opened 5 years ago

mees commented 5 years ago

Hi, when I try to do the second step of the training export SPEAKER_PATH_PREFIX=tasks/R2R/speaker/snapshots/speaker_teacher_imagenet_mean_pooled_train_iter_20000 it fails because the files available after training are speaker_teacher_imagenet_mean_pooled_train_iter_20000_enc and speaker_teacher_imagenet_mean_pooled_train_iter_20000_dec, which one should we export to the path? Or did something go wrong during the speaker training?

Tangolin commented 3 years ago

Hi @mees, did you manage to get it up and running in the end? The implementations in data_augmentation_from_speaker is really confusing to me as well. Especially with respect to the need for a follower during data generation