Question about pretraining

cshizhe / VLN-DUET

Official implementation of Think Global, Act Local: Dual-scale GraphTransformer for Vision-and-Language Navigation (CVPR'22 Oral).

106 stars 8 forks source link

Question about pretraining #9

Closed heyzude closed 1 year ago

heyzude commented 2 years ago

Hi, thank you for opensourcing your work!

I have a question about pretraining phase.

I'm about to test Duet on R2R, but I wonder when you got the R2R performance on your paper, did you pretrain the Duet model not only on R2R dataset but also on R4R, or other tasks like REVERIE?

I guess pretraining not only on R2R but also on other tasks can boost up the Duet performance on R2R val_unseen split.

cshizhe commented 1 year ago

We only pretrained on R2R.