Mask-level Ferret?
joshmyersdean opened this issue · 2 comments
joshmyersdean commented
Hello again!
I was wondering how you were able to have Ferret output segmentation masks? From the paper it looks like it is restricted to yo bounding boxes?
Thank you!
LiWentomng commented
@joshmyersdean Hello, thanks for your interests.
All models, including Ferret, only generate the semantic understanding, not outputing segmentation masks. Regarding open-vocabulary segmentation, all models use the groundth-truth of mask to conduct the performance evaluation of their semantic interpretation capabilities.
joshmyersdean commented
Thank you!