Question regarding training EVF-SAM from scratch

[jadenpark0]

Hi, thanks for the interesting work!

I am trying to reproduce the results.

When I train only on the referring segmentation data following the same setting, and evaluate on refcoco|unc|val, what would be the expected performance? I am only observing ~0.45 g/cIoU.

I want to verify if my script has implemented what is described in the paper properly :)

Thanks!