Evaluation on refcoco datasets.

[juxingyiwan]

Dear authors:
I'm really interested in your work. We are reproducing outstanding visual grounding works on refcoco datasets, could you please release evaluation scripts on refcocos?

[liujunzhuo]

Dear authors: I'm really interested in your work. We are reproducing outstanding visual grounding works on refcoco datasets, could you please release evaluation scripts on refcocos?

Hi,
I have a similar question regarding visual grounding evaluation. I've been using prompts from the paper, like "Please segment the man. (with grounding)," but sometimes end up with multiple targets, and sometimes, there's no <seg>. It's not ideal. Any tips or solutions you've come across?
Also, I'm having issues with batch inference - the results are different when the batch size is not 1. Do you happen to have any insights into this matter?
I appreciate your help!

[ApolloRay]

same question.

[XuYunqiu]

same issue