How to Construct a Ground-Truth Test Dataset

[hungnh1125]

Hi,

I noticed that you used SAM and Grounding DINO to generate segmentation masks.

Could you please explain how you merge the outputs from SAM and Grounding DINO to create the ground-truth in the GranD-f dataset?
Additionally, could you describe the process of creating the final dense caption?
Is your method fully automated, or does it require manual verification?
I am interested in applying your method to the GranD dataset.

Thank you.