Generic grounding

[ekazakos]

Hi,

Thank you very much for the repo! The provided code supports grounding for interrogative sentences. I'm interested in generic grounding, similar to the examples in the paper, where the question to the model is something like "What do you see in the video?", and the answer is a caption with tagged noun phrases and bounding boxes.

Could you please provide a script for this?

Also, can the model run without audio input?

Thank you!