Dense/Object captioning feature encoding

[kietngt00]

Hi,
Thank you for publishing your interesting work!

I have 1 question about the function _get_instruction_response in captioner.py. As I understand, this function can optionally take a box or a click, and extract the point cloud feature for this box/click for captioning.

However, your current code does not use box/click inputs. Is that your intention to ignore these inputs?

Much appreciate!

[Believeht029]

Thank you for your attention! This is the visual prompt, our model do not use it.