Why the 3D grounding result shows 2D box instead of 3D box?

[jkstyle2]

Thanks for sharing your great work.
As title said, I'm wondering why the result from 3D grounding task is 2D, instead of 3D box.
3D-LLM was evaluated on ScanQA dataset, which normally expects 3D location and the answer to a question for a specific object as output.
Is this a just visualization example of 2D box? I wonder if the 3D-LLM can estimate the 3D box as its output.

Thanks for your answer in advance!