About the token cost of one 3D scene

[KaKa-101]

Could you tell me how many tokens you used to represent the whole 3D scene before sending them to LLM?
Thanks a lot.

[ch3cook-fdu]

We represent a 3D scene with 32 learnable tokens following BLIP2 and InstructBlip https://github.com/Open3DA/LL3DA/blob/main/models/ll3da/captioner.py#L183. We introduce 8 additional tokens for each visual prompt.