Question about D^3 Fields' Semantic Representation Capabilities

[QinengWang-Aiden]

I want to express my appreciation for this impressive work!

I have a specific question regarding the semantic representation capabilities of D^3 fields:

Can D^3 fields represent multiple semantic attributes at a single position? I'm particularly interested in scenarios involving multiple object properties, such as physical attributes and visual attributes.

To illustrate my question with a concrete example "shoe alignment task with three shoes":

• Two black shoes that form a pair
• One white shoe
• All shoes placed randomly

In this scenario:

• Can the D^3 fields simultaneously encode both color information (black vs. white) and structural information (matching pair properties)?
• Would it be possible to selectively organize only the matching black pair while ignoring the white shoe?

I would greatly appreciate any insights from you!

[WangYixuan12]

Hi Qineng, thank you for your questions and interest in our work! For your first question - yes it can encode both color and semantic features. For your second question, yes you can do that. However, you may need to modify the correspondence rule yourself.