Rationale behind deciding the radii of ball query based groupers.

[arcticthunders]

Hello,

This is my first time raising an issue, so please let me know if this is not the right place to ask questions like these.

My question is regarding a model that involves N1 points and a D-FPS layer that produces N2 points. From these N2 points, we construct several balls of radii 0.2, 0.4, and 0.8. What is the rationale behind choosing these radii? Is there a rule of thumb or point distribution statistic that was used to determine these values?

I am asking this because I am curious if this is a significant factor when fine-tuning the model on a different dataset that uses different sensors and therefore different point densities.

Thank you for your help!