Additive refinement strategy

[Kurtaga]

If I understand correctly, current COPC writers (such as PDAL) use a replacement-style hierarchy - where parent and child nodes contain overlapping points. While, an additive refinement approach would have deeper nodes contribute only new points, rather than repeating those already represented at higher levels.

This leads to increased file size and redundant data between levels. For my use case (a real-time point cloud renderer), an additive strategy would likely be more efficient, both in file size and performance.

From what I can tell, the COPC specification doesn’t really enforce a particular refinement strategy - meaning it is up to reader/writer to enforce that?

I’m wondering if anyone has experimented with an additive encoding approach, or if there’s already a tool or writer that supports it.
If not, I’m considering implementing one and would appreciate any feedback or thoughts on whether all this makes sense.

[wonder-sk]

Actually COPC only supports additive refinement strategy 🙂

[Kurtaga]

Hmm, so just to confirm - if I want to render everything correctly, at say, depth 3, should I render all nodes from depth 0 up to 3, or only the nodes that are exactly at depth 3?

[wonder-sk]

You should render all nodes from depth 0 up to 3

[Kurtaga]

Ohh, okay that's great!
I'm now double checking things and it checks out
Thanks a lot :)

[connormanning]

the COPC specification doesn’t really enforce a particular refinement strategy

We do cover this in the first paragraph of the EPT description of the nodes, which is linked to from the COPC spec here, but it's true that the COPC spec as a standalone document doesn't make this clear. Since there are various links from the COPC to the EPT document, some of the more conceptual context like this only exists in the EPT spec and is not reproduced here.