Fast model of CCellularGridND about dgtal

[JacquesOlivierLachaud]

In fact, in my former "digital" library "Imagene", there was a compact (and somewhat faster) model of KhalimskySpaceND. The idea was simply, if you have a bounded domain, to encode all information describing a cell (topology + coordinates) as some 32 bits or 64 bits integer. Then, set of cells (for instance with same dimension) could be encoded as bitsets. Another advantage was that the sign+topology was coded as a (d+1)-bit word, so incidence relations were precomputed as LUT.
So this issue is related to back porting this model of CCellularGridSpaceND into DGtal.
In fact, there is (an old) branch on my repository FastKSpace
https://github.com/JacquesOlivierLachaud/DGtal/tree/FastKSpace
where this is done...
The only problem is that this approach is efficient if you encode the digital sets as bit sets, which was not done at the time. I am afraid also that this branch is no more mergeable due to the improvements made by Roland to have "Unbounded" and "Toric" KSpaces.

[phcerdan]

Asking here out of curiosity. What is the difference between KnSpace and the current Khalimsky Space? Where is the bottle neck in current implementation of K-Spaces? And... more generally: is there any "winner" in terms of computational efficiency for a data structure for Cellular Grids and/or Cubical Complexes? I find the literature a little bit too vast to form a clearer idea on this. So asking for insight here.

I am pretty sure the following idea has been explored before, but, what is your opinion on ND cellular complexes in an image grid, which have the occupancy of lower dimension cells encoded in a bitset?
There are 26 lower-dimension cells in a voxel (8 points, 12 edges, 6 faces) plus the d-cell voxel itself.

Maybe it is not superflexible, neither generic for dimensions greater than 3, but it seems... compact.

Anyway, just thinking out-loud, sorry to invade this issue! @dcoeurjo @rolanddenis

[phcerdan]

As a reference to what you were illustrating, I link this paper of yours in arxiv: https://arxiv.org/pdf/0906.2767.pdf

And also your branch FastKSpace, but rebased on top of current master: https://github.com/phcerdan/DGtal/tree/fast_kspace

[JacquesOlivierLachaud]

Thanks for the link and the rebase.