Subsampling by volume density?

[I&R_Digital]

Hi, we are new here, and before anything wanted to thank CC crew for all the joy of using it, and all the pain it saved us.
we were using it for a few years, mostly on projects in our previous shipyard which we tried to solve with Reverse Engineering.
anyhow, we have new company and would like to offer same services to a broader audience.

we are delivering .stl file of reconstructed point cloud for our customer, and would like to send him a point cloud also. in this case, .stl is about 270kB, and .e57(trimmed and subsampled) is about 28 MB. file isn't large in this case, but it could be for other cases.
regarding that, we have a question:
is it possible to do subsampling by volume density? for example, set one number from scalar field and erase all points denser than that. for proof and for 3d reconstruction denser points are irrelevant, and because they are denser, removing them would lighten the output file.
we tried several(random, random%, space, included SF, octree) subsampling methods(maybe we didn't found proper one), and those we tried do not seem to reduce density by some threshold, there are still "dense" clouds, judging by scalar visibility bar.

regards,
Ratko
I&R Digital Engineering

cc density.jpg (589.54 KiB) Viewed 4298 times

[daniel]

So I would still choose the 'space' method, which guarantees that there's no pair of points that are closer from each other than the set distance. It's a kind of low pass filter on the density. So you should find a value that's large enough, that no area will be 'too dense', but some areas might have an even lower density from the beginning.

[I&R_Digital]

Hi Daniel, thanks for suggestion.
after tinkering with space settings I've got relatively satisfactory result regarding visibility and size.
volume/surface density didn't drop much as a result, even tough file size did. i was hoping to leave (in above example) lower half of densities untouched, and just reduce upper half to average values.
maybe next step would be Space subsample with "use current SF", and then adjust both spaces(less subsampling on smaller values, more on greater ones) to have more impact on greater values?

[daniel]

If you manage to find the right parameters maybe?