[filter] Revert Morton

[pblottiere]

Hi,

This PR adds a new option for the MortonOrder filter: revert. If this option is true, then the indexing is made according to a revert Morton code. It allows to have a good spatialisation.

For example, with a pipeline Splitter -> Revert Morton, we may successively select N representative points of a patch:

We are using this filter in LOPoCS with the pgpointcloud writer. BTW: Oslandia/lopocs#23.

Let me know what you think about it and if I have to update something.

[hobu]

@pblottiere Can you tweak the docs to include this picture and a section about what the option does with links to LOPoCS? https://github.com/PDAL/PDAL/blob/master/doc/stages/filters.mortonorder.rst

[abellgithub]

I'm thinking that it would be nice if the original code could be modified to fit this model: don't have a special sort for the map, but return an integer that represents the order. This would eliminate duplicate code. I could do this as a second pass, I guess.

[abellgithub]

Also, is there some documentation that explains the math/logic of this ordering? "Revert morton" doesn't mean anything to me. I didn't find anything Googling, either.

[hobu]

Also, is there some documentation that explains the math/logic of this ordering? "Revert morton" doesn't mean anything to me. I didn't find anything Googling, either.

I think maybe the option should be called reverse for "Reverse Morton", or maybe invert for "Inverted Morton". It's something they made up for the LOPoCS to create dispersement of data so when they data are fetched, they are presorted for pleasant looking visualization.

[abellgithub]

Perhaps it would be better to pass a function to the filter that would allow one to generate an index based on a coordinate? We could build-in the current and proposed ordering. Maybe this is a to-do.

[pblottiere]

Can you tweak the docs to include this picture and a section about what the option does with links to LOPoCS?

Done.

[pblottiere]

I'm thinking that it would be nice if the original code could be modified to fit this model: don't have a special sort for the map, but return an integer that represents the order. This would eliminate duplicate code. I could do this as a second pass, I guess.

I didn't want to update the original code but I may propose a commit to clean duplicate code if you want?

[pblottiere]

Also, is there some documentation that explains the math/logic of this ordering? "Revert morton" doesn't mean anything to me. I didn't find anything Googling, either.

for the LOPoCS to create dispersement of data so when they data are fetched, they are presorted for pleasant looking visualization.

Exactly. But I'm sure that @Remi-C will be happy to talk to you about it if you want more information :).

[abellgithub]

My only concern is that Morton-ordering implies a scheme for ordering 2-D points in 1-D while preserving locality to some extent. Unless I'm mistaken, this algorithm isn't attempting to preserve locality. Perhaps this is a naming issue and nothing more.