Documentation

[jpan127]

The readme specifies there is a PDF file, but the releases only provide a .tex file.

A quick google search shows that it is convertible to a PDF file, however pdftex throws up:

entering extended mode
(./manual.tex
! Undefined control sequence.
l.1 \documentclass
                  [letter,10pt]{article}

Can you provide a PDF file (seems much simpler for everyone), or explain how to properly convert it? Thanks!

[cnuernber]

What about this prebuilt pdf?

[jpan127]

Cool! Sorry where was this located?

[cnuernber]

I googled around. It is hosted on uni servers.

[jpan127]

Oh I see, then I still believe it should be included in releases.

[target-san]

Sorry for interfering, but is there some kind of more detailed docs? In particular, the manual linked above mentions almost nothing besides KD-tree. And initial whitepaper compares KD-trees with K-means clustering trees almost exclusively. Though I see following algorithms in v1.9.1:

• autotuned_index
• composite_index
• hierarchical_clustering_index
• kdtree_cuda_3d_index (CUDA version of normal KD-tree?)
• kdtree_index
• kdtree_single_index
• kmeans_index
• linear_index (just linear search?)
• lsh_index

It would be really nice to see some kind of rough description for each of them and complexity estimations. I'm asking because I'm dealing with point clouds through PCL, and performance isn't the most desired. Computing normals over point cloud gives following approx. speeds:

• 155K are processed in about 2 seconds
• ~3M points give ~40K points per second
• ~45M points give ~7-8K points per second

My assumption is that performance is severely impacted by the growth of default KD-tree depth.