[Liblas-devel] compression benchmarking

[Michael P. Gerlek]

I've been asked a few times about comparing Martin Isenburg's laszip
compression scheme to LizardTech's MrSID/MG4 compression scheme, both for
speed and compression ratio.

Unfortunately, due to a restriction in LizardTech's EULA I can't publish any
MrSID numbers.  And in any case, one would need to be careful in comparing
the two, because they are aimed at different workflows: laszip is designed
to be a lossless archival format, with no support for random access into the
file, while LizardTech's format is designed to support lossy compression and
random access.

Anyway, in honor of the imminent 1.6 liblas release, here's some data for
y'all to chew on:

* .laz files get an average compression ratio of about 8:1

* with las2las, on my machine (2.2GHz Xeon) I get 1.36 mpps encode and 0.88
mpps decode
   - "mpps" is "millions of points per second", a new term of art in the
lidar benchmarking world

* Martin's native "laszip" tool is 20-40% faster than the comparable las2las
tool in liblas
   - this is due to a tradeoff in features/functionality, and possibly
different I/O approaches

My numbers are from a benchmark set of three "typical" datasets.  Your
mileage will undoubtedly differ, especially based on your CPU speed and I/O
subsystem.  If you'd like the full details and the spreadsheet of results,
email me.

I like benchmarking, so perhaps I'll periodically update these numbers as
liblas improves, add representative data sets, etc.  I'd like to add times
for random access (decoding a subset of the file) too at some point, esp. if
LT will consent to have me publish numbers for their tools.

-mpg
 [disclaimer: yes, I used to work for LT and yes, I'm embarrassed/ashamed
that clause is in the EULA]