I am a software developed in a small forestry open source application
and my job is it to fit cylinders. I have a solid validation pipeline.
Today I was provided by my core team with an exectuable. My method
compiled by them, switching from pcl 1.7.2 to 1.8. In a few hours I fit
round about a million cylinders into my test data.
I could never drop in my validation under 17.5 % rel rmse. I have a
publication in fact with 20% rel rmse under review. The exectuable
dropped to 15%. That is huge. I still do not know if this is really pcl
related or the core team of my tool did a lot more work on my plugin.
Well they are not supposed to work on my plugin, but there can be still
In any way, if somebody knows the MLESAC cylinder fitting routine (or
the RANSAC behind) was between the two named releases, give me
clarification please. I had a quick view on github, but I am really not
experienced to navigate there. I saw here
https://github.com/PointCloudLibrary/pcl/issues/1373 that SAC was
refactored and looked also the three issues behind it (one cylinder
Besides knowing that where my huge error decrease comes from I would
like to thank the person(S) behind it. Does the team at that time had
one dedicated to MLESAC (or complete SAC).
That is the most positive result I saw since long on the screen,