SACSegmentation Linux and Windows Different Results

Question

Given identical inputs and identical code, a different plane model is found on Windows compared to Linux using SACSegmentation in PCL. For one example, the identified plane on Linux has a roll angle of ~7 degrees where the Windows plane is ~0 degrees. Consecutive Windows results are identical as well as consecutive Linux results. Could this be related to GCC and Visual Studio using different sizes for types? Or Windows using a better default_random_engine ?

Same calculation on Linux and Windows --> different results

<random> generates same number in Linux, but not in Windows

I know the Windows plane is the better identified plane. How can I best remedy this difference so that the Linux version also finds this same plane?

Answer 1

It's about the random engine. If you only run a few iterations (in relation to your cloud size) it might give unreliable results. You could try to add more plane candidates to your RANSAC engine by the .setMaxIterations( ) method to see if the results get closer to each other.