Ransac stopping criterion

[Riccardo Pedrelli]

Hi everyone,

I recently stumbled across this paper, showing an approximation in the classic RANSAC implementation that could lead to an undersampling of good models :

https://arxiv.org/html/2503.07829v1

I was wondering if this was accounted for/ was fixed in the latest versions ASP

Thank you!

Riccardo

[Oleg Alexandrov]

It is good to know that such fundamental algorithms as RANSAC still get revisited. 

In ASP we do not "adaptively stop RANSAC". We run it for 1000 iterations, which is totally outrageous, but quite fast with multiple threads in parallel (and faster than the other steps in interest point matching).

RANSAC can surely fail if there are not enough features in images or if the inlier threshold is not good, but that can't be fixed by the number of iterations.

--
You received this message because you are subscribed to the Google Groups "Ames Stereo Pipeline Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ames-stereo-pipeline...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/a259e4c5-1820-438b-8811-47fc55103f34n%40googlegroups.com.

[Riccardo Pedrelli]

Thank you so much Oleg for the clarification!