Vanilla Gauss-Newton?

[Paul Ozog]

Hi Sameer,

Is it possible to have Ceres perform standard Gauss-Newton?  I could set the min/max Levenburg-Marquardt diagonal to be 0... but I think I also need to disable the trust region.

Again, all I want to do is iteratively solve the normal equations, and repeat until the solution converges.  Some popular techniques we use in robotics/SLAM actually work better this way vs. using LM.  Thanks.

Paul

[Sameer Agarwal]

Paul,

Pure gauss newton is not convergent. It needs to be combined with either a line search or a trust region method. The closest thing you want is the dogleg solver and then you can choose between traditional or subspace dogleg.

Sameer

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/dd4897b2-ef5b-4d9d-87f4-088bc968a708%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.