Re: [eigen] New Levenberg Marquardt stuff

[ Thread Index | Date Index | More Archives ]

Hi list!

I have successfully used the Eigen LM solver in the past and I am very happy with it. I agree with
Thomas, that LM is so closely related to linear algebra, that it definitely qualifies to be shipped
with Eigen.

Also since Eigen is headers only, it is no problem to simply omit the parts you do not need. That is
you do not pay for what you do not use.

Thus I strongly favour keeping LM in Eigen. If Ceres contains an LM solver, which is superior to
Eigen LM in terms of speed or accuracy, it could be considered to replace the current implementation.
But this has yet to be proven, in my opinion. Remember that the MINPACK implementation of the More
algorithm, is a de-facto standard, which has proven its reliability in numerous applications.


Mail converted by MHonArc 2.6.19+