Re: [eigen] New Levenberg Marquardt stuff

[ Thread Index | Date Index | More Archives ]

On 07.12.2012 20:57, Keir Mierle wrote:
At the risk of starting a war:

At the risk of accepting the new war (and a bit off-topic) ...

I never quite understood, what made you publish yet another general least-squares solver, after g2o, iSAM and SLoM/MTK (all available on I can't deny that ceres offers things neither of the other do, but for large parts its a re-invention of the wheel (e.g., what you call "local parametrization" is part of SLoM since 2008.)

Have you considered punting on the LM library entirely?

Ceres is almost uniformly better than what's there now, is heavily battle
tested in production both large (Street View) and small (Android, in every
Nexus 4), is BSD licensed, uses Eigen internally, has extensive sparse

Well, it might use Eigen internally, but ceres' interface dealing a lot with double-pointers is far away from the clearity of Eigen ...

It's not clear that Eigen dev's time is best spent working on LM when Ceres

As mentioned already by others, Eigen's LM module existed way longer than Ceres, and there existed a bunch of LM implementations before that. Furthermore, I agree with opinions raised by others that including a generic non-linear LS solver into Eigen might/should not be the goal of Eigen, which IMHO should basically be a *linear* algebra package. As long as no further dependencies are introduced (neither for compiling nor for licensing), I don't mind if some small extensions are added, though (clearly not the case for ceres having 2--5 more dependencies).


Dipl.-Inf. Christoph Hertzberg
Cartesium 0.049
Universität Bremen
Enrique-Schmidt-Straße 5
28359 Bremen

Tel: +49 (421) 218-64252

Mail converted by MHonArc 2.6.19+