Loss Function

[SF]

Hi all,

I'm investigating how effective is the loss function in Ceres solver in down weighting the influence of outliers. I'm currently playing with HuberLoss and CauchyLoss. I have tried with several configurations, but it appeared to be only CauchyLoss(0.3) is the best configuration for my problem, regardless of how far the outlier is compared to my inlier. So, my question is, is there a proper guide on how to tune the scaled parameter of HuberLoss and CauchyLoss?

Thank you.

[Sameer Agarwal]

Shin Fang,

The right way to do this, is to think in terms of a likelihood function, by first modeling the errors that you expect in each of your observations/residuals. That will lead to the choice of shape of the loss function (it is the negative log likelihood) and the width/scale of the robust loss function will be some function of the variance/spread of the probability distribution.

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/62f36c39-1bba-4d69-a94c-3a96b24b9b4d%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.