inverse square root of covariance matrix for residual weights

[phreak...@gmail.com]

Hi,

I'm struggling to find an appropriate approach for obtaining the inverse square root of my covariance matrices in order to weight the residuals. For some covariance matrices, it seems my matrices are close to not being

symmetric definite positive and thus my approach using LLT decomposition gives NaN. I also found that Eigen has a operatorInverseSqrt but it seems to give a different result. It is not the same upper triangular matrix from LLT, but a symmetric one

with similar values to that obtained by LLT. I'm unsure if this resulting matrix is an appropriate replacement (since it appears to handle numerically difficult matrices better, i.e. does not give me NaN) as-is or if I should maybe grab the upper triangle of this matrix.

Best,

Matias

[Sameer Agarwal]

If your matrices are rank deficient or near rank deficient, then I think the first order of business should be to figure out why that is, and if using those covariance matrices is okay.

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/37f23f35-b5c2-4b4a-9240-79f193794f55%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.