the cholesky_AAt in Non-Rigid ICP can't work

[AzureSilent]

Hello, When there are many vertices whose matching reliability weight is set to 0, the Hessian ATA is not full rank.
So only the slow solve method scipy_spsolve can work. The sksparse.cholmod.cholesky_AAt just throw an ERROR:
sksparse.cholmod.CholmodNotPositiveDefiniteError: c:\cholmod-scikit-sparse-window\suitesparse-metis-for-windows-1.3.1\suitesparse\cholmod\supernodal\t_cholmod_super_numeric.c:911: matrix not positive definite (code 1)

how do you handle with this? or I got the wrong version of scikit-sparse?

[jiaxiangshang]

I am also meeting this problem, have u solved?

[patricksnape]

The easiest thing to do would be to add some regularization to allow the matrix to be invertible. So just add a small number to the diagonal.

[jiaxiangshang]

Really thanks for your help

[mihaoning]

hello，may i ask how to add numbers to the diagonal of a sparse matrix？and how to avoid ill-conditioned matrix？thanks a lot！

[patricksnape]

Something like sparse_matrix.diagonal() += 1e-6?