Folks,
I have added Nested Dissection based fill reducing ordering to Ceres Solver. Currently this is only available when using SUITE_SPARSE as the sparse linear algebra library.
You can try by setting
Solver::Options::linear_solver_ordering_type = ceres::NESDIS
This should help with SLAM/pose graph optimization problems where SPARSE_NORMAL_CHOLESKY is expensive with the default AMD ordering and SPARSE_SCHUR doesn't help either. For some problems we see 5x-10x reduction in solve times.
I'd be curious to hear if you find it useful.
Thanks to NeroBurner for the initial patch that this work is based on,
Sameer