Question: why are WOLF Eigen matrices RowMajor?

The Eigen default is ColumnMajor! Why did we change this?

added question label

@joanvallve @artivis you are the only ones I can imagine to have made this change. Any idea?

Briefly introduced here #14 (closed)

In ceres::CostFunction::evaluate(double const *const *parameters, double *residuals, double **jacobians) (see link), the Jacobians are provided thru a "row major" array. This is important only in the case of analytical factors (factor_analitic.h L105) in which this array is mapped using an Eigen::Map<Eigen::Matrix>.

The API of the factors is going to be redesigned. Most probably, using Eigen::Map<Eigen::Matrix> over some ceres::jets that also contain the Jacobian rows. Then I think this storage order is important to be maintained.

In Eigen, they recommend not to mix different storage orders. Also, they recommend to use colmajor as the default and most tested one.

I guess this can be a problem with the use of other libraries, such as manif?

I would leave this issue opened to be taken into consideration whenever we face the new factor API design.

Indeed, now manif and wolf are incompatible because of this.

We can possibly assume that transferring matrices from row to column major could be doable at small scales (like the jacobians of any of the factors). The questions is what to retain as default.

Personally, I would like to have the same default as Eigen, but I still do not see the implications (either positive or negative) of changing it.

mentioned in merge request !328 (merged)

closed