从特征矩阵制作特征向量的标准向量

Question

I have an a Eigen::MatrixXd that I made from std::vector of Eigen::Vector3d . It was easy.

I made some transform manipulation on that matrix and I want return result as std vector of Eigen::Vector3d .

How could I make

std::vector<Eigen::Vector3d> form Eigen::Matrix3d?

Answer 1

Better stick with the std::vector object. Here is how I usually deal with such cases:

std::vector<Vector3d> vecs(n);
auto mat = Matrix3Xd::Map(vecs[0].data(), 3, vecs.size());

This creates a view on the data owned by vecs . Then play with mat as you like (except resizing of course,). eg::

mat = my_affine * mat;

No need to copy back the values from mat to vecs , but of course, if you have a Matrix3Xf or MatrixXf already at hand and want to copy it to vecs , then simply write:

mat = other_mat;

Provided that vecs.size() == other_mat.cols() , otherwise you need to resize vecs first and re-create a Map with new size.

Answer 2

I agree with @ggael and @RHertel that you should stick with one representation. If you need to dynamically insert Vector3d objects, then std::vector is probably the better solution (and you can still access it like an Eigen object, using Eigen::Map as shown by ggael).

Likewise, if you have a Matrix3Xd and want to use it column-wise in a standard algorithm, you can do that using .colwise().begin() and .colwise().end() which by itself won't copy any data (this requires the development branch of Eigen -- or the upcoming 3.4 version).

This also gives an easy way to create an std::vector from an Eigen::Matrix3Xd :

// `mat` needs to have 3 rows (at runtime)
std::vector<Eigen::Vector3d> vec(mat.colwise().begin(), mat.colwise().end());

Godbolt-Demo: https://godbolt.org/z/uCqZni