Currently, I am initializing a sparse matrix as follows:

Code: Select all

Eigen::SparseMatrix<float> A;
A.resize(pixels*3, pixels*3);

My question is that would this actually allocate the full memory (i.e. pixels * pixels * 9 elements)?

Now I have a scenario where I know before hand how many non-zero elements would be in the sparse matrix. In this case, should I be using `reserve` rather than `resize`?

Second, I am currently using `setFromTriplets` to fill the matrix as a batch but this also seems to be very memory intensive. if I want to reduce memory footprint, would using `insert(i,j)` after calling `reserve` or `resize` be a better idea?

Typically constructing the matrix from triplets is best.

From my reading of https://gitlab.com/libeigen/eigen/-/blo ... seMatrix.h your code does not allocate the whole matrix but just the column or row index (according to storage order) array.