Description: image editting problem, Y. Yoshiyasu, Keio Univ, Japan
|(bipartite graph drawing)|
|number of rows||240,000|
|number of columns||120,000|
|structural full rank?||yes|
|# of blocks from dmperm||1|
|# strongly connected comp.||7,516|
|explicit zero entries||0|
|nonzero pattern symmetry||0%|
|numeric value symmetry||0%|
|kind||computer graphics/vision problem|
|Additional fields||size and type|
The problem is template-mesh deformation to match with silhouettes. In this process, there are two kinds of linear systems to solve. This system (Yoshiyasu/image_interp) is a smooth vector field construction from images, which is harmonic interpolation (minimizing laplacian: Lx=0) of intensity gradient field p. This can be solved by normal equation and cholesky factorization, x=(A1'*A1)/(A1'*b1), where A1=[L;C] and b1=[zeros(size(length(L),1);1);C*p]. C is a square diagonal matrix containing weights. This is for a 400x300 image, so Ix=reshape(x,400,300) must be done to get the vector field. After solving y direction for Iy, the result is visualized with quiver(Ix,Iy). At each iteration the both C submatrix and the right-hand-side change but L remains unchanged. [Note by T. Davis: since C is of high rank, update/downdate will not be effective, since it is meant for low-rank changes.]
|nnz(V) for QR, upper bound nnz(L) for LU, with COLAMD||316,991,703|
|nnz(R) for QR, upper bound nnz(U) for LU, with COLAMD||12,952,004|
For a description of the statistics displayed above, click here.
Maintained by Tim Davis, last updated 12-Mar-2014.
Matrix pictures by cspy, a MATLAB function in the CSparse package.
Matrix graphs by Yifan Hu, AT&T Labs Visualization Group.