304_ConjugateFields not converging

[jockyuiz]

The conjugacy solver in this tutorial is not converging for me. I don't know what's going wrong but it seems the non-conjugacy value gets even higher after each iteration...And the final result gets crazy. Do you know what may potentially go wrong here for me?
For constraints and mesh, I think I use the original data in the shared path. And I think I do not change anything in the code. But I don't know if I somehow rewrite it when running other programs.
Iteration data: https://pastebin.com/E3tdiWjr
inspired_mesh_b.dmat: https://pastebin.com/nX8S7LWM
inspired_mesh_bc.dmat: https://pastebin.com/VJ2Yzu6t

Also a more general question: for all the examples we tried to interpolate the field given some fixed constraints. But can we simply use a field that already exists on the mesh, i.e. the principal curvature, or take the gradient of some scalar function? For this example, a directional field that aligns with principal curvature should already be conjugate... So may we just use this directly instead of finding a deformation for an interpolated field?

[avaxman]

I will be looking now at your input. For your question: there is libigl gradient: https://github.com/libigl/libigl/blob/main/tutorial/204_Gradient/main.cpp
and libigl principal directions:
https://github.com/libigl/libigl/blob/main/tutorial/203_CurvatureDirections/main.cpp
More general versions (for instance for N-fields) will be released in the next release (with some other exposed differential geometry operators that already exist in the internal implementation).

[avaxman]

hmmm it is the original data and consequently works well for me in both linux and windows... could you say which system you use exactly?

[jockyuiz]

hmmm it is the original data and consequently works well for me in both linux and windows... could you say which system you use exactly?
I am using Windows 10, x64-based system. How can I debug here to see where might go wrong?

[jockyuiz]

I will be looking now at your input. For your question: there is libigl gradient: https://github.com/libigl/libigl/blob/main/tutorial/204_Gradient/main.cpp
and libigl principal directions:
https://github.com/libigl/libigl/blob/main/tutorial/203_CurvatureDirections/main.cpp
More general versions (for instance for N-fields) will be released in the next release (with some other exposed differential geometry operators that already exist in the internal implementation).

Those examples in libigl seem to be vertex-based fields. When we transfer vertex-based directional fields to face-based, is there a good way to do that? Summing up the values at three vertices and simply take the average seems not working well for me