the quads will probably collapse under the current alg construction (look into this). assuming the resulting quadmesh is sufficiently not respecting the underlying size function, what if we 1 made a copy of the quadmesh and then for each quad 2 convert it into 4 tris and 3 perform the admesh truss solver but do not allow it to perform any mesh mutations at all except for vertex locations so that after its solves we take the revised vertex locations and update the original quadmesh? might need to then fem smooth the quads after that update but maybe after all that the quads can be both high quality and also respect the size function? or maybe all of this is unnecessary because the quadmesh sufficiently respects the underlying size function idk
the quads will probably collapse under the current alg construction (look into this). assuming the resulting quadmesh is sufficiently not respecting the underlying size function, what if we 1 made a copy of the quadmesh and then for each quad 2 convert it into 4 tris and 3 perform the admesh truss solver but do not allow it to perform any mesh mutations at all except for vertex locations so that after its solves we take the revised vertex locations and update the original quadmesh? might need to then fem smooth the quads after that update but maybe after all that the quads can be both high quality and also respect the size function? or maybe all of this is unnecessary because the quadmesh sufficiently respects the underlying size function idk