I have a doubt that it makes sense to compute the closed-form affine/rigid/tps in align_in_real_world_coords=False.
EDIT:
in fact the transform you compute is correct for non-cube shapes if align_in_real_world_coords=True, because before fitting you:
- denormalize coordinates
- then apply affine
in here:
So if one wants KM to work fine with non-cube shapes but does not have affine matrices, he can just set align_in_real_world_coords=True, and then feed identity affines.
I have a doubt that it makes sense to compute the closed-form affine/rigid/tps in align_in_real_world_coords=False.
EDIT:
in fact the transform you compute is correct for non-cube shapes if align_in_real_world_coords=True, because before fitting you:
in here:
So if one wants KM to work fine with non-cube shapes but does not have affine matrices, he can just set align_in_real_world_coords=True, and then feed identity affines.