Closed minnerbe closed 11 months ago
I just validated the new model against the 'old' alignment (with InterpolatedModel2D
). There are 5-10 (of 1729 tiles) that have a relative error of more than 5%, the rest is well below that (about 1% or even less).
This is the same for comparing two runs of the 'old' model on the same stack, since there is a considerable amount of randomness in the optimizer. Therefore, I conclude that this PR does not significantly change the alignment results.
I added a model that interpolates an arbitrary number of
? extends Affine2d<?> & Model<?>
. In comparison to the nestedInterpolatedModel2D
approach, the code isThe new model is used in the alignment-solver. There are a bunch of tests that compare the new model to the old approach, and the alignment of one layer of wafer 53 looks fine. Still, it would be nice to compare the computed alignment with a previously computed one. Is there any quick way to do this?