Unoriented sphere der

[ThibaultLejemble]

This PR implements de differentiation of the UnorientedSphereFit.

• the eigen solver and the matrix Q are now stored as class data members because they are needed by the UnorientedSphereDerImpl derived class
• the test fit_unoriented uses the new differentiation but only checks the coherence of the results when normal vectors are swapped. I tried to check that the curvature values are correct, but I couldn't find an appropriate epsilon value that is small enough for the comparison to work...

[jmaleo]

I have conducted a series of experiments using the newly implemented derivative class, focusing on evaluating the coherence of results under different conditions of normal flipping.

The parameters employed in these tests included:

Three different ratios for normal flipping: 0.0, 0.25, and 0.5.

• A variety of radii.
• two iterations of MLS for each test scenario.
• Three levels of noise applied to normals, acknowledging that higher noise levels typically degrade results. It's important to note that the same noise level was maintained across all flip ratio tests to ensure consistency in the comparative analysis.

Upon analysis, I observed that there were no difference in the outcomes across the various flip ratios.

[nmellado]

Do you mean that your experiments validate the estimator (wrt. to those already implemented) ?

[ThibaultLejemble]

Do you mean that your experiments validate the estimator (wrt. to those already implemented) ?

I think @jmaleo was saying that he performed some extra tests that validates the invariance on normal directions.

[ThibaultLejemble]

We believe there is a small error in one of the mathematical equation, I will clarify this, so this is not ready for review and merging.

[nmellado]

Shall we undraft it and review it ?

[ThibaultLejemble]

Shall we undraft it and review it ?

Yes, ready for review!

[nmellado]

Sounds good, can you please clean the history ? then we will be good to go I think.