These are chat archives for fiji/fiji

14th
Nov 2017
tpietzsch
@tpietzsch
Nov 14 2017 14:19
fantastic! @rimadoma, more material for us in the hackathon :-)
Richard Domander
@rimadoma
Nov 14 2017 17:14
@tpietzsch yes, we can try to wrap it all into a neat little Christmas package ;)
@kapoorlab I noticed that sometimes the matrix being solved in yuryPetrov doesn't have an inverse, only a pseudoinverse matrix. Is this why SVD succeeds v. cholesky decomposition?
Richard Domander
@rimadoma
Nov 14 2017 17:23
@kapoorlab Also other notes on yuryPetrov:
  1. Doesn't guarantee that the quadratic surface solved is an ellipsoid.
  2. The eigenvectors of the surface may form a left-handed basis.
Varun Kapoor
@kapoorlab
Nov 14 2017 17:35
@rimadoma Hi, Yeah I prefer SVD coz taking inverse there is just taking transpose of some orthonormal matrices, and for zero eigenvalue matrices it does a pseudo inverse, about your other points is there a reference to better understand the inner workings of yuryPetrov? Please share it with me if you have it.
Varun Kapoor
@kapoorlab
Nov 14 2017 17:42
I wrote a recursive Ransac finder for ellipsoids and tested it on a 2D data: https://github.com/kapoorlab/Bubbleator, the 2D test image with overlays found: ellipse, this is using SVD and bisector method, I would not expect much difference if root finder was Newton Raphson.
Richard Domander
@rimadoma
Nov 14 2017 17:45
@kapoorlab I'm pretty sure the method is based on this paper: https://www.researchgate.net/publication/4070857_Least_squares_ellipsoid_specific_fitting
@kapoorlab The original code itself has very few comments, and the points mentioned above I figured out myself the hard way ;)
The original yuryPetrov I mean.
Richard Domander
@rimadoma
Nov 14 2017 17:55
@kapoorlab Maybe you'll find the documentation in my implementation helpful. See classes SolveQuadricEq and ellipsoid.QuadricToEllipsoid.
Varun Kapoor
@kapoorlab
Nov 14 2017 18:00
Wow thanks, I will look at it Asap. :+1: