D28767: SEP Focus improvements

[Hy Murveit]

murveit added a comment.


  **Addressing your comment for line 789 of fitsview.cpp.**
  //> I disagree : it was like this before, but small stars ellipses become offset because of rounding issues on the pixel center.//
  
  See the images in this google photos album: https://photos.app.goo.gl/CHgXVZBREA19LVgN8
  
  Note that using the sep 'width' that is calculated is very inconsistent. It is slightly larger than twice the HFR (I don't mind that, I could go with that if you strongly preferred it) but occasionally it is way too large. It is not at all hard to find these aberrations. If you wanted wider circles, I could use 3xHFR or something like that, but the width seems wrong too often. I guess the width somehow includes the entire flux above background, which can be fooled by fluctuations in background or something like that.
  
  **Re you comment on fitssepdetector.cpp line 135:**
  //> What if 0.8 of the source set means 0 sources? Is the remaining algorithm safe? Shouldn't we discard sources based on their distance in sigma units instead? (that value may not be available, sure)//
  
  I guess I was planning on waiting on Rob's investigations to get "the perfect" parameters (e.g. 20% vs 2sigma). Top 20%  is very rough, but it tended to do the job. As Pit observed in the forums, the HFR should be similar for all stars.
  The code seems safe as long you can safely sort a 0-length container, which should be ok. If you prefer, we could put this here:  if (catalog->nobj == 0) goto exit;
  
  **Re more test fixtures:**
  //> I would also like you to add more FITS fixtures captured with the CCD Simulator or real pictures, with various FWHM values (use the focuser simulator) to verify your improvements.//
  
  I can add a few other fits files here with tests.  One of my subs is 30Mb, seems too much for this? Don't have any defocussed ones saved, perhaps I'll shoot some tonight (and crop them).

REPOSITORY
  R321 KStars

REVISION DETAIL
  https://phabricator.kde.org/D28767

To: murveit, mutlaqja, lancaster, TallFurryMan
Cc: kde-edu, narvaez, apol
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-edu/attachments/20200412/f9609d63/attachment.html>