D28767: SEP Focus improvements

Hy Murveit noreply at phabricator.kde.org
Sun Apr 12 21:26:49 BST 2020

murveit added a comment.

  Thanks. I'll discuss your other comments in a bit, but first here is detail on the speedup. The basic idea is that previously the code was calling sep_flux_radius, an expensive call, on all detections unnecessarily. Instead, I first sort of the detected stars' oval size, since we are only keeping the largest stars, and only call sep_flux_radius on the stars we intend to keep. I checked and sorting by HFR after calling sep_flux_radius on all stars vs first sorting by oval radii and only calling sep_flux_radius on those give almost the same set of stars. 
  The extent of the speedup is directly related to the number of detections from sep_extract, and how much larger that is than the desired number of stars (I kept that at 100). So, on the "double stars" image, with 11K detections, the speedup is an order of magnitude. On the cropped focus image, with only 3 stars detected, there is no speedup. In general, on full-frame images, we detect more than 100 stars, even with short exposures, so there is a speedup. For my full-frame images it tends to be about a factor of 2.
  Here are the images I've tested with:
  fullframe: a full-frame 4656x3520 gocus image, green 5s
  cropframe: 192x192 focus image, green 5s
  doublestars: a fullframe sub from an RGB camera, 300s exposure
  sunflower: a fullframe 4656x3520 sub from a monochrome camera, 240s Red filter
  They can be found in https://drive.google.com/drive/folders/1jlwxCh7J27_jmQZugD8RgB1ExY276XPP?usp=sharing
    QBenchmark on fixture: 252ms
    doublestars. SEP: 9.470s (11208 detected 100 final)
    sunflower    SEP: 0.864s (680 detected 100 final)
    fullframe:   SEP: 0.646s (427 detected 100 final)
    crop frame   SEP: 0.006s (3 detected 3 final)
    QBenchmark on fixture: 119ms (273 detected 100 final)
    doublestars. SEP: 0.939s (11208 detected 100 final)  
    sunflower    SEP: 0.473s (680 detected 100 final)
    fullframe:   SEP: 0.410s (427 detected 100 final)
    crop frame   SEP: 0.005s (3 detected 3 final)
    QBenchmark on fixture: 1,739 msecs (273 detected 100 final)
    doublestars. SEP: 66.464s (11207 detected 100 final)
    sunflower    SEP: 5.623s (680 detected 100 final)
    fullframe:   SEP: 4.180s (427 detected 100 final)
    crop frame   SEP: 0.081s (3 detected 3 final)
  RPi4 4Gb Raspbian WITH SPEEDUP
    QBenchmark on fixture: 818ms (273 detected 100 final)
    doublestars. SEP: 6.187s (11207 detected 100 final)
    sunflower    SEP: 2.424s (680 detected 100 final)
    fullframe:   SEP: 2.363s (427 detected 100 final)
    crop frame   SEP: 0.083s (3 detected 3 final)
  Also, note the large difference between a fast macbookpro and an RPi4 4Gb!

  R321 KStars


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/e2786e74/attachment.html>

More information about the kde-edu mailing list