Starnet++ within Siril does not remove all stars from a comet stack

Software Siril
1

Siril version 1.20 (version 1.21 has other issues) using starnet on a sequence leaves a lot of faint star trails.
I am using Siril on Mac Catalina.
The starless stack looks good until I stretch it (Histogram auto stretch or asinh). Then many faint star trails show up.
I had hoped that performing the star removal on the sequence would produce a better result.
Here are some of my steps

  1. Perform Background Extraction and PCC on the reference image, and save that change.
  2. Register the stars with Global Star Registration
  3. Stack that new sequence (stars)-has comet streak
  4. Perform a comet registration-
  5. Stack that result-has expectedly all streaked stars
  6. Perform Starnet star removal on the star registration sequence
    • Used the opened (linear) reference image from the sequence
    • Let starnet do the stretch
    • Tried with and without 2X upsampling, no difference
    • Selected “apply to Sequence”
  7. Opened the starless comet image to stretch.
  8. Any amount of stretching reveals numerous faint star trails.
    I do have a properly processed star mask to use for star recomposition.
    The stretched starless comet image requires a great deal of further processing (in Affinity Photo) to minimize the star trails and sharpen the comet.
    What else can I do to prevent getting those faint star trails?
    Fred
2

StarNet in Siril works as StarNet without Siril. Because Siril just call StarNet. Are you sure to check the stretch option if your image is linear? Without screenshot, always difficult to imagine, but what you are saying is that StarNet do not remove enough star. Maybe it is an issue you should report to StarNet dev.

3

Do you mean quite broad, faint star trails? Something like this, from my image of 2022 E3 ZTF:

image
image852×571 46.5 KB

If so, that appears to be a limitation of Starnet: the stars aren’t perfectly removed, and when the comet-registered starless sequence is stacked there is just enough similarity between the residuals of the stars in successive frames as they move through the image that rejection doesn’t cope very well. You could try experimenting with different rejection methods and sigma values, and have a look at the rejection maps to see what is being rejected, but ultimately Siril is doing what it’s supposed to with the data it’s given so fixing the issue properly depends on improvements in Starnet.