Another esoteric stacking artifact; centers of stars have black or pure-color pixels.

Hey again! I stacked 50% of my 7500 shots of andromeda from multiple nights (no calibration shots) and wanted to see how things fared. It’s a nice result, except for the fact that I can’t plate solve it because of this interesting artifact. See the following FIT file for what I’m talking about:

And this screenshot for a quick peek at what I’m referring to, in every bright star:

Has anyone seen this before? I’m unsure what could have caused it, except for maybe some concept of the stacking results (average with rejection, winsorized sigma clipping, 3/3 low/high) leading to those channels overflowing or underflowing? I just can’t see that.

Hi! Someone reported it before but I never reproduced it… Can you tell us what type of input files or sequence you have? Does it appear only after stacking?

Yes it’s probably an overflow of unsigned integer values.

First, can you see it on your debayerd lights ? Before stacking?

Second, do you stack in 16 or 32bits ?

EDIT: Ok this is 32. Did you applied a process after stacking?

We need to understand at which step do you have this. I would say debayering.

We also need your logs.

OK apparently you did a deconvolution.
So this is really not a stacking effect, but a deconvolution effect I guess.

OK apparently you did a deconvolution.

I did? I don’t recall doing a deconvolution at that stage in the pipeline. I restacked using 50% weighted fwhm and still got a similar result on those stars, sadly. I hope I’d recall if I tried running a deconvolution step on over 7500 registered images (lol).

Can you tell us what type of input files or sequence you have? Does it appear only after stacking?

I don’t know if it matters, but I’m using fits cubes because I can’t work on more than 2048 files at a time, so I just converted all of my data into debayered FITS cubes (since I’m not calibrating), and then used the conversion tab to merge all 3 of these massive FITS cubes (multiple nights of data, multiple ISOs, etc. etc.). I hope that answers the questions. Yes, it definitely only appears after stacking.

Here’s that 50% Weighted FWHM image:

What’s so strange to me is that this did not happen when I stacked 90% of the 7500, but the plot does look like this:
Perhaps there’s something inside these frames on the right that is messing things up? I could spend another day letting it re-register on a different reference image, I suppose.

Update: Maybe I’m a moron?

Here’s a little collage I just made from the brightest star in the debayered, unregistered lights:

Obviously, if I pull up the red channel, those are totally black pixels.
What I’m completely perplexed by is how this happened and didn’t affect the 90% fwhm stack that I made, but it affects any of these 3500-sub stacks.

Another thing I don’t understand, then, is how we got weird effects on that star that aren’t just cyan, but are all sorts of pure combinations of colors. What’s up with that?

Yes you did. It is written in you FITS history.

Perhaps in that last FITS file, but I’m completely uncertain how it would have applied a deconvolution on all of those lights (or the registered lights, for that matter). Is that where you saw it, or in the most recent one? I might have unintentionally saved the FITS after deconvolution during some prior processing, but I didn’t think that SiriL would save those directly when I did so. Does it say exact where that deconvolution was done (before registration)? If it wasn’t really early in the pipeline, I don’t see how the debayering led to that.

Yes there is an overflow in the debayer part. Which algorithm do you use?

The debayer section of the settings menu says I’m using VNG4.

Please try using RCD, the default algorithm, far better.

1 Like