pixel rejection levels when stacking

Hello, I am new to Siril,was using Deep sky stacker before, but I like Siril a lot because of every bit you can tweak to get better results. Following the documentation, when stacking the bias, darks and flats with Windsorizing algorithm with default sigma I get rejection levels about 0.000% - 25.xxx% but recommended values are below 0.5%. It goes down a bit when I rise high sigma but nothing near the 0.5% so I switch to Median Sigma Clipping or linear fit clipping (with default sigma values of 5), rising the high sigma between 4 and 5,so I get below 0.5%

Is there anything I am doing wrong? Or doesn’t that matter too much?
For now I am taking photos with a simple entry-level dslr and lens without tracking, single exposures of 2-3 seconds.

Thanks in advance.

The doc must be updated at this point (@argonothe) . Because there is no reason trying to stay below 0.5%. Of course, 25% is too strong, but 0.5% means nothing.

I think you are calibrating your dark? And you have subtracted your master-bias? Is this the case?

Yes, my work flow is subtracting master-bias from flats and darks, and preprocess lights subtracting master bias and dark optimisation enabled.
So below which value should I stay? Maybe below 5 or 10%?

The problem then is because your bias has probably a too high level.
If you don’t subtract bias from dark and you don’t use dark optimization it should be fine.

There is no absolute answer. In fact you could also combine your darks without pixel rejection.

Ok, thank you very much. I have one more question about that. Sometimes at the last step when stacking the processed and registered lightsI get negative values like -2% - -10%. I would suppose that instead of rejecting it was adding pixels and something went wrong ?

Thanks and keep up the good work and great support .

Negative values? It is not possible.
Would like to see your logs.

Which version of Siril?

Siril version: 0.99.6
Log language is in german but I can repeat that and output it in english.2020-10-17T10.57.37Z.txt (136.3 KB)

Yes please. In english that would be better. Very strange!!!

So, I retried. Stacking 20 percent weighted fwhm with linear fit clipping resulted in normal values, but 50 percent left me with negative values. From the resulting image it looks like registration rotated some images differently, must have taken them not as centered as the other ones I think.

2020-10-18T12.34.30Z.Txt (19.4 KB)

That’s something we don’t see every day.

1 Like

I can’t reproduce with a large set of data.
However, I don’t understand how it is possible.

A few nights ago I did some imaging of Orion and the Pleiades. Will process them today, let’s see the result.
By the way the script capabilities are great, you can save so much time not having to do the Steps manually. If I can make a request, I would like to be able to select another algorithm than winsorized for stacking, like sigma clipping, that would be awesome.

Thanks again!

Already done in dev version.

I don’t know what you think but for me it is not possible to get negative values …
The only way to do it is to get a variable overflow. No?

irej[channel][0] / nb_tot * 100.0

while irej is unsigned, and double nb_tot = (double)(naxes[0] * naxes[1] * nb_frames);
That makes me crazy.

We’d really need to reproduce it to understand what’s happening I think. Maybe it’s a compilation issue…

Indeed. Which packages do you use?
Windows? Linux?

I am using Windows only. Currently running windows 10, and on a windows 7 laptop too.

If it helps I could upload the files I used so you can reproduce it.

If it’s not a huge dataset I believe that would help indeed! Thanks

No, only the r_lights are necessary I think. Will upload them tomorrow on Dropbox or similar.
I made another stack of Orion with over 500 lights, even greater negative values are the result xd
Exif: 1.6sec exposure, iso800, f4.5 sigma 70-300mm
Visually the result seems fine, I can see good details of Orion taking into account that no tracker was used and exposures are very short.

We have 201 parallel blocks of size 59 (+59) for stacking.
19:18:32: Starting stacking…
20:38:16: Pixel rejection in channel #0: -6.578% - -737.116%
20:38:16: Pixel rejection in channel #1: -5.298% - -615.811%
20:38:16: Pixel rejection in channel #2: -7.791% - -689.547%
20:38:17: Rejection stacking complete. 524 images have been stacked.
20:38:18: Integration of 524 images:
20:38:18: Pixel combination … average
20:38:18: Normalization … additive + scaling
20:38:18: Pixel rejection … linear fit clipping
20:38:18: Rejection parameters … low=5.000 high=5.000
20:38:21: Background noise value (channel: #0): 0.101 (1.535e-06)
20:38:21: Background noise value (channel: #1): 0.084 (1.288e-06)
20:38:21: Background noise value (channel: #2): 0.088 (1.341e-06)
20:38:27: Saving FITS: file stack_result.fit, 3 layer(s), 6034x4012 pixels
20:38:33: Execution time: 2 h 03 min 28 s.