46 replies to this topic

[Smurf]

(I was copy this thread from https://www.cloudyni...ng-image/page-2 hoping someone would notice..)

 

 

Hi. My equipment is:

Telescope: Sky Watcher 200/1200 Dobson, GoTo Alt/AZ mount.

Board: Fysetc E4 by OnStep.

Camera: Samsung Galaxy S20.

 

After successfully seasons of auto-stacking Lights in Siril with Scripts, now I am learning Manual stacking lights, but my stacking not work! I got always this two warnings:
MAD is null. Statistics cannot be computed.
Normalization failed. Check image 156 first.
Stacking failed, please check the log to fix your issue.

 

My Stacking settings are:
Method : Average stacking with rejection.
Normalisation : Additive with scaling.
Rejection : Winsorized Sigma Clipping.
Weighting : Number of Stars.
Image  Rejection: sele cted
Output normalisation: yes
RGB equalization: yes

 

Errors I got in any seasons. Registration goes normal (like 157 of 335, normally for me).

 

Did I try repair what logs suggest me?

Yes. In every stacking show me about 10 numbers image warnings.. I remove (uncheck) that 10 pictures, stack again, and after that it shows me new warnings with other numbers of pictures. That process I can repeat unit ALL pictures are deleted, there is no end! (tested at 50 pictures, all I deleted). So, it seems that Siril not accepts any of pictures with current setting: „Normalisation: Additive with scaling“ (or any other normalisation).

 

WEIRD THING:
How is possible that SCRIPT (OSC without DBF) successfully finish stacking process with included „Average“ and „ Additive with scaling" (in text log)?? I confused, why not work manual? If that settings working with script, that means it must wotrking in manual too, right?

 

METHOD WHICH WORK:
Otherwise, I found a method which works: disable Normalisation, like:
Method: Average stacking with rejection.
Normalisation: None

 

Is it normalisation important and how much? Does it ruin a picture if is disabled?

I think yes, because it seems that script stacking gives me better quality results.

 

Also working Sum stacking and Pixel (min and max) stacking, which have normalisation disabled by default, but it seems to give me the worst results in the image.

 

I remind you that I have ALT/AZ mount, where I have field rotation. My target not all time in season „in center“, I check it every 15-30 minutes and sometimes must recenter position. Maybe because of that normalisation failed.
I saw in log some .log "memory" word, is it problem in memory PC?

With pictures I am sending my main stacking log (which not work) and script stack log (which work) to compare.

 

Attached Thumbnails

• 
• 

Attached Files

•  logs.zip   127.2KB   3 downloads

Edited by Smurf, 11 January 2025 - 05:01 AM.

[archiebald]

It may not be related but a while ago I was having problems with failed stacks that I finally solved by switching the Pixel Rejection method from "Winsorized Sigma Clipping" to "Linear Fit Clipping".

 

I don't fully understand the reason why, but it works, whereas the same data fails with Winsorized .Sigma

 

According to Siril documentation, the Linear Fit algorithm "performs very well with large stacks and images containing sky gradients with differing spatial distributions and orientations".

 

EDIT -

=================

Full list of methods.... from here... https://siril.readth...g/stacking.html

 

 

Rejection methods

• Percentile Clipping: This is a one step rejection algorithm ideal for small sets of data (up to 6 images).
• Sigma Clipping: This is an iterative algorithm which will reject pixels whose distance from median will be farthest than two given values in sigma units(low,high).
• MAD Clipping: This is an iterative algorithm working as Sigma Clipping except that the estimator used is the Median Absolute Deviation (MAD). This is generally used for noisy infrared image processing.
• Median Sigma Clipping: This is the same algorithm as Sigma Clipping except than the rejected pixels are replaced by the median value of the stack.
• Winsorized Sigma Clipping: This is very similar to Sigma Clipping method, except it is supposed to be more robust for outliers detection, see Huber's work [Peter2009].
• Generalized Extreme Studentized Deviate Test [Rosner1983]: This is a generalization of Grubbs Test that is used to detect one or more outliers in a univariate data set that follows an approximately normal distribution. This algorithm shows excellent performances with large dataset of more 50 images.
• Linear Fit Clipping [ConejeroPI]: It fits the best straight line of the pixel stack and rejects outliers. This algorithm performs very well with large stacks and images containing sky gradients with differing spatial distributions and orientations.

Edited by archiebald, 11 January 2025 - 05:49 AM.

[rj144]

How did you get to stacking though?  In order to stack you need to calibrate and register your pics.  How did you do that?

[Smurf]

@archiebald
I tried your method:
Method : Average stacking with rejection.
Normalisation : Additive with scaling
Rejection: "Linear Fit Clipping".

But still the same, while "normalisation" is ON, can't stack.

@rj144
Registration went successfully, you can see in the registration .log which I posted in my first post.
And we don't do calibration step only for lights, right? I skipped that.

Edited by Smurf, 12 January 2025 - 06:16 AM.

[Smurf]

Somebody knows maybe?

 

I can send like a part of my raw season if someone can try to stack them..

 

Or I will be forced to switch to the Deep Sky Stacker.

[lock042]

Did you check you calibrated images as suggested by logs?

[Smurf]

Did you check you calibrated images as suggested by logs?

I wrote in first post:

"Did I try repair what logs suggest me?

Yes. In every stacking show me about 10 numbers image warnings.. I remove (uncheck) that 10 pictures, stack again, and after that it shows me new warnings with other numbers of pictures. That process I can repeat unit ALL pictures are deleted, there is no end! (tested at 50 pictures, all I deleted). So, it seems that Siril not accepts any of pictures with current setting: „Normalisation: Additive with scaling“ (or any other normalisation)."

 

So, all pictures what logs suggest me to check, must be all deleted.

But pictures are not looks so bad.  

 

I don't know how script accepts all this and finishes stacking well with normalization.

Edited by Smurf, 18 January 2025 - 05:58 AM.

[lock042]

Don't delete them. But load a calibrated image and apply statistics on it.
Send a screenshot of the entire screen

[Smurf]

 But load a calibrated image and apply statistics on it.

I am not sure how do you mean "calibrated" image, and which statistic, from .log or..

 

Ok, I made it. Actually I found one of better season and I stack lights manually. I thought that I tested all seasons manually, but I was wrong.

It seems that I have more bad seasons which only work with script, but not manually.

 

This spring I will try to play with the darks included too.

Attached Thumbnails

• 

[lock042]

When I say entire screen, that means literally the whole screen...
Then here you did not chose a calibrated file. Please, if you want help, follow what I asked.

If you don't know what is calibrated files, type calibrated on the search entry of the documentation. It will help you.

[Smurf]

When I say entire screen, that means literally the whole screen...
Then here you did not chose a calibrated file. Please, if you want help, follow what I asked.

If you don't know what is calibrated files, type calibrated on the search entry of the documentation. It will help you.

Actually, that picture above was not the answer to your question. It is a log picture from another 'better' season which I recorded. And finally stacked light successfully.

 

Now I created a picture, if that what you're looking for.

Attached Thumbnails

• 

[lock042]

More than 50% of your image is black. This is why MAD is null. Cropping the sequence should help to process the image in the circle.

[BlueMoon]

Pardon my asking, but seeing MAD mentioned brings to mind a question I've had. The description suggests MAD is useful for "noisy infrared image processing". In using MAD for pixel rejection, are there other appropriate use cases outside of infrared? Just curious. Cheers.

[lock042]

Pardon my asking, but seeing MAD mentioned brings to mind a question I've had. The description suggests MAD is useful for "noisy infrared image processing". In using MAD for pixel rejection, are there other appropriate use cases outside of infrared? Just curious. Cheers.

Don't get me wrong. MAD rejection is indeed recommended for what you're saying. However, here, MAD is just an underlying algorithm for calculating a value needed for normalization. However, MAD has a 50% breakpoint. So, when more than 50% of pixels are outliers, this estimator is no longer robust.

[BlueMoon]

However, MAD has a 50% breakpoint. So, when more than 50% of pixels are outliers, this estimator is no longer robust.

Ah, I see. Thank you for the explanation. Cheers.

[Smurf]

More than 50% of your image is black. This is why MAD is null. Cropping the sequence should help to process the image in the circle.

Thank you. Good try! But not work.
I was going in export sequence. That let me do a crop sequence.
First I cropped just around the black circle. Export. Load cropped sequence. And same error in stacking appear.

 

Second time I cropped a little more "in the circle", and same error.. Statistics was "little better"-got MAD 1 value only on Green channel(think). Red/blue nothing.

 

Ok, please check my this two .gif files:

• Aggressive season .gif (stack not work): https://ibb.co/CwSSp8R
• Calm Season .gif (stack work): https://ibb.co/VVrJQrf

(Those gifs are only about 60 of 400 frames trough full season.)

 

You will see a big difference where aggressive 'jumping' too much in all directions and hard "tilt" some picture. Why is that? Think that is a problem.

 

Otherwise, in DSS that not happened. Aggressive season is always "in place" (centered) and no tilting there.

Edited by Smurf, 20 January 2025 - 06:15 PM.

[belliott4488]

Note to all: after a bit of heads-scratching I've guessed that when Smurf, the OP, says "season" they actually mean "session". Maybe?

 

HTH

[Smurf]

Note to all: after a bit of heads-scratching I've guessed that when Smurf, the OP, says "season" they actually mean "session". Maybe?

 

HTH

Yes, you are right. Sorry, my bad english. Tnx

Edited by Smurf, 20 January 2025 - 06:35 PM.

[lock042]

Share your images please.

[belliott4488]

Yes, you are right. Sorry, my bad english. Tnx

I would say that your English is 99% good - this was just part of that 1% that wasn't quite right! 

[Smurf]

Share your images please.

Tnx.

Here are raw files:

https://easyupload.io/qggz6t

(File will be auto-deleted at 20.02.2025.)

Edited by Smurf, 21 January 2025 - 05:37 PM.

[rj144]

Just randomly picking a few of your files, your stars aren't "stars":

 

 

 

 

Your stars need to be points, not lines, or squiqqly things to stack.

 

How are you taking these pics?

 

In the stacking with the script, it says there are 405 pics, but in the folder you uploaded there are almost 500 pics.  Are these the same pics as what you ran through the script?

Edited by rj144, 21 January 2025 - 09:41 PM.

[Smurf]

Registration 492 pic, 355 registered, 157 failed.
Script Stacking 283 stacked, 209 failed.
Light.zip 492 files.

 Are these the same pics as what you ran through the script?

Yes, you can try to run the script, around a half must pass.

 

I know that sessions have bad stars shape, but it is one of the better m27 what I recorded (but don't have many).

 

I was taking these pics with my phone (see signature) and Ex.15sec ISO 50.

 

What I figured out is when you use multiple registration, any next time number of pictures decrease, so like ex. 492 at start registration, 380 at next registration, 360 at next... all to the fixed number where registration finished 100%. So, maybe that's way show 283 stacked to me.

Edited by Smurf, 22 January 2025 - 11:50 AM.

[Smurf]

@lock042 What you say?

[lock042]

With some adjustments in the star detection, I get:
14:10:37: Total: 5 failed, 222 registered.

14:10:37: Execution time: 32.29 s

Then I cropped the sequence as I told you.
After that, only one image has a NULL MAD. I unchecked this image from the processing.

Next, stacking worked (I normalized the output).

After stacking, I cropped the image to remove all black edges.

Next, background extraction, manual color calibration, starless and GHS on it (+saturation with GHS).

Finally, star recompilation.

 

All done in Siril

 

Attached Thumbnails

•