Subframe pre selection

Hey there.

I am wondering.

I'm in the process of aquiring quite some data on the bubble nebula, including all subs I'm at around 25h and planning to add the following night.
In the past I trusted in WBPP and just went with it, but I noticed, even if I only add a smaller percentage of sub par subs with nice looking nebulocity but slighty potato shaped star, the stack ended up pretty potatoish.

I'd love to know how you work with WBPP.
Do you just exclude the really bad subs and run with it or are you only using your best subs for the stack?
I feel like using only the best subs with perfectly round stars (tracking went great, most subs looking good), but I'd end up with less signal.

I used the option in blink to check the quality of the subs and noticed, that the sub with the highest score was a potato one.
So I guess I have trust issues when it comes to WBPP.

Also I dislike the whole deco, starnet process. Stars always end up looking iffy and I'd prefer leaving the stars as they are.
Since I'm using a 3nm filter they are pretty tight anyway.

What would come out better, deleting the obvious bad subs manually but use as much subs as possible or pre selecting only the top subs.
Quality vs quantity.

I feel like I collected a lot of great data and want to get the best out of it.

Really looking forward to your answers.

CS
Sascha

I use AstroPixelProcessor for registering and stacking.
I use all subs except the ones I manually delete (e.g. when a cloud rolls in or so and the sub is bad).

Depends how I feel… If I'm in a bad mood few subs survive the blink…
Don't write BXT off for stars - use it with 0.00 for stellar sharpen and halo adjustment and it will still improve those potatoes significantly…

Given my location, I don't have many clear nights. Regardless, I've found that I get better results by being selective and aggressively culling the data I use.

First I use Blink to check for obvious issues with any frames and discard any problematic ones (e.g. a large plane flying over with lights etc).

I then use SubFrameSelector and mostly check PSF Signal Weight (the higher the better), FWHM (

.0), and eccentricity (<~0.5, my flattener corrects the field very well so you may choose to be more tolerant up to 0.6). Make sure you input the proper subframe scale in arcsecs/pixel.

Whatever doesn't get exluded goes onto WBPP.

Hope this helps

I never used subframeselector, gotta give it a try.
very helpful, thank you.

Maybe I try 2 stacks, one where I strictly exclude everything with slight issues manually and one with tools only and look how many subs survive each procedere and how the stack looks like in comparison.

I get the feeling that I stacked too much bad subs in the past and WBPP wasn't sufficient enough to account for it.

I started using subframe selector recently and it's been a great way to find and toss problematic frames. My pre-WBPP workflow is the same as Ped describes above

I just imaged the bubble nebula. Great target. I just blink and toss obvious star trails and cloudy subs. Don't use subframe selector. 25 hours?! I am really looking forward to seeing that result!

I remember, there was a video from Adam Blocks on youtube. He described the functionality of the then new normalize scale gradient script in PI. He showed an example, which reference frames WBPP would choose compered to the nsg script. The weights of the reference frames were night and day. It's not that one is either good or bad. They are choosen with different goals in mind. So WBPP might not always choose the one you would use.
Now it's up to you to decide wheather thats a good or a bad thing

.

CS Christian

I do pretty much the same as Ped above, but possibly a bit more relaxed as the seeing in the UK is usually pretty tragic. We've had such mixed cloud cover recently (at least not rain!) so I've as an absolute minimum had to Blink subs, rejecting visually the obvious ones... I then load the rest into SFS and:

Remove subs where the noise is increasing exponentially (usually around 15 mins before end of nautical night)
Remove low PSF Flux outliers to combat high cloud not necessarily obvious via Blink
Remove subs with high FWHM AND Eccentricity, usually loading up and overlaying the offender with a good sub and flicking between the two to check how bad the difference is and decide from there - let BlurX deal with the lower FWHM and higher eccentricity subs later after stacking
Chuck the rest into WBPP and let it do its magic!

Hmm blurx on subs before stacking?

Hmm blurx on subs before stacking?

No sorry, I meant let it deal with it after WBPP! Edited the post...

Might work before stacking ¯\_(ツ)_/¯

I generally first blink the frames rapidly and then do another selection in SIRIL, where I reject the "strange" frames (Bad FWHM and roudness)

I used to blink all frames and subsequently applied the subframe selector to sort out according eccentricity, FWMH and SNR. This is followed by WBPP.

This is tedious and takes processor time. So out of interest in did a comparison on a large number of frames. Just WBPP, no blink and no subframeselector. To my surprise the less effort approach was also fine.

Worthwhile to note that my mount is an old lady, having a not flat guiding graph at all times. Location is inside a big city.

Still I mostly apply subframe selector.

Give it yourself a try on say 500 frames.

Dirk

Hey folks,

thank you so much for all the replies.
I fiddles around for 6 hours yesterday and stacked 3 different approaches.

1. Blink through every single of my 410 frames and have a close look at every frame, with zoom and star check and everything.
It took quite a while, I was very strict with eggish stars and contrast.
Only 200 frames survived.

2. Blink through rapidly, only excluding obvious subs with satellites, planes, clouds, trailing.
Then I used subframe selector with FWHM, Stars, Median, Eccentricity.
334 frames survived.

3. Blink through rapidly, then go straight to WBPP
392 frames survived.

I was confident that 2 and 3 will result in eggish potato stars, the result surprised me.

1. Looking good, perfect round stars, decent signal.
2. Better signal, stars still looking close to perfect, very slight potato if you pixelpeep.
3. Similar signal but stars looking noticably potatoish when zoomed in.

Number 2 is my favorite, doesn't take long, you get the rejection automated but can decide to what degree you want to reject frames.
It shows that WBPP is really good but not perfect, or I just have to get better at instructing it correctly.

I will stick to Number 2 for now.

It took forever to stack 30h of data 3 times, but it was worth it, finally I got this sorted out for me.
Now I have to process the 28.7h master light which will take some time, with so much effort putting into a single pic the pressure is on

CS
Sascha

I used to perform the preprocessing "by hand" without WBPP: starting from Blink to discard the guiding errors and clouds and then going through all the pipeline, using various formulae for rejecting/accepting/giving weight to each frame in SubframeSelector. It was a lenghty and complex process.
When I switched to WBPP, I realized that all this effort was not worth since I get same (or even better results) with a very simple procedure: use blink to discard frames with guiding errors and high clouds and then give WBPP the responsibility of discarding and weighting the non filtered frames.

I think subframe selector is a really straight forward process, if you get the hang of it it takes like 10min. max.
For me it produced better results than wbpp only.

wbpp does work really good but it annoyed me that I got more ecentricity than I like in the final stack,
with the sfs I have more control.

elongated stars drive me nutts

I also use AstroPixelProcessor (APP) for the stacking chore. The interface for doing so is more to my liking and the data sorting and mapping features make it easy to eliminate bad ones. One of the data points is indeed how elliptical the stars are. I can sort by that criterion, which produces a graph, and I can delete any below a threshold I choose.

Dave Rust:
I also use AstroPixelProcessor (APP) for the stacking chore. The interface for doing so is more to my liking and the data sorting and mapping features make it easy to eliminate bad ones. One of the data points is indeed how elliptical the stars are. I can sort by that criterion, which produces a graph, and I can delete any below a threshold I choose.

Yes I heard this is a great tool as well, but for now I like to focus on pixinsight.
Eccentricity is such a valuable option, unfortunately I became a pixel peeper and I hate if the stars in my final image are elongated.
With this tool I can finally decide to what degree it is acceptable for my liking and how much signal I'm willing to sacrifice in order to get perfect stars.

I have to say, I’m lazy.
To my dismay, it seems like WBPP doesn’t discard anything, and my narrowband masters come out with different sized stars. User error to be sure, but It would be nice to have an automatic way to just exclude highly eccentric or boated star frames. I seriously don’t want to do it manually. (And I don’t want to know all the math to write my own script)

As “powerful” as pixinsight is, my god it feels like I’m back at work, dealing with java developers and a million “fixes” again.

I’m going to have to look into sfs.

I use NINA and then export the image history as a CSV file.

I used to check the CSV and discard subs based on:

- Total RMS.
- Eccentricity.
- Median value compared to the best frame.
- Number of stars ratio compared to the best frame.
- FWHM

After that initial check, I blinked the rest of the subs.

It was so time consuming, so at the end I wrote a python script that does all the previous steps (all threshold values are parametrized). The "bad" subs are moved to another folder so I can check for false positives.

I know there are tools for doing that, but I just wanted to build something by myself just for fun.

I was planning to upload the script to Github under GPL license. So, if anyone wants to try it, no problem about that.

CS.

Guiem.

Is there an “easy button” way to do this?!?!?

One change I made recently has simplified my life a bit. I set up the weighting formula to favour star fwhm, eccentricity, and number of stars. Then under integration parameters, I set the minimum weight to 0.40 (default is 0.05). Before making this change, WBPP never rejected any subs. Now I see rejected subs and much better results. I no longer use subframe selector.

Guiem Kimi:
I use NINA and then export the image history as a CSV file.

I used to check the CSV and discard subs based on:

- Total RMS.
- Eccentricity.
- Median value compared to the best frame.
- Number of stars ratio compared to the best frame.
- FWHM

.....

I know there are tools for doing that, but I just wanted to build something by myself just for fun.

I was planning to upload the script to Github under GPL license. So, if anyone wants to try it, no problem about that.

Guiem.

Hello Guiem...
Looking to do something similar. Would like to preview and autoreject subs as well as factor which registration and stacking script to best use based on these type os parameter. I only image small galaxies and want more consistent image quality. Interested in seeing what you have put together if you are willing to share. Did you post on Github... All for fun and enjoyment....

Ken

I just manually pick out and delete the subs with severely streaked stars (like, anything more than 1.5 stars width, deleted)

also, the subs when my telescope is pointed at the fence or my house, and the subs with clouds, no matter how high they are, no clouds for me

AND, the subs taken with the worst seeing, like, the FWHM is just way too large for me to accept

all this is manually (and quickly) done with ASI Fits viewer, then the remaining subs are tossed into WBPP

How do you pre select your subs for WBPP