Drizzle Advice please

An extensive manual is planned for the release of the stable 2.0 of APP. But when you hover with the mouse over the buttons in drizzle integration, you get pop ups with explanation as well, do those explain enough? Smaller droplets will allow for sharper results, but they also will be noisier. They only work well when you're clearly undersampled and have rather big drizzle steps regularly.

@vincent-mod 

Sorry to say that it is helpful but doesn't answer my questions.  Maybe its my senior citizen "learning disability" or something like that. 😆  

The one example in the hover instructions (.5 and 2) is all I see that are actual suggestions.  What about 1.5 and 2, or 1.5 and 1, or 2.25 and 1, etc. What happens with certain combinations?  What will be the results? I just don't understand what numbers to use and when, for what result.

I appreciate you all being there, even if I don't fully understand.

Regards,

Andy

Oh no problem, that's why we are here. 🙂 The lowering of the droplet-value is basically creating smaller droplets, the maths behind this may be a bit much for now, but it basically means that you're trying to fill in data where it previously wasn't due to undersampling. If you dithered very well during data collection, you have multiple points of proper data in that area and with drizzling you can recreate that data on that spot to get sharper results in the end. Lowering the value gives more details (sharper), but requires more and more data that is properly dithered. So what to pick is difficult to say and depends on the data-set, you have to experiment with that.

I reactivate this topic, because it sets a good starting point for my questions related to drizzle. (I really like to know more about what i can achieve in app, aside what i could already read in this forum about drizzle)

If I recollect correctly, drizzle can help getting more out of my undersampled data (if some other requirements are being met, e.g. lots and lots of data, dithering - all things i try to do anyways).

As I assume, my images will be undersampled most of the time. My equipment ranges up to a focal length about 300mm, what's about a minimum of 2,3"/px for my Lumix GX9s sensor (pixel size 3,34um, 5184x3888), quickly getting higher on shorter focal length.

My plan is now to get familliar drizzling at least my widefield images.

Questions:

Is there a rule of thumb, how to calculate a good starting point for droplet size at a given "/px? (I am used to calculate numbers, so if there is some math involved be so kind to let me know!)

Can scaling help getting more out of the given data? (up oder down? Or is that related to droplet size and scaling just scales the results of drizzle?) I would assume that upscaling could help shaping the stars a bit?

How much is "lots and lots of data" (considering i am using a dslr, were noise is an issue most of the time). Or better said: is there a minimum to start with? ("Dont try drizzle with less than 1000 lights" or something like that?)

Is there a better way to compare results, than just try and see, if it looks better? (I assume one has to check for artifacts)

I like to test bayer drizzle but my low focal length lenses suffer from chromatic abberation (so I try to color align afterwards as mabula pointed out in another thread).

Other tips and hints you can give (so I could reduce excessive efforts doing trial-and-error, not knowing what to do).

Ok, don't bother my last post for now!

I found this post of Mabula and he suggests to rather go for the empirical solution:

https://www.astropixelprocessor.com/community/tutorials-workflows/bayer-drizzle/#post-15814

These were very helpful informations:

Posted by: @mabula-admin

↑

This has the consequence, that if you relate Drizzle to Bayer Drizzle, you would double the droplet size with Bayer Drizzle compared to Drizzle to get the same results in terms of image reconstruction/sharpness and noise levels in the result.

So that's why bayer drizzel gave me a "grainier" result than normal drizzle with same settings!

Posted by: @mabula-admin

↑

So, my advice is to take this with a practical approach instead of a theoretical approach. For instance, If you Bayer Drizzle, I would simply keep scale at 1, and make 3 stacks with different droplet sizes, 1.5 pixel, 2.0 pixel and 2.5 pixel. Then study how this affects the results for noise and sharpness.

Ok, I see it depends; you have to try...

So my only question(s) that remains for normal Drizzle:

- If I keep scale at 1, do I go for droplets <1 or >1? (Sorry, I just don't get it)

- And how does scaling relate to droplet size (if I scale up, do I have to lower droplet size?)

Hi @xyfus,

If I keep scale at 1, do I go for droplets <1 or >1? (Sorry, I just don't get it)

With Drizzle: always set the droplets lower than 1.0 😉 to benefit from increased resolution/detail in the result when compared to not drizzling. And then you want to set the scale higher than 1.0. The higher the scale the more resolution can be obtained with lower droplet size. But, the requirements to get decent results with lower droplet size and higher scale is that you need more data/more aggresive dithering to not get drizzle artefacts in the result and high noise levels.

With Bayer/X-Trans Drizzle, the same holds only that to compare with regular drizzling, you need to double the droplet size to compare it to regular drizzling.

And how does scaling relate to droplet size (if I scale up, do I have to lower droplet size?)

A higher scale factor asks more data with a fixed droplet size to get decent results. So if you get a good result with scale 2.0 and droplets of 0,5, then you can try to use droplets of 0,5 and scale larger than 2.0. Or, keep scale at 2.0 and check if resolution in  the result improves with lower droplet size.

Regarding the droplet size, it is a compromise between getting better resolution and more noise in the result. Drizzle injects noise and lower droplets increase noise in your result, so it is a question of what resolution you desire and how much noise you are willing to accept 😉

Mabula

@mabula-admin 

Thank you for your answer, this did clarify some things to me! 🙂

I did run a test for bayer drizzle with a crop of M4 i did take with a 300mm lens on my Lumix GX9 (3,34µm pixel size) so that should be 2,3" / pixel. Seeing wasn't that bad, but since M4 stood roughly above horizon to the south (and about 60km south of my garden lies Frankfurt, so the sky gets really bright at lower angles in this direction), I do think this makes a valid candidate for testing. (I will do another test with a shorter lens and upscaling, which will surely show much more amazing improvements).

371 lights about 221x60s ISO1600 and 150x120s ISO800 on 5 sessions

1) normal integration (with calibrated color aligned lights)

2) Bayer Drizzle droplet size 2 (color aligned afterwards)

3) Bayer Drizzle droplet size 1,5 (color aligned afterwards)

4) Bayer Drizzle droplet size 1,25 (color aligned afterwards)

All no scaling!

droplet size of 1,5 and 1,25 do show for me an aplified noise while stars get rounder/smaller (but little effect overall). I think I have to try and see how much this is affecting the full image. So I will do a full bayer integration now @1,8 droplet size and remove lp (this will show better if the increased noise is tolerable)

edit: first time i did use the cropped integration for comparing different settings. i really dont know why i didnt make more use of this before. you did add so many helpful "little" features, i am really looking forward for that long announced manual!

@xyfus Hi Sebastian,

cropped integration? How does this work? Fist time I hear this!

Thank you in advance 🙂

Hello @juergenn 

On 6) INTEGRATE  panel you can set a composition mode. Standard it is set to "full" you can change that to "crop".

APP then opens your reference frame where you can draw a crop and set it. Now if you start integration it only compares this crop and only integrates this area. So it is very fast.

After its done you can change some setting but leave crop as is. This way you can easily compare different settings.

Sebastian

@xyfus Hidden treasures in APP 😉

Hi @xyfus and @juergenn,

Excellent, yes, the crop composition mode is really usefull to test good drizzle settings on a dataset 😉 Depending on data, sky quality, optics etc, you need to use different settings to get the most pleasing results, which are also subjective I would say. But it is also nice to use crop to modify the actual Field of VIew as shown in the reference frame directly and zoom in on the object.

Mabula