Large mosaic optimal registration methods

Just found a recent comment from @mabula-admin (reproduced below) which I think probably answers my question, albeit indirectly. Looks like I should bite the bullet and process everything in one go.

-------

Hi Patrick @artaios and @itarchitectkev,

So instead doing a 8x8 I can try doing for 2x2 and add those into another larger 2x2 later to get the same result but with less hardware consumption at a time.

This is something that I can not recommend, the optical distortion correction will suffer in this workflow, the best results will be delivered when you properly calculate the whole mosaic at once 😉

I agree that the OS does not matter really. The hardware does, and more CPU cores/ Threads will simply have a more faster APP :-), for big mosaics, 64GB is plenty at the moment. More is not needed since we still have an image size limit for big mosaics of 2^31 pixels = 2 gigapixels. We will try to remove this limit going forward of course. Wiht 64 GB, you can already create huge mosaics. Finally, the faster SSD drive that you can afford will help as well.

Mabula

Hi Jon, @jdwood,

Thank you very much for your question.

It is a very good question with a rather difficult and technical answer. Like you already found on the forum, it will always be better and more robust if you can calculate the mosaic from the actual individual mosaic panels. The reason is that if you first combine a 2x2 mosaic using, and then use that as a light, APP can not identify and fit a logical optical distortion pattern on this. And the reason is clear why, a 2x2 mosaic will have the optical distortion mostly corrected, but there might be a residual left in the 4 kwadrants and this residual is not a pattern that any optical device would create. APP's distortion correction is actually unique among all Astrophotgraphy software in the sens that it uses real optical distortion models to correct these issue.

If you would have a telescope with zero optical distortion, then your workflow makes perfect sense and should work well. But in my experience, all telescopes have optical distortion, even great astrographs like the Takahahsi FSQs...

Your comment about the overlap for the 4x4 panel also makes sense, but I think that should be overcome if you increase the star count in 3) analyse stars with a factor 4. So instead of 2500 stars, let it try to find 10000 stars and then try to register. Maybe you can try that if you haven't tried yet. The proof is in the pudding 😉

Mabula

Thanks @mabula-admin. I have made significant progress on this and the 64 panel mosaic is starting to look good (albeit still requiring a bit of work to sort out the background mottling): Orion mosaic In fact the registration and integration of the entire thing only took a few hours on an M1 Apple Studio.

One further quick question relating to large mosaics if I may? I know that you have plans to enable APP to work with larger files (no downscaling) in the future but, for the present, what is the best time to do this in order to maintain as much resolution as possible?  

I see that there are downscaling options under several tabs (Register, Normalise and Integrate) and APP will also force a downscale at Integration if the resultant file looks too big - but I also found an old thread where you suggested that the best option for downscaling would be to use the Cubic B-spline interpolation under the Batch/resize tool.

Is Cubic B-spline used for downscaling in those other tabs or is it best to process at full resolution and then use the tool to downsize at the very end (even if APP itself forces a partial downscale during the Integration step)?

many thanks, Jon

Hi Jon @jdwood,

Yes, going forward we should make it possible to use unlimited file size, so unlimited in pixel dimensions theoretically 😉

The Cubic B-Spline is a very good algorithm for downscaling indeed. It has not the sharpest reconstruction as a resample/interpolation filter, but when downscaling, this is less of a problem. The upside of Cubic B-Spline is that it will reduce some extra noise, which makes it very nice for downsampling with for instance 0.5x.

If you know that your mosaic would be downsampled, simply change the filter in 6) integrate to Cubic B-Spline and it should work very nicely.

APP's limit is that the total pixels of the resulting image can not be larger than 2 GigaPixels in mono, or 2GigaPixels / 3 for RGB data. So you can calculate this if needed.

Hope this helps?

Jon, you also had the issue with the stripes in the mosaic result right on macOS? There is a chance that beta32 actually has fixed that issue. Could you test that?

Mabula

Thanks Mabula @Mabula-admin,

That is exactly what I wanted to know.

Yes - I did have a problem last year with fine lines across a large Milky Way Mosaic. I will dig out those old data and see if the latest beta helps.

Jon

Great Jon @jdwood,

I will test the mosaic lines issue myself in the coming days. I just received a new mac with 128GB ram to test it, so I will report back in the topic https://www.astropixelprocessor.com/community/main-forum/strange-horizontal-lines-in-mosaic-integrations/#post-28829 . If you can test it as well, that would be really helpfull.

There was another thread as well regarding this issue from Walter @walsc : https://www.astropixelprocessor.com/community/appreleases/image-artefacts-on-multipanel-mosaic-2-0-0-beta17/

Walter, can you test and confirm possibly as well if this issue with the lines is still there in 2.0.0-beta32 on macOS?

Thanks,

Mabula

@mabula-admin OK, I'll be back tomorrow night with a result, thank you!

Hi Mabula @mabula-admin (cc. Walter @walsc)

I can no longer generate the error (thin lines appearing across the image with specific forms of LNC processing) on the same dataset that previously presented problems. So, as far as this one dataset is concerned, I think that it appears to have been fixed with recent changes? 🙂 

Hi Jon @jdwood,

That is very good news! Thank you very much for testing.

I actually think that this issue on macs with the weird integration lines is directly related to this ArrayOutOfBoundsException that Walter encountered on Windows https://www.astropixelprocessor.com/community/main-forum/how-to-speed-up-registration-process-on-large-mosaics/#post-31013 .

Might sound confusing, but this bug fix for Walter's problem is in the latest release and it actually fixes how all pixels are integrated preventing problems with integrations that approach the 2GigaPixel limit as was with Walter's issue and also with these mac issues because it happens when downscaling is done to keep it just under the 2GP limit.

It is weird that on Windows, an actual error is thrown, while in macOS the error is not thrown but continues and then produces weird results, which from my developer's eye I can understand. I know there are slight differences in how our development platform deals with some issues depending on the platform.

Hopefully Walter can also confirm that this mac weird lines issue is gone. I will test myself with Walter's and your data with beta32 and the older versions to see if I can produce the issue with the old versions and that it is gone with beta32. Than we know for sure 😉

Mabula

Hi Jon @jdwood and Walter @walsc and @nhmorgan79,

I can confirm that the issue with the horizontal banding on macOS with big mosaics and downscaling below 2 GigaPixels is solved in Beta32. All looks and works fine now. In 2.0.0-beta26 the issue still manifested with the exact same settings as in beta32 on Jon's data. I have closed the issue now.

Thanks and please accept my apologies for not fixing this sooner.

Mabula

@mabula-admin Thank you, my confirmation will be given as soon as possible. My Mac is substantially slower with my big mosaic, the image-registration process is not the M1's strength. Result will be ready tomorrow, and I do not see it before 8PM because of my long workshift.

Hi Mabula @mabula-admin 

That is great news! Many thanks for your exhaustive efforts in trying to track this down😀 

Jon

I can confirm it is working, no stripes any more.
It took longer than expected because I had to free up space on the mac.

Posted by: @walsc

↑

I can confirm it is working, no stripes any more.
It took longer than expected because I had to free up space on the mac.

That is awesome Walter @walsc, all my testing has confirmed that the issue is gone with beta32 🙂 so the issue is removed from our issue list now.

Thanks for testing !

Mabula