Can you tell me exactly what options you use to set APP up for creating the mosaic?

@vincent-mod Honestly, I've tried so many that I don't think I remember them.  Since I was just integrating two panels, I started with normal registration mode (and later tried mosaic), scale-start of 5 and scale-stop of 12 (and later 15), followed the instructions to enable/disable dynamic distortion correction/same camera & optics, tried registration mode with quadrilaterals and triangles, advanced normalization, MBB 20 (for 20% overlap between panels), etc.  Unfortunately I started on another project and no longer have the settings in front of me.

Thanks for trying to help!

No worries! If you still have the panels I can give it a go as well. You can upload like 10 raw frames per panel and calibration files to our server if you want:

@vincent-mod I've uploaded the individual mosaic panels that were producing the horizontal lines, but for whatever reason it's not doing it now.  I had 6 attempts to create that mosaic and they all had lines to one degree or another, but now it doesn't.  Hmm.  I had since quit and re-launched APP, but I think I'm using all the same settings I was during the first 6 attempts.

If you still want to take a look, I uploaded the panels I just stitched to /AstroPixelProcessorUpload/MountainAir_MosaicLines.

Thanks for taking a look!

Well, it may be a data connection in that case, like an issue at that time transferring or reading the data. You are processing on an internal (preferably SSD)?

@vincent-mod Yes, internal SSD.  However, this happened several times so it wasn't just a one-time read issue.  The lines changed between integrations as well.  I must have changed something that made it better, but I have no idea what!

Very interesting, still seems more hardware related given it's so random. Anyway, if it happens more often again, please let us know and we'll dive into it.

Vincent, sorry to bother you with this again.  I tried a couple more mosaics, and I get these lines about half the time.  This one seems a little different, almost like the top and bottom frames didn't have enough overlap (you can see the individual subs in the top frame) -- but if you look at each panel individually, they did have plenty of overlap.

Note, these mosaic panels were imaged with NINA's framing assistant wizard, set to 20% overlap.  The panels were stacked separately, with same camera/dynamic distortion on, MBB set to 5% and LNC1 3 iterations.  The panels were registered with and without mosaic mode, triangles and quadrilaterals, scale start 5/stop 15, same camera & optics off, adv normalization, and no LNC and 20% MBB.

I'm uploading the two panels, and the integrated mosaic (which I tried in both normal and mosaic mode), to the same MountainAir_MosaicLines folder.  ETA 20 minutes.

Oh, and I should note -- the lines change only when I change options.  If I re-run with the same settings, they seem to be the same.  Also, I have never had any errors opening files on this relatively new MacBook Pro M2 Max, and my disk utilities say everything is fine.  I can open 4GB files repeatedly, so I don't think it's hardware.

Thanks in advance, I'm dying to know what I'm doing wrong.

Well, it might be a bug ofcourse, so not necessarily something you did wrong. 🙂 If you could upload a set of data, raw lights and calibration frames that show the issue I can properly test it. Thanks!

I'm in a hotel with a very stingy limit on upload bandwidth.  The stacked panels and stitched panel in MountainAir_MosaicLines took an hour and a half!  I'll pick a couple subs of a single channel plus masters, make sure the issue persists, and I'll let the upload run overnight.

Ah ok. 🙂 I hope it succeeds!

So I looked at the panels, I think those 4 are all (please inform me when data is uploaded fully so I don't miss it, thank you)?

Checking first the registration (by only starting registration first and then switching to the registered mode on top of the preview image panel), I noticed dynamic distortion correction not working properly. So I switched that off and reran registration and this time it came out correct. Beautiful data btw:

I used the latest beta of APP.

I am seeing similar lines with my mosaic integration. These appear quite thin with LNC 1-3 but thicker with LNC 1-6 - see first two files below. They also extend beyond the image into the surrounding black area. 

If I turn off dynamic distortion correction as you noted above then they go away (final image). These are very large mosaics (already automatically downsized by APP) but I don't think that memory is my issue. This is using beta 22 with M1 Mac.

Thanks for reporting, yes this may be either an issue with dynamic distortion correction or your data doesn't require this correction. I'll ask Mabula about it.

Hi @vincent-mod, I tried your no-dynamic-distortion-correction trick on the other images I posted (NGC 7000 P1 & P2 on the upload site) but it didn't work.  There is plenty of overlap, yet I get what looks like the same strange integration line despite having dynamic distortion correction off for the mosaic step.  This seems to be coming from the edges of the integrations not being properly trimmed and blended.  It does seem like it registered correctly.  Any other insights?

APP:

Manual:

I'm also wondering if you're running into a limitation we still have for large mosaics, if you downscale the mosaic, do they still appear? You can scale down in the integrate tab (6) all the way down. Set that to 0.8 for instance or 0.5 just to test.

Hi @mountainair and @jdwood,

Thank you very much for reporting these strange problems. I will investigate them a.s.a.p. now! Sorry it took a while due to me having a 2 week holiday, but I am back now 😉

We have another report that LNC creates weird lines:

https://www.astropixelprocessor.com/community/appreleases/image-artefacts-on-multipanel-mosaic-2-0-0-beta17/

And here is also reported that weird holes can appear which is in fact a know bug that I will fix.

I will work on the uploaded data from Walter and mountainair to get this issue solved.

Jon, can you upload your mosaic data as well so I can use it to test and make sure that we solve this ?

You can upload it here:

> https://upload.astropixelprocessor.com/ <

username: uploadData

password: uploadTestData

Please make a folder with your name and description like Jon-mosaic-lines

and let me know once uploaded, thanks!

Hi @mabula-admin

I think that you should already have some of my mosaic files uploaded re. the recent thread:

Image viewer crashing - beta 20 

but happy to reload some if you need them.

@mabula-admin No worries, I hope you enjoyed your vacation.  Thanks for identifying that this may be a problem, and for diving in to fix it.  Note that this happens in my smaller mosaics as well (the two examples uploaded were 2x1) so it is not only a problem with huge mosaics.

Let me know if you need the raw subs, but for NGC 7000 we're talking 35GB.

Hi @mountainair & @jdwood,

I have been investigating this issue using the 45 panel mosaic of Walter mentioned here:

https://www.astropixelprocessor.com/community/appreleases/image-artefacts-on-multipanel-mosaic-2-0-0-beta17/#post-28985

He indicates that this was happening in beta17.

I have run your 45 panel mosaic now more then 20x times on different computers, on Windows 10, and on macOS intel and macOS with Apple Silicon. Everytime I ran it with different LNC settings and each time it worked without producing the weird horizontal artefacts that you get. So I am not closer to understanding the issue nor solving it unfortunately.

I suspect that it is in fact a data IO issue somehow, so I need to ask you if you use the Harddisk intensively for another application while APP is performing the actual integration or not?

So i need to ask you as well:

I suspect that it is in fact a data IO issue somehow, so I need to ask you if you use the Harddisk intensively for another application while APP is performing the actual integration or not?

One thing that would be nice to know and might indicate where the bug is is the following: how many CPU threads have you set for APP to use when processing the data? It is the top left CPU thread number. And what happens if you set it to a different amoung of CPU threads? Can you check if the problem is the same or different?

Mabula

@mabula-admin You might be right Mabula,

I have just recently had some success (with LNC) by setting scaling to 0.2 or 0.5 rather than letting APP scale the image by itself - but I was not expecting any memory issues. I have M1 Mac with 64GB, currently set to use 9 threads with APP. No other applications running while doing any processing.

Jon

@mabula-admin Hi again Mabula,

I ran some more tests today to confirm the above observation.

So, just for background, the full mosaic I am running is 36 panels each of 300MB. M1 Mac with 64GB memory, 9 threads running. I am using CPU not GPU because the latter causes an image viewer/Java crash when the final mosaic is produced. Registration is start 5, stop 10, mosaic mode, use DDC, same camera unchecked. Normalise has 'neutralise background' unchecked. Integrate is LNC 2nd degree, 3 iterations, MBB set to 10%.

If I let APP do everything it will tell me that the image has to be downscaled by 0.79650. If I run with this then I get the horizontal lines as per the first image below.

If instead I choose to downscale myself by 0.5 in the Integration window, there mosaic looks OK, as shown below.

These are the analytical results for reference

So, either there is something going on when APP downsizes by itself because it is at capacity...or perhaps my Mac is at capacity and causing a problem...but no other software running while I am processing...and it doesn't take too long?

cheers, Jon

@mabula-admin Generally, I'm not running any other apps or using the disk on this machine.

Hardware:

• 2023 16" MacBook Pro M2 Max
• 96GB RAM
• 2TB SSD

Because it is generally dedicated to APP, I generally do change it to use all 12 CPUs.

One other thing to note:  I just stacked a single photo from 61mb subs (4784 x 3194), and ended up seeing one of the lines.  This was not a mosaic.  This was a fairly small stack, 23 OIII subs.  The line didn't appear in any of the other channels.  For what it's worth, I did just try re-running the integration with 11 and then 8 CPUs but the behavior was the same every time:

What DOES make a difference was your earlier recommendation to disable LNC.  In fact, it eliminates a big gradient problem I was trying to figure out in my OIII channel in addition to fixing the strange line:

Before, with LNC 1 3 iterations:

After, with no LNC -- no line and no strange gradient:

Note, I couldn't find a source for the gradient in any of the individual OIII subs.  This seems to have come out of nowhere.

I can honestly say I have never seen this issue and I do a lot with Mosaics and LNC (1+3), I never use LNC with non-mosaic images, for my mosaics I always Change the following settings:

3 Analyse Stars
Change starts to 2500

4 Register
Pattern recognition: Pentagons
Scale Stop: 15
Use Dynamic distortion correction: Enabled
Same camera and optics: Disabled
Registration Mode: Mosaic

6. Integrate
LNC Degree: 1
LNC Iterations: 3
Enable MBB: Enabled
MBB Value: 12% (My frames overlap by 10%)

I also follow the rule of stacking each panel manually, and then performing the mosaic from the stacked panels, rather than loading in all of the lights, calibration frames etc and expecting the Mosaic to come out from that as that process did not work for me in the past.  I also find it faster to do it this way.

But I have never had any issues with what I am seeing in this with the lines

Simon

Hi all @mountainair @stastro @jdwood & @walsc

Yes, this issue is still open and despite extensive testing it never seems to occur on my test computers using your uploaded data nor my mosaic data.

THis is another example by Walter:

https://www.astropixelprocessor.com/community/appreleases/image-artefacts-on-multipanel-mosaic-2-0-0-beta17/

I have run this one now over 30x and never the problems shows up no matter which settings I use.

There is one thing in common, all users that have reported this issue are using macs, it does not seem to occur on windows nor linux, or does it?

I did test myself on 2 different macs, 1 imac pro Intel and 1 mac mini M1 but still i can not reproduce this issue somehow.

Are you all having the working directory on an internal harddisk, or are you using an external location?

The issue has the appearance of harddisk IO problems somehow. If there were bugs in our read/write code, surely it would happen on all platforms and more regularly I would think. But I will still investigate thoroughly in the read/write code of the integration and especially LNC code which does read eacht integration layer, corrects it and writes it back again... Another cause could actually be IO problems on the computer to the location of the working directory, that is why I ask where the working directory is located when this issue manifests.

Mabula

Thank you for not giving up.
My image files are located in a (sub-)folder on the desktop, so it is the main HDD.

Once these stripes occured (at around 40 images), I started making downscaled images (0.5x) and that worked fine until I had more than 70 images, then these stripes occured again.

Now I have more than 90 images, it nearly takes 24 hours to render a mosaic plus 2nd degree LNC (1 iter) with the downscaled images, I did not even try the fullscale images.

Posted by: @jdwood

↑

@mabula-admin Hi again Mabula,

I ran some more tests today to confirm the above observation.

So, just for background, the full mosaic I am running is 36 panels each of 300MB. M1 Mac with 64GB memory, 9 threads running. I am using CPU not GPU because the latter causes an image viewer/Java crash when the final mosaic is produced. Registration is start 5, stop 10, mosaic mode, use DDC, same camera unchecked. Normalise has 'neutralise background' unchecked. Integrate is LNC 2nd degree, 3 iterations, MBB set to 10%.

If I let APP do everything it will tell me that the image has to be downscaled by 0.79650. If I run with this then I get the horizontal lines as per the first image below.

-- attachment is not available --

 

If instead I choose to downscale myself by 0.5 in the Integration window, there mosaic looks OK, as shown below.

 

-- attachment is not available --

 

These are the analytical results for reference

-- attachment is not available --

 

So, either there is something going on when APP downsizes by itself because it is at capacity...or perhaps my Mac is at capacity and causing a problem...but no other software running while I am processing...and it doesn't take too long?

 

cheers, Jon

I have the same problem, and I figured out if you get your resulting image with these horizontal lines, just hit the "integrate" button again, the next result should be good, if not then you might do this again, then the stripes are gone.

Posted by: @walsc

↑

Thank you for not giving up.
My image files are located in a (sub-)folder on the desktop, so it is the main HDD.

Once these stripes occured (at around 40 images), I started making downscaled images (0.5x) and that worked fine until I had more than 70 images, then these stripes occured again.

Now I have more than 90 images, it nearly takes 24 hours to render a mosaic plus 2nd degree LNC (1 iter) with the downscaled images, I did not even try the fullscale images.

Walter, when you say you have 40, 70... images, are these individual light frames, or are these mosaic panels?
If they are individual light frames, are you performing the calibration and stacking suring the same process as performing the Mosaic, or are you creating the mosaic from already created stacked files?
When you also say HDD, is this a spinning disk or an SSD?

With "images" I mean mosaic panels - stacked, cropped and light pollution removed, no further processing.
It is a Mac Studio, there are no spinning discs inside. I am 51 years old, I am used to this term, sorry.