Milkyway Panorama blending.
I shot an 85mm panorama of the Milkyway last month in two separate attempts - I'd love to use APP to stitch together the frames and wondering what the best approach would be.
So far I have calibrated and stacked all my frames and attempted many integrations with little success.
All shot in Bortle 2 conditions.
The first pano I shot on my Sony a7iii + 85mm f/1.8 lens at 3 minutes per sub, f/2.8, and iso 640.
The second pano I shot on my Sony a7iv + same lens at 3x60s, f/2.8, and iso 640 as well.
My computer has an i9-12900h, 40 gigs of DDR5 ram, and 1TB SSD at about 3,000mbps r/w.
First off - a question: What is the best way to integrate some of the data from the first pano into the second one shot on my newer camera? Since I got some 3-minute subs on that first attempt I think I could (hopefully) improve my SNR on key areas like Rho Ophiuchi. Should I first stitch the pano from the new camera, then try to integrate the mosaic with a single frame at a time? Since I shot these without a go-to mount, the panels are all done by hand and not precisely in the same parts of the sky between the two sessions. Or is this a lost cause and not worthy of my time/effort?
On that same note, I have about an hour of integration time on Rho with the same lens from a Bortle 4 zone from a few months ago and wondering if you think it's worth using since it's so much extra data but in much more polluted skies... I know that's a tough one to answer but worth throwing out there.
Secondly, in terms of blending, I'm having issues. At first, I tried to integrate all 60ish of the frames from both sessions... first of all, that required like 700gb of free space, and even when I did that both attempts to integrate just straight up failed after like 25 hours or more of time... this is when I started to get the idea to just use the second session and try and add bits of integration time to it.
At first, my issue was that the projection was bending the top of the Milkyway, but I think I have resolved that by choosing a more central reference frame. Also at the same time, I was getting blurry stars in key areas like the lagoon nebula and so on.. that seems to have been resolved by going through all the lights and cropping out any of the edges of frames where the stars started to stretch, particularly where key objects were on the edges. Now my main issue is just what you see here:
IIRC, this is with 10k stars analyzed, scale start/stop 1/10, Pentagons, Dynamic Distortion Correction on, Same camera OFF (even though for this particular pano I only used data from my second session which actually was same camera, but any time I tried to run this with that setting on it would give me a failure message which I can paste here if needed), registration mode mosaic, and LNC 2nd degree/5 iterations (this is the part I'm fuzzy on, could have been slightly different). MBB 10%.
I attempted to increase LNC to 4th degree and 10 iterations and was greeted with this beauty:
For reference, an image stitched from single uncalibrated frames (second session) in Microsoft Image Composite Editor:
Certainly not perfect but blended quite nicely... Any help here is greatly appreciated. I have spent countless hours wrangling this beast and haven't even gotten to attempt post-processing it yet.
So, are you first creating the panels of the mosaic before trying to actually make a mosaic? It's best to do that first, giving you about 10 or so panels if I'm seeing that right? That should combine pretty fast and in a better way. You can then also first create the panels with the different camera data.
@vincent-mod I tried to do that but got some errors while attempting to put together a few frames at a time. When you say creating panels, you mean just putting together something like 3-4 frames at a time instead of stitching them all at once? You think that'll solve my blending issues? I'd definitely be willing to try that again.
Yes sort of like that. Usually a mosaic consists of panels (the area's you used to divide the object into), these panels you create first. I can have a look at your data as well if you want, though my connection here is spotty at the moment. I can ask Wouter or Mabula to have a try if that's needed.
@vincent-mod I will give that a try again... was having issues stitching small panels but now that I'm in this forum I can at least troubleshoot when it goes wrong again. This will actually be helpful in integrating the extra data too as I just shot several more minutes in key areas last night.
Here's a folder with all the lights and calibration masters if you would like to give it a try!
@vincent-mod Okay so I attempted to create a smaller panel rather than stitching everything at once - I am not sure how small to make panels (this panel ends up being about 1/3 of the size of the total mosaic) and it's very good in terms of blending but unfortunately, I still have two major issues.
First, you can see in this example that the stars are not fully aligned between panels and get doubled up in certain areas. I'm not sure which setting to use to rectify this.
From a distance, you can see the blending between panels in terms of LNC is really great.
Up close, though, you can see the issue of misalignment. It is only in certain areas though, and when I change the reference frame the misalignment happens in different areas.
The second concern I have is the projection method - what's the best way to choose a reference frame here? Centered to the panel, or as close to center to the final mosaic as I can? I would love for the Milkyway to come out straight and not have any stretching on Rho Ophiuchi as well, just like my ICE projection from my OP shows.
I have additional data that is now calibrated if you'd like to give any of this a try yourself, I can upload it.
@Mabula-Admin maybe you know what's going wrong here? I have spent many hours trying to stitch this mosaic and it's a commissioned project that will be printed in a very large format. I have tried 10,000 stars and it still does not align properly.
Sorry for the delay Michael, bit busy down here. I've asked Mabula to give it a look as well.
Hi Michael @mikeabr,
The first thing you need to do is create the mosaic panels itself using regular registration. Don't try to mosaic from mosaics.
Look at this tutorial first:
The 11 mosaic panels are first created using normal integration with normal registration (not mosaic mode).
Then the mosaic is simply created from the mosaic panels with mosaic registration.
Do not set the star count higher than 3000-4000 stars, that will only make things unstable I think.
If it does not register then, you need to either increase scale stop, use triangles or increase the distortion margin I think.
@mabula-admin Thanks for taking the time to respond! I followed the tutorial closely but maybe I'm just getting confused on what really defines a panel.
I have data from four separate nights and don't have distinct "panels" per say, just many light frames. One night I opted to do shorter exposures and stack, so I did stack those using regular registration. I now have something like 60ish light frames (fits files) which have been calibrated in APP. I'll try again to integrate them all at once, but I had some issues with that in the past (described in first post) so I was trying to do smaller amounts at a time to see if I could dial in the settings.
I'll give it a shot again with your suggestions and keep you posted.
@Mabula-admin I attempted to stitch again just 16 of the 60ish frames using those settings and here's what I got:
Here are all of the settings I used:
Again, the colors/brightness seem to blend quite well, but not the actual star alignment.
Could you try disabling distortion correction?
@vincent-mod I just ran an instance with the exact settings of my last post but with DDC off (and that switched to projective instead of calibrated projective) and here's what I got:
Much better than before but still not quite aligned in ALL places.
Rho looks great now
but other parts are worse now
Which settings can I tinker with to get this right? Just want to remind that this is only 16 of the panels out of the 60ish that I will need to use for the final integration, which shows much more of the milkyway above this region.
This whole project is a commissioned print which will be printed in a very large format, so I need all the resolution I can get here.
Mm, tricky, I'll ask Mabula again to have a look, we might need the data ourselves to find out (which you shared, so that's not too hard 😉 ).
The workflow is still not as advised in the mosaic video tutorial.
To make the mosaic work you need:
1) create mosaic panels using normal registration (no distortion correction) of the subs per mosaic panel. A panel is defined by a single field of view. So in your lights, the first 3 lights would be one panel, they show the same field of view.
2) Once all mosaic panels are created. Then you can load those panels as lights and register them as a mosaic using dynamic distortion correction enabled.
By not following this workflow, the current APP implementation will be unstable as you are experiencing.
I will try to process the mosaic myself in the coming days and will let you know if I can create it correctly or not.
@mabula-admin Hey Mabula,
Thanks for your reply. I think I may have been unclear - My first attempts at stitching the mosaic were exactly as you described. Of course, it was so long ago at this point (and so many attempts ago) that I am not 100% sure where I may have gone wrong, but I distinctly recall following your tutorial closely before I ever posted here.
I am very much looking forward to hearing how it goes for you, thanks for taking the time. If you feel the need to make another tutorial you are welcome to use my data.
Just to clarify, the folder I provided here previously isn't all the data I would like to integrate in the final project since I got more integration time on certain areas of the mosaic on different nights. The folder includes just one session where I managed to cover most of the mosaic, though. And in my workflow, I did as you suggested: integrated each stack into a panel of its own (3x60, some panels got a few extra exposures for whatever reason). In my workflow, I also cropped out the bad edges of certain frames in order to avoid having coma in key areas that happened to fall on the edges of certain frames... but in the set that I provided that is not nearly as prevalent an issue.
Hi Michael @mikeabr,
I have run a first test on your data.
First I would like to point out that your MasterFlat does not work very well, the calibrated data shows overcorrection of the vignetting profile unfortunately, creating dark circles in the middle of the calibrated images. I know that shooting good flats is difficult with short focal lengths though. If the masterflat would be good then the blending of all panels would be easier of course.
But let us focus know first on getting the data properly registered.
As a first test I looked at your 101 uploaded frames and was able to divide it into 33 panels where each panel has the same field of view.
I did not create the 33 panels for this test. I only loaded 1 single frame for each panel to see if I can make that register correctly.
Without a lot of effort I was able to get it to align nicely I think. I was using APP 2.0.0-beta4.
3) Analyse stars set to 2500 (so not too high, this is not needed since your panels have plenty of overlap 😉 )
4) Register, quadrilaterals, scale start 1, scale stop 10 (defaults for mosaic) dynamic distortion correction on, same camera and optics off (but on should work as well and more stable so will test that as well). Mosaic mode. Calibrated Projective model since we need to be able to project differently than rectilinear. Rectilinear only works for Field Of Views up to 120 degrees but then will already stretch the data strongly... We want to project this with Equirectangular or Mercator. So at the bottom of 4 set the projection to equirectangular or mercator.
5) for first test, just defaults
6) also defaults, except I set scale to 0,5x to run it a bit quicker for a first test.
This is what i get, already looks very good to me. You can see why I mention that your MasterFlat is not optimal ;-). Dark circles in all the panels are visible.
The reference was chosen to be the bottom right panel. I will keep it as reference and will try to adjust the projection center to get a vertical milkyway. as you mentioned next.
I checked for misalignment but I only see it very little in the extreme corners of the panels where the optical distortion is biggest. I would think that a high % of Multi-Bend Blending will solve that easily.
Side by side the 3 possible projections: rectilinear, equirectangular & mercator:
Mercator projection looks almost the same as equirectangular. Rectilinear is only for FOVs less than 120x120 degrees. The other 2 can project the whole circle of 360x180 degrees.
Mercator looks nicer than equirectangular, but that is personal of course. I will continue with mercator projection and will try to remove the seams/illumination differences now.
firs step, 5) advanced normalization with Multi-Band Blending at 50%: