Thank you for your time sharing knowledge and giving a run with my photos.
I definitely see what you mean with respect to the edge falloff. The left 
side is worse than the right it looks like, and some are worse than others, 
but overall the centers are sharper.
I imagine the alignment could've been off resulting in keystoning and that 
would explain why it doesn't line up without adjustment. I had to do a lot 
of refinement of points and allow transformations to get it nearly perfect. 
I think in my project my max misalignment was around 1.2px and the average 
was like 0.3px after the transformations and visually the largest control 
point disparities were as good as I could hope to do it manually.

I am using a reversed 24mm lens on quite a lot of tubes and I thought I 
would be able to avoid soft edges since it's already sort of cropped into 
the middle of the lens, but I'm also thinking it might be because of the 
lens' spherical focal distance against an almost perfectly flat subject. I 
will print a calibration pattern and use it to verify coplanarity as well 
as check for sharpness falloff. I'm used to stitching panoramas in other 
programs like Photoshop which are quite a lot needier in terms of overlap 
so it's good to hear I can get away with less. I might shoot the next one 
in APS-C mode before I decide whether I need to try to source a better 
lens. Also intending to build another level onto my mechanicals to ensure 
that the camera is fixed compared to the subject.

This subject is a Cistus (hybrid probably) flower. I specialize in 
ultraviolet-induced visible fluorescence floral photography but I started 
exploring UVIVF floral microscopy during a project and wanted to continue 
doing it. The first panorama I did was a Phacelia and it was small enough 
Capture One handled it. It seems like C1 is pulling sharper parts of 
images, but I've also been pickier on this one and saw more flaws after I 
spent so much time on it. I have another one even larger than the Cistus 
(guessing around 1 gigapixel) that I need to develop but I needed to figure 
out how to use Hugin adequately and get the Cistus done first as a 
validation for the workflow.

The learning curve is steep, but I'm seeing how powerful it is. I am 
impressed with the focus stacking as well. Everyone always touts Zerene and 
Helicon as the only reasonable options, but Hugin does it rather well. I 
use Zerene or else I'd probably do Hugin for my stacks. As challenging as 
it is with the GUI, for everyone who did it/does it with CLI I am really 
impressed.

For now, I'll try your cropping concept since the result you shared looks 
generally satisfactory at this point and more consistent than my output.

I'll keep going and if I discover anything new/edifying or if I come to any 
serious conclusion about the source of my aberrations, I will post back if 
you're interested in seeing it through.

Thank you!

On Saturday, May 3, 2025 at 11:13:46 AM UTC-7 lukas wirz wrote:

Hi Craig, 

> I had also been looking for the GPU prioritization over CPU, but 
somewhere 
> else I read that the GPU may be slower than CPU still. 

There are two components with GPU support: hugin/nona does GLSL and 
enblend does openCl. I haven't done any benchmarking in a while, but I 
think the current consensus is that GPU nona is worth it while GPU 
enblend is not (although I use it, it's not terrible at least). 

> I will be giving that a shot, as well as entering parameters for enfuse 
> (when I did it in the fusion options there was no difference. I threw the 
> outputs into Photoshop with a "difference" blending mode and everything 
was 
> identical.) 
> 
> As far as you know, does the fusion for most detail and the stitching for 
> panorama have to be completed in two steps? 

Yes, conceptually the blending (ie, finding a seam line and glueing 
together two components, aimed at not too large overlaps) and fusing 
(ie, considering pixels from either image in the whole overlap area, 
aimed at almost entirely overlapping images) stages are treated 
separately, and enblend / enfuse are two programs that do one each. One 
could have a program that takes aligned images and does a blending / 
fusing combination but I'm not aware of one, and it wouldn't integrate 
as well with hugin. 

> I've temporarily uploaded a selection of the source files where I saw the 
> issue the most significantly here: 
> 
https://u.pcloud.link/publink/show?code=kZCRRb5ZeOE3383m2v81I7UxerzaLy9x6ez7 
> I'll keep trying it myself, but if you end up trying it out with my 
images 
> and have input based on that, I would be happy to see how your test goes. 

I looked at your images for a bit. They're very pretty, is that an 
anemone? I have two main comments about that: 

-- For perfectly aligned images it should be possible to only fuse them 
without blending, and I think that may be possible with your images. 
However I'm not really getting that to work. 
What I previously said about fusing options in Preferences/Programs 
partly wrong, that sets only the default for new projects while the 
setting for the current project is in the stitcher tab / processing. 
One can verify that options have been set by opening the pto file. But 
even then I'm not seeing any difference between default / 
entropy=contrast=1.0 / entropy=contrast=0.3. I should look into that 
some time but not today. 

-- Looking at the sharpness of your images I'm not seeing significant 
differences, maybe #6 is a bit worse, but the rest look mostly the same 
to me. However, all of the images are *much* sharper at the centre than 
off-centre, as many lenses are. The images are also overlapping much 
more than they need to. So I'd just crop the images heavily and then 
run exposure corrected LDR, really nothing fancy. That looks like this: 
http://78.46.190.157:8080/anemone_pano.tgz (~520mb) It might be more 
elegant to dump all images into a program and get this automatically, 
but I think with the current images you wouldn't beat symmetric cropping. 

Two useful/interesting things would be fusing such a pano setup, I think 
that should work but I must be doing something wrong. And having a 
blending program that comes up with a seam lines based on a cost 
function for areas. That would also solve your problem here, but 
enblend doesn't do that. 


cheers, lukas wirz 

-- 
A list of frequently asked questions is available at: 
http://wiki.panotools.org/Hugin_FAQ
--- 
You received this message because you are subscribed to the Google Groups 
"hugin and other free panoramic software" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion visit 
https://groups.google.com/d/msgid/hugin-ptx/4a8bc7a7-b3e1-4acf-8096-68802c4fce54n%40googlegroups.com.

Reply via email to