I'm occasionally seeing bad results from Verdandi in which parts of the 
panorama are discolored:

<https://private-user-images.githubusercontent.com/119951/348439722-4fbb1066-640f-4d74-9bdb-41ea82b430a0.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3MjA4MzQ0MTcsIm5iZiI6MTcyMDgzNDExNywicGF0aCI6Ii8xMTk5NTEvMzQ4NDM5NzIyLTRmYmIxMDY2LTY0MGYtNGQ3NC05YmRiLTQxZWE4MmI0MzBhMC5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjQwNzEzJTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI0MDcxM1QwMTI4MzdaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT03ZjY3MDBiNjFhYzU5YzExYWZjMGFkYmRkMzFkMjdjZjY4MGJlZDkwN2Y3OWQ2MmU1NDFkYzBlNmIyMmIxOTg3JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZhY3Rvcl9pZD0wJmtleV9pZD0wJnJlcG9faWQ9MCJ9.i1RT8KbT_rKjZPcB4xPjnj4Sf74s_DTukdH95l8Vv6Q>

This looks similar to an issue previously reported and fixed in 2019: 
https://groups.google.com/g/hugin-ptx/c/mpnWNfJZ4b8/m/LMFVNMdbBQAJ

In testing, I found that 2023.0.0 and mercurial tip are affected, while 
2022.0.0 results are good.

Weirdly, it seems that the nona output is nondeterministic for the same 
images and PTO file. The nona output itself looks fine, and the artifacts 
are introduced by verdandi. The severity and coloring of these artifacts 
varies between different sets of remapped images from running nona multiple 
times, but does not seem to change between different runs of verdandi on 
the same remapped TIFs.

There are only a few commits touching verdandi between 2022.0.0 and 
2023.0.0, but 8556:17d25b5ae116 
<https://sourceforge.net/p/hugin/hugin/ci/17d25b5ae116/> stands out. It 
re-adds a call to Multigrid that had been commented out with the note "this 
results in artefacts, maybe there is still a small error in other lines" in a 
2019 change <https://sourceforge.net/p/hugin/hugin/ci/62b0662d5bee> that I 
think was the bugfix from the thread linked above.  I tested commenting out 
that line again and the results look better.

I don't understand the algorithm enough to know how this helps, what other 
problem it might be working around, or what the proper fix would be, but 
I'm hoping someone can take another look at it.

Here are two sets of remapped images that exhibit the problem: 
https://test-bucket.stscn.net/pano/verdandi-color-bug/1_remapped.zip 
https://test-bucket.stscn.net/pano/verdandi-color-bug/2_remapped.zip
Run verdandi --wrap --seam=blend --compression=LZW -o out.tif -- part*.tif. 

Let me know if I can help with more details, testing, or anything else.


Thanks,

Kevin

-- 
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 on the web visit 
https://groups.google.com/d/msgid/hugin-ptx/30b64bb3-bf96-4c8e-b484-5ccf75d2a934n%40googlegroups.com.
  • [hugin-ptx] Colored... Kevin Mehall
    • [hugin-ptx] Re... 'T. Modes' via hugin and other free panoramic software
      • [hugin-ptx... 'T. Modes' via hugin and other free panoramic software
        • [hugin... 'T. Modes' via hugin and other free panoramic software
          • [h... Kevin Mehall
            • ... 'T. Modes' via hugin and other free panoramic software
              • ... Cristian Marchi
                • ... 'T. Modes' via hugin and other free panoramic software
                • ... Cristian Marchi
                • ... Kevin Mehall
                • ... 'T. Modes' via hugin and other free panoramic software

Reply via email to