hugin produces bad picture

[Alexander Rabtchevich]

Hello

Here is a scaled down picture, produced with hugin 2017.1.0.64cc39ce23d4 . The result doesn't depend on the type of blending software - internal one or enblend-mp and enfuse-mp are used. The scenario was - loading pictures, CPFound + Celeste, geometrical optimization (positions), fotometric optimization (LDR), adding several vertical lines and automatic straightening.

I recall it is the second time I see the similar output. The previous was more than a year ago.

If needed, I can provide with original tiffs.

With respect,
Alexander Rabtchevich

[T. Modes]

Strange,

I can't see anything strange in the pto file. And with dummy images I can't reproduce the issue.

Can you test if the remapped images are okay and the issue is with the blender or if the remapped images are already buggy (on the stitcher tab select remapped images)

One more idea: Have your images icc profiles embedded? If so check the icc profile for the red image.

[Alexander Rabtchevich]

Here is a the link to original files. It is possible to reproduce exporting.

https://yadi.sk/d/um_AyDWy3NqRLZhttps://yadi.sk/d/um_AyDWy3NqRLZ https://yadi.sk/d/um_AyDWy3NqRLZ
https://yadi.sk/d/um_AyDWy3NqRLZ
вторник, 17 октября 2017 г., 19:06:12 UTC+3 пользователь T. Modes написал:

[T. Modes]

Am Dienstag, 17. Oktober 2017 18:32:59 UTC+2 schrieb Alexander Rabtchevich:

Here is a the link to original files. It is possible to reproduce exporting.

Sorry, but this is to big. I don't want to download 1 GB.

From the originals 2 images would be sufficient (e.g. DSC03288 and DSC03289)

When you remap the images you can also decrease the size, which shows also the issue.

[Alexander Rabtchevich]

Could you please specify the required steps? Should I upload two first images or scale them and upload scaled ones?

вторник, 17 октября 2017 г., 19:44:54 UTC+3 пользователь T. Modes написал:

[T. Modes]

Am Dienstag, 17. Oktober 2017 21:01:52 UTC+2 schrieb Alexander Rabtchevich:

Could you please specify the required steps? Should I upload two first images or scale them and upload scaled ones?

If you can reproduce the issue with the downscaled images it would be fine if you could upload only the downscaled ones (The pto file for the downscaled version would also be fine).

[dkloi]

Here's a downscaled version of the 24202 pixel wide stitch. Hugin 2017.0.0.eac5e8cc546e on Windows 10 Ent 64 bit. Don't see any problems. CPFind, added some VCPs, position optimization, remove outliers, position optimized again, photometric optimisation, stitch at maximum optimum size.

[Alexander Rabtchevich]

I'm afraid some changes in the image size can lead to disappearing of the red spot. At least I recall the same workaround worked previously. But it is not the solution.

[Gunter Königsmann]

Individual bright white and red pixels that disappear when scaling the
images down? That suspiciously sounds like the "fire ants" thread in
this group. That in turn sounded like nona uses floating-point
calculations for determining where which images start and stops
neglecting that using floating-point introduces small errors that might
add up. But I might be wrong in this case.

One example for floating-point errors I always use in maxima is that an
exact 0.1 cannot be expressed in floating-point. My computer uses
3602879701896397/36028797018963968 instead - which nearly is 0.1, but
not quite. Normally this error is neglectible. But it might cause
something to start half a pixel too far to the right/left...

[T. Modes]

Am Mittwoch, 18. Oktober 2017 05:51:51 UTC+2 schrieb Alexander Rabtchevich:

I'm afraid some changes in the image size can lead to disappearing of the red spot. At least I recall the same workaround worked previously. But it is not the solution.

Sorry, but so we don't get on.

Because nobody else can reproduce the issue we need your help:

1.) Open the pto file from the first post.

2.) Go to stitcher tab, additional tick option "Exposure corrected, low dynamic range" under "remapped images".

3.) Stitch project now.

4.) Now check output files: especially the file PREFIX0003.tif - has this already the red color or only the blended images PREFIX.tif

5.) Reduce output size (please state number used).

6.) Repeat steps 3 and 4.

Please report the results here.

One more issue: I don't think that you used a cylindrical lens for shooting. The lens projection is probably rectilinear. With this lens projection the cp error in the project is decreased significant. But this should be unrelated to the red spot.

@Gunter Königsmann: Please read the full thread before posting totally unrelated statements. Look at the image in the first post and you will see that this issue has nothing to do with your "fire ants".

[Alexander Rabtchevich]

Looking at the enclosed image at the beginning of the thread, I guess it is more than the rounding error. The large part of the stitched image is a red spot which is absent in the original images.

среда, 18 октября 2017 г., 8:21:21 UTC+3 пользователь Gunter Königsmann написал:

[David W. Jones]

FWIW, I downloaded the original TIFF images. Did CPFind, Find Vertical
Lines (it found none). Optimize Positions (incremental, starting from
anchor), Clean control points, Optimize Everything without translation.
Optimize for Low Dynamic Range. Set it to rectilinear projection, full
size. The stitch came out just fine.

That's a very peaceful looking neighborhood.

I did learn a bit: I feel reassured that my present laptop (2.4GHz
4-core/8-thread i7 with 16GB RAM) could handle it, since I'm hoping
sometime to replace my present 6MP DSLR with a 24 or 36MP one. I've been
eying the new AMD and Intel processors, wondering if I'd need to jump to
one of those. I have a couple of strip panoramas what cover many miles
of coastlines (some in Hawaii, some in Canada), so nice to know I'll
still be able to process such panoramas with higher resolution images. :)

On 10/17/2017 10:44 AM, 'dkloi' via hugin and other free panoramic
software wrote:
> <https://lh3.googleusercontent.com/-gaEFRn2ccmY/WeZqip7wWlI/AAAAAAAEVZg/J4XdStdsM5kImh6imX4Co07UF8SbjUnxgCLcBGAs/s1600/test.jpg>

>
> Here's a downscaled version of the 24202 pixel wide stitch. Hugin
> 2017.0.0.eac5e8cc546e on Windows 10 Ent 64 bit. Don't see any problems.
> CPFind, added some VCPs, position optimization, remove outliers,
> position optimized again, photometric optimisation, stitch at maximum
> optimum size.
>
> On Tuesday, 17 October 2017 17:32:59 UTC+1, Alexander Rabtchevich wrote:
>
> Here is a the link to original files. It is possible to reproduce
> exporting.
>

--
David W. Jones
gnome...@gmail.com
wandering the landscape of god
http://dancingtreefrog.com