Exposure fused from any arrangement problem

[Hans Bull]

I think a similar subject has already been discussed, but I cannot find the thread.

Trying to stitch nine images fused from any arrangement I get unsharp areas at the right of the image, but the unsharpness is not present in any of the various "exposure" and "exposure_layers" intermediate tiff files (see attached file - left is the blended-fused final output). Switching the blender to multiblend instead of enblend produced the same problem, so the problem does not seem to be enblend-related and even seam-related. To me it seems like a mismatch between the intermediate files. Optmized parameters were yprxyzv with lens distortion from lensfun.

Any idea?

Best,

HB

[Hans Bull]

Finally found the old thread, referring to the 2013 version:

https://groups.google.com/forum/?fromgroups#!topic/hugin-ptx/QZIRpWBotas

--no-ciecam does not help, though. The problem persists in the ubuntu ppa:hugin/next (Raring, 64bit) I am using.

[Terry Duell]

Hello Hans,

On Sat, 30 Aug 2014 05:49:10 +1000, Hans Bull <bul...@gmail.com> wrote:


> Trying to stitch nine images fused from any arrangement I get unsharp
> areas at the right of the image, but the unsharpness is not present in
> any of the various "exposure" and "exposure_layers" intermediate tiff
> files (see
> attached file - left is the blended-fused final output). Switching the
> blender to multiblend instead of enblend produced the same problem, so
> the problem does not seem to be enblend-related and even seam-related.
> To me it seems like a mismatch between the intermediate files.Optmized
> parameters were yprxyzv with lens distortion from lensfun.
> Any idea?

Why optimize x,y,z?
What result do you get if you just optimize y,p,r,v or everything without
translation?

Cheers,
--
Regards,
Terry Duell

[Brandon]

I do not see a problem on the right, but I could be over looking something the big thing that I see is center left all of the white around the windows and the street light. They look like the camera shuck on a long exposure image(i know that is not what happened here)

Does it do it on a low range stitch? I am wondering if your control points between the images have some issues and are not quite lined up.

[Hans Bull]

Terry,

optimizing only y,p,r,v unfortunately yields only marginally different results. --primary-seam-generator=graph-cut or --no-optimize does not help either. Photos are hand-held. In any case, a "exposure corrected, low dynamic range" stitch gives a perfectly aligned image (like the one on the right). That's what I do no understand.

Best,

HB

[Terry Duell]

Hello Hans,

OK.
Are you able to make the image files and the .pto file available for
others to test, perhaps on Dropbox or similar?
A nine image set may be quite a sizeable package...is there a subset of
the images that still shows the same problem?

[Hans Bull]

Thank you so much, Terry! I uploaded everything here (using JPG instead of the original 16bit TIF):

https://www.dropbox.com/s/mr2csuxuj5uyji9/_DSC7681-_DSC7689___x.tar.xz?dl=0

The archive includes the stitched files for immediate comparison. Colours are not so important, since I use to convert the result to b/w, using darktable. 

Best,

HB

[Terry Duell]

Hello Hans,

OK, I have the files.
Let's also have your .pto file so I can see your control points and image
EVs.

[Hans Bull]

Hello Terry,

I thought I included the pto :( but I didn't. Sorry. 

Here it is: https://www.dropbox.com/s/dkw4jgioevv8o3c/_DSC7681-_DSC7689___x.pto?dl=0

And here the stiched image w/o problems: https://www.dropbox.com/s/1ha2vhsdjwjltmp/_DSC7681-_DSC7689___x.jpg?dl=0

The fused one should be included in the archive.

Best,

HB

[Terry Duell]

Hello Hans,

On Sun, 31 Aug 2014 16:21:05 +1000, Hans Bull <bul...@gmail.com> wrote:

> Hello Terry,
> I thought I included the pto :( but I didn't. Sorry.
> Here it
> is:
> https://www.dropbox.com/s/dkw4jgioevv8o3c/_DSC7681-_DSC7689___x.pto?dl=0
> And here the stiched image w/o
> problems:
> https://www.dropbox.com/s/1ha2vhsdjwjltmp/_DSC7681-_DSC7689___x.jpg?dl=0
> The fused one should be included in the archive.

OK, I have run your .pto and produced a stitch using "exposure fused from
any arrangement", and the result does seem to be sharp, but there are a
couple of points about this.
My stitch can be found here ...

https://dl.dropboxusercontent.com/u/36440732/_DSC7681-_DSC7689___xfaa_blended_fused.jpg

Firstly, you didn't provide image _DSC7681-0.JPG, so I used a dummy image
to open your .pto in hugin and did the stitch with that image deselected,
so it is possible that image may be influencing your result.
Secondly, it looks as though you have shot these images as a normal
pano...i.e. from the one location, and your control points have the
appearance of being auto generated. In these circumstances one wouldn't
normally optimise the translation parameters (x,y,z).
If you want to optimise x,y,z, control points should be confined to a
plane surface (e.g. all on a wall), which does appear to be possible for
these images, but your set does include control points that are clearly
not in the same plane as the majority of your points.
Finally, I have used hugin-2014.1.0.

Have another look at this project with image 7681 deselected and see if
you still have the same sharpness problem.

[bul...@gmail.com]

Thank you so much, Terry! And thank you for the tips! In fact I do not see the "sharpness" problem in your stitch! 

Just uploaded _DSC7681-0.JPG here (though I included it as well in the archive): https://www.dropbox.com/s/1ha2vhsdjwjltmp/_DSC7681-_DSC7689___x.jpg?dl=0

I'm using Hugin & nona 2014.0.0.51ff237f209e, enblend-mp 4.1.2, all on Ubuntu 14.04 64bit.

Did you use graphcut as the primary seam generator? Did you use gpu remapping for nona?

Best,

HB

[Terry Duell]

Hello Hans,

On Sun, 31 Aug 2014 17:06:38 +1000, bul...@gmail.com <bul...@gmail.com>
wrote:

> Thank you so much, Terry! And thank you for the tips! In fact I do not
> see
> the "sharpness" problem in your stitch!
> Just uploaded _DSC7681-0.JPG here (though I included it as well in the
> archive):

Image 7681 definitely not in the archive that arrived here!

> https://www.dropbox.com/s/1ha2vhsdjwjltmp/_DSC7681-_DSC7689___x.jpg?dl=0

The above link is to one of your stitches, not the 7681 image. It probably
is worth having another go at providing that image just to be sure that
the result is the same here when using it.

> I'm using Hugin & nona 2014.0.0.51ff237f209e, enblend-mp 4.1.2, all on
> Ubuntu 14.04 64bit.
> Did you use graphcut as the primary seam generator? Did you use gpu
> remapping for nona?

I use the default primary seam generator (nearest-feature-transform), and
do not use gpu remapping.
I used Enblend 4.1.3, although given that Multiblend didn't make any
difference for you, I suspect the blending isn't the culprit.

If I still get the same (sharp) result using all your images, we will need
to have another think about what is going on...maybe others have seen this
previously.

[T. Modes]

Am Sonntag, 31. August 2014 09:07:05 UTC+2 schrieb Hans Bull:

Did you use graphcut as the primary seam generator? Did you use gpu remapping for nona?

 
These options has nothing to do with your problem.
You have an alignment problem. When using "exposure fused" you need an exact alignment of all images in the stack.
Otherwise you get the unsharp areas, because the exposure fusion is using the information from all images in the stack and so will see also slight misalignments.
So check the cp in the images with the problematic area.

Thomas