black "holes" in panorama rendered in HDR
dishio
I am discovering Hugin and the many possibilities it offers. I must
say, it looks very impressive.
I have been playing with a while now and I have observe some "bugs".
I use Hugin 0.7.0 under MacOs X.
Say I have a panorama composed of 7 images all taken with different
exposures.
Because these photos are taken with different exposure, I stitch the
images in Hugin using in the exposure tab the Photometric
Optimization: "High dynamic range, variable white balance, fixed
exposure". Then in the stitcher tab, I ask to render the panorama as:
Output, Merge to HDR/Blended HDR panorama.
So far so good. Unfortunately I observe, for some panoramas, that the
hdr output file (exr or tif) contains " black holes". See the
following link for an illustration of these holes:
http://dishio.eu/stuff/hugin_bug.png
In this case I tried to open the .exr in qtpsgui. Other software show
the same problem.
Now the interesting part is that if I render the same panorama in
"normal/blended panorama", so no HDR, the usual 8 bits LDR output, the
black holes are not present and as far as i can see the rendered
panorama is just fine.
I observed that this "bug" occurs on some of the panorama that I have
tried to create. Trying to re-render the panorama will consistently
lead to the same bug. On some other panoramas, the hdr rendering will
just work fine.
When Hugin works it works really well!
Thx in advance for any hints.
dishio
tennevin yves
https://sourceforge.net/tracker/index.php?func=detail&aid=1850361&group_id=77506&atid=550441
Yves Tennevin / esby
dishio a écrit :
David Brodsky
dishio wrote:
> Dear list,
>
> I am discovering Hugin and the many possibilities it offers. I must
> say, it looks very impressive.
> I have been playing with a while now and I have observe some "bugs".
> I use Hugin 0.7.0 under MacOs X.
> Say I have a panorama composed of 7 images all taken with different
> exposures.
> Because these photos are taken with different exposure, I stitch the
> images in Hugin using in the exposure tab the Photometric
> Optimization: "High dynamic range, variable white balance, fixed
> exposure". Then in the stitcher tab, I ask to render the panorama as:
> Output, Merge to HDR/Blended HDR panorama.
> So far so good. Unfortunately I observe, for some panoramas, that the
> hdr output file (exr or tif) contains " black holes". See the
> following link for an illustration of these holes:
> http://dishio.eu/stuff/hugin_bug.png
> In this case I tried to open the .exr in qtpsgui. Other software show
> the same problem.
this may be caused by the fact that exr produced by nona can contain
pixels whose values cannot be represented by exr's type half. These
pixels are then represented by infinity thus corrupting whole image.
You can see recent thread about this:
http://groups.google.com/group/hugin-ptx/browse_thread/thread/ae3ddd4825136598#
I've managed to store values as 32 bit floats in the exrs but enblend's
output contained other glitches (like black dashed lines on the sky)
that I don't know how to interpret or debug.
So you may have two options: use enfuse to make hdr (as ldr :-) ) from
multiple exposures or blend each exposure into one exposure layer and
then make the HDR. I don't know how big pictures are you trying to make,
so this can be another problem.
I hope this will help you a little.
Regards,
David Brodsky
dishio
> Assuming it's the usual (same) bug...
>
I see that...
The description of the symptoms corresponds to what I experienced.
Unfortunately, turning the final output in Hugin from exr to tif does
not help.
I guess that must be because the intermediate stages to create the
final hdr (tif or not) output are in .exr.
At least that's what i can see from the log file.
Now any idea how to work around this issue? Either specifying the hdr
rendering to include the "--fix-halfmax" switch or to use tif in the
intermediate stages...?
I am also thinking of manually overriding the exposure all the photos
and settig them down by 5EVs for instance. Maybe that would prevent
clipping.
dishio
dishio
> this may be caused by the fact that exr produced by nona can contain
> pixels whose values cannot be represented by exr's type half. These
> pixels are then represented by infinity thus corrupting whole image.
>
Thx for the link.
Well i found a very practical solution. Because the exr is clipping, I
just decreased the exposition of all the photos composing a panorama
by a few EV (3 in my example)....and tada, it works.
Now this is funny that on has to decrease the exposure of the photos
when one is making a HDR...