Artefacts due to Enblend memory bug

[蜥蜴]

Please take a look at my equirectangular panorama
http://img11.imageshack.us/img11/7831/test24.jpg
What happened to those exposure line?
Also the nadir image result is equally weired.
If I use normal view with horizontal horizon ,it looks like something broken:
http://img18.imageshack.us/img18/6176/48761613.png
But if I chang the view and looking down, it appears fine:
http://img59.imageshack.us/img59/1462/47884439.png
I don't get it, is there anything I'm not doing right?

I've use manual mode for everything when shooting (e.g. white balance, aperture, exposure time..... )
I using prebuilt hugin for Windows 7 64bit.(HuginSetup_2011.2.0-beta1_64bit_Windows_2.exe )
Also take a look at the discussion at google form
 https://groups.google.com/forum/?pli=1#!topic/hugin-ptx/RQz4-I_LGss

[蜥蜴 (f039281310)]

Well, the original photo can't be uploaded to here, so please download at
http://www.mediafire.com/?byxuvph5npyek

[Bruno Postle (brunopostle)]

I stitched your project and see the same seam artefacts, I'll attach a scaled down version.

This is an enblend bug, though I have never seen it before. I used dummy images so the problem is caused entirely by the geometry of photos in your project (and not their content). The project looks fine, you have a good overlap, this should work.

A possibility are that enblend has some implicit expectation that panoramas are shot left-to-right, and that your right-to-left shooting style is exposing the bug. Though this seems unlikely.

You can remove all ordering by setting '-a' in the blender options before stitching, you could try this.

[Bruno Postle (brunopostle)]

I tried stitching with the enblend -a parameter, but this clearly isn't working (see attached scaled down version of the output).

The seams for the lower 70% of the image are very wrong, though the top 30% is working as expected.

This image also has the horizontal line artefacts that have previously been associated with memory cache size issues, my Hugin is using the installation default preferences for this.

So it seems that this panorama is too big for my 2GB system.

Reducing the size is likely to help, currently it has dimensions of 28634x14317. This is the size suggested by Hugin, but Hugin is very conservative about these calculations, a size of 20000x10000 will stitch much quicker without any real loss of information.

[蜥蜴 (f039281310)]

Yes, I also use the -a parameter and got the same result (see attachment). I'm not sure if the horizontal line artifacts is cause by lack of memory since my computer has 12GBs memory , so enblend have at least 10GBs of memory. Will it take such a large amount ?

[Bruno Postle (brunopostle)]

Reducing the size to 20000x10000 works for me, I see none of the artefacts, see attached stitched output. This is a good workaround if you want to carry on shooting at this resolution.

enblend has known bugs related to the memory cache which are responsible for the horizontal lines and presumably the other problems you see. You can workaround the cache problems by disabling it with the -b and -m parameters, though other people have more experience of this than me.

[蜥蜴 (f039281310)]

Yes, I've try to resize it to 22500 but still didn't work. But 20000 works fine.

[蜥蜴 (f039281310)]

I understand the -b and -m parameters are used to set unage of memory. But how should I set it? I have 12GB of memory and work with Win7 64bit .Operation System usually takes no more the 2GB after startup.

[蜥蜴 (f039281310)]

OK, I've used parameter '-m 10240' to enblend and it works! NO MORE ARTIFACTS !!! It is a 28184*12641 panorama. I think Bruno is completely right about that. It's probably a enblend bug I guess.

[Yuv (yuv)]

Thank you for providing a complete test case. It will be useful to test if the bug is really fixed later on.

[Thomas Pryds (thomas-pryds)]

I experience the problem with thin horizontal lines in my pano of 24388×12194. Lots of lines with standard settings, though less with setting enblend's -m parameter to 1536.

I have put the pto file, the mk file and all images (though heavily JPEG compressed, but same size pixel-wise) in a ~3 MB zip file here: http://goo.gl/93k20

If you need more info, please let me know.

My system info (from Hugin's about dialogue) is:

Operativsystem: Linux 2.6.38-13-generic i686
Arkitektur: 32 bit
Ledig hukommelse: 755452 kiB

Hugin
Version: 2011.4.0.cf9be9344356
Sti til ressourcer: /usr/share/hugin/xrc/
Sti til data: /usr/share/hugin/data/

Biblioteker
wxWidgets: 2.8.11.0
libpano13: 2.9.18
Boost: 1.42.0
Exiv2: 0.21.1

[Christoph Spiel (cspiel)]

The ImageCache causing the problem was withdrawn in the
Development Branch, which finally leads to version 4.2. So the
bug is gone there, because the code that caused it was removed.

We have a tentative replacement for the ImageCache in
the "mmap_view" branch:
    http://hg.code.sf.net/p/enblend/code/rev/7a3964af671a
However, the developers are still unsure whether the new code is
(a) fast enough and
(b) offloads sufficient image data from core memory to disk.
Right now it is uncertain whether "mmap_view" will make it into
the mainline Development Branch.

[tmodes (tmodes)]

Closing ticket in Hugins tracker.

[Christoph Spiel (cspiel)]

I have set the status to "In Progress", because the "mmap_view" branch
needs more testing from the users.