black lines from graph-cut seam generator

Kevin, I updated the enblend seam visualization code.

Would you please be so kind as to use the latest development version and send me the visualization output from the images you are having problems with?

Hi Mikolaj,

Ok, pulled it down and have version 4.1-e378b04ea3b7 installed now.  Same results as before.  PSM gives the correct output and GC gives the black lines.

>________________________________
> From: Mikolaj Leszczynski <email address hidden>
>To: <email address hidden>
>Sent: Thursday, May 3, 2012 10:40 AM
>Subject: [Bug 989908] Re: black lines from graph-cut seam generator
>
>Kevin, I updated the enblend seam visualization code.
>
>Would you please be so kind as to use the latest development version and
>send me the visualization output from the images you are having problems
>with?
>
>--
>You received this bug notification because you are subscribed to the bug
>report.
>https://bugs.launchpad.net/bugs/989908
>
>Title:
>  black lines from graph-cut seam generator
>
>Status in Enblend:
>  New
>
>Bug description:
>  Using enblend 4.1-e378b04ea3b7 I'm seeing black seam lines when the
>  graph-cut seam generator is used.  When I switch to use the nearest-
>  feature-transform seam generator they go away.  It's a large stitch
>  involving 13 images.  You can see the difference between the outputs
>  here:
>
>  http://www.bluelavalamp.net/hugin/enblend-seam/gc-pano_exposure_0006-small.jpg
>  http://www.bluelavalamp.net/hugin/enblend-seam/psm-nft-pano_exposure_0006-small.jpg
>
>To manage notifications about this bug go to:
>https://bugs.launchpad.net/enblend/+bug/989908/+subscriptions
>
>
>

Hi Kevin,

The latest revision should be 0c303c9ed15b, it's important - seam visualization output was changed between the two revisions.

Your stitch will most probably come out wrong again, but what I want to see is the seam visualization output (add the --visualize option to enblend arguments). If you could also narrow it down to the problematic group of images/seams it would be perfect.

Oh, sorry Mikolaj, I didn't do the --visualize as you asked before.  I pulled down what I think is the development version but I'm getting the same version number which is different that what you have here.  Here's the command I'm using to pull down the code:

hg -v clone http://enblend.hg.sourceforge.net:8000/hgroot/enblend/enblend

and the version number that comes back:

./enblend --version
enblend 4.1-e378b04ea3b7

It looks like I'm not getting the development version, what's the proper command to download it?

Thanks

>________________________________
> From: Mikolaj Leszczynski <email address hidden>
>To: <email address hidden>
>Sent: Friday, May 4, 2012 6:33 AM
>Subject: [Bug 989908] Re: black lines from graph-cut seam generator
>
>Hi Kevin,
>
>The latest revision should be 0c303c9ed15b, it's important - seam
>visualization output was changed between the two revisions.
>
>Your stitch will most probably come out wrong again, but what I want to
>see is the seam visualization output (add the --visualize option to
>enblend arguments). If you could also narrow it down to the problematic
>group of images/seams it would be perfect.
>
>--
>You received this bug notification because you are subscribed to the bug
>report.
>https://bugs.launchpad.net/bugs/989908
>
>Title:
>  black lines from graph-cut seam generator
>
>Status in Enblend:
>  New
>
>Bug description:
>  Using enblend 4.1-e378b04ea3b7 I'm seeing black seam lines when the
>  graph-cut seam generator is used.  When I switch to use the nearest-
>  feature-transform seam generator they go away.  It's a large stitch
>  involving 13 images.  You can see the difference between the outputs
>  here:
>
>  http://www.bluelavalamp.net/hugin/enblend-seam/gc-pano_exposure_0006-small.jpg
>  http://www.bluelavalamp.net/hugin/enblend-seam/psm-nft-pano_exposure_0006-small.jpg
>
>To manage notifications about this bug go to:
>https://bugs.launchpad.net/enblend/+bug/989908/+subscriptions
>
>
>

Kevin,

sorry for the confusion, it seems I forgot to change the version number. You should have the version we need. You can do a sanity check - if hg tip returns a patch with the description 'Extend visualization to work also for graph-cut' all is well.

I checked with the hg tip command and I get the 'Extend visualization to work also for graph-cut' description.  Using this version the output is still the same, the black areas are still there.

Ok. Thanks for your patience so far. Please send me the visualization (--visualize) output, so that I may address the issue.

All four visualize tif files are available here:

http://www.bluelavalamp.net//hugin/enblend-seam/

If you need me to do any more tests just let me know, thanks for looking into this!

----- Original Message -----
> From: Mikolaj Leszczynski <email address hidden>
> To: <email address hidden>
> Cc:
> Sent: Monday, May 7, 2012 3:57 AM
> Subject: [Bug 989908] Re: black lines from graph-cut seam generator
>
> Ok. Thanks for your patience so far. Please send me the visualization
> (--visualize) output, so that I may address the issue.
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/989908
>
> Title:
>   black lines from graph-cut seam generator
>
> Status in Enblend:
>   New
>
> Bug description:
>   Using enblend 4.1-e378b04ea3b7 I'm seeing black seam lines when the
>   graph-cut seam generator is used.  When I switch to use the nearest-
>   feature-transform seam generator they go away.  It's a large stitch
>   involving 13 images.  You can see the difference between the outputs
>   here:
>
>   http://www.bluelavalamp.net/hugin/enblend-seam/gc-pano_exposure_0006-small.jpg
>
> http://www.bluelavalamp.net/hugin/enblend-seam/psm-nft-pano_exposure_0006-small.jpg
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/enblend/+bug/989908/+subscriptions
>

Hi Kevin,

Unfortunately I can't locate the issue using visualize images alone, I will need your images in tiff format, just as they are right before the stitch, with all other data set up.

It would be perfect if you could also narrow it down as much as you could. Ideally I would like to see a pair of images where the issue appears, but several will also do.

Thank you for helping us nail this bug!

Hello,

Is this safe to assume that this is the same bug that was already reported in 2009?
http://www.panoguide.com/forums/qna/7107/?page=1
http://tech.dir.groups.yahoo.com/group/PanoToolsNG/message/34892

Just got hit by it (or something very similar) using hugin & dependencies from Debian (enblend 4.0+dfsg-1 - therefore 4.0-753b534c819d + whatever Debian diff). I may also have a slightly different kind of noise (dots) in addition to lines.

I would need to severely trim the panel from its current 115 frames to be able to upload something for you to reproduce. Before doing so, I'm just "testing the water" to see if there's something else on that bug that I should know first. After 36 hours of rendering, I feel a bit like walking away.

Regards,

JMR

Update...

I'm now down to a more manageable 2-frames panel, but still in fairly high resolution (2 x 3200x4500 => 60000x6000 or 2 x 25MB => 50MB). Problem still present.

I also tried the official 4.1 release on XP 32-bit. Problem present there too, although the lines are in different locations.

Nothing fancy in the command line - actually extracted from Hugin's generated makefile:
enblend --compression=LZW -f59793x6071+1554+9345 -o ZZZTest4.tif -- ZZZTest40000.tif ZZZTest40001.tif

While downsizing, I didn't got the bug everytime. I'm under the impression that it happens more often if the canvas is larger - the size of the sources images seems less important.

On the positive side, the final "writing output file" step is 1-2 orders of magnitudes faster on 4.1/XP32 than on 4.0/Linux64 - a very nice improvement.

Comments?

Regards,

JMR