"enblend: warning: unable to run Dijkstra optimizer" for large panorama

[Ulrich-windl]

Reported for Hugin on Windows (32-bit version of Hugin): https://bugs.launchpad.net/hugin/+bug/1633352
I suspect it's an enblend bug. I'm no expert on enblend, so I could use some help how to diagnose, isolate and fix the problem.

[Christoph Spiel (cspiel)]

The warning is benign. It is not related to the size of a panorama,
but to the location of the tentative seam-line after
Simulated Annealing.

Use `--visualize' to check if there is a connection between the
seam line and the supposed artifacts. For a more detailed look
at the optimizer's job try
    `--parameter=debug-path'.
You will want to divert this output to a log-file.

Probably any of the optimizers' (S/A and Dijkstra) parameters
that can be set at the Enblend commandline may influence the
occurrence of the warning. So, you could run some experiments
along this line, too.

If you really want to jump into the source, "postoptimizer.h"
and "path.h" are your starting points. Though I doubt whether
one can improve on the situation, I'll be happy to accept your
patches. TIA!

[Ulrich-windl (ulrich-windl)]

When running enblend with --visualize and path debugging, the program
crashes. The effect is stat output suddenly stops, and when I click the Hugin windows, a message pops up that the program is not responding, then that it had crashed. Most unfortunately the log is not saved then! (I made a screen shot of the window when output had stopped.
However when only using --visualize, the program does not crash. Maybe there is some LZW issue, because for the same image sometimes an LZW warning ("enblend: error: Deflate compression support is not configured") appears, while it does not with a different --dijkstra parameter.
Still with different dijkstra parameters there is always a severe artefact in the resulting image.

Attached is a demonstration of layering the mapped images in GIMP without enblend, so you can see that there are no severe problems in the whole panorama. Another image shows the mapped images and the corresponding vis-files and the artefact with some annotation by myself in red.

I'm rather clueless what to do.

[Christoph Spiel (cspiel)]

> When running enblend with --visualize and path debugging, the program
> crashes. The effect is stat output suddenly stops, ...

        It is possible that `--parameter=debug-path' does not jibe
well with OMP. (Any `--parameter' lets you rummage in the guts of
Enblend or Enfuse.) Try a non-OMP version or set `OMP_NUM_THREADS=1',
though neither trick may cut it.

> Maybe there is some LZW issue, because for the same image sometimes an
> LZW warning ("enblend: error: Deflate compression support is not
> configured") appears ...

        This error is related to your TIFF library, which ought to fall
back to `no compression' in such a case.

> Still with different dijkstra parameters there is always a severe
> artefact in the resulting image.

        Which is what I have suspected in #1: "The warning [of the
Dijkstra optimizer] is benign."