Re: [PanoToolsNG] Enblend performance?
- paul womack wrote:
> paul womack wrote:Hmm. I have now done my own first pano
> (that he thought he knew how to speed up Enblend)
> Well, I don't. After some research, and substantial
> help from some members of this forum, here are my findings
> (summary at the end)
> Conclusion: enblend does a LOT of maths, in addition to IO.
> With regards to IO, best practise is to do rows then
> columns (at least for the sort of rectangular panos I
> was testing), and use the -a flag.
> In practise, it appears that EITHER of the approaches will suffice,
> which means that a single call with -a will serve quite well.
(more questions in other posts will follow).
The output was 3683 by 9779 (or 9779 x 3683)
Merged screen shot of pano previews here:
The enblend time was 90 minutes on my laptop,
thrashing the nuts off the disc.
Acting on a well informed hunch, I used ImageMagick
to generate 90 degree rotated cropped tiffs, and blended these
(making the seams horizontal).
enblend time 5 minutes!!
I think the difference is the size of an individual scan line
in the final output image, causing bad paging behaviour,
which is what I suspected all along.
Since I haven't managed to compile enblend,
I can't analyse much further.
It's just too frustratingm, since a simple compile
with -f would give answers "on a plate"