[ptx] auto fine-tune on large images

[pablo.dangelo at informatik.uni-ulm.de]

Hi Rich,

> Pablo d'Angelo wrote:
> > Rich schrieb:
> >
> >> auto fine-tune is a nice tool, but on large images autotuning single
> >> point can take a lot of time.
> >
> > How large are large images? And how many of them are in your project.
>
> this is a 360 panorama in two rows, totalling 30 images. first i merged
> both rows separately (by running enblend on them on a separate machine
> at the end - it had more ram and as a result process was much faster)

Hmm, this is indeed very big. From your comments below, one image will probably
need more than 200 MB of RAM. This is quite a lot, and I have to admit that
while writing hugin, I was thinking about "normal" images taken with a camera.
There are some places where images are stored as duplicates, once in the image
cache and once as wxBitmap, suitable for direct display on the screen. Making
all that memory efficient requires some changes that can't be made before the
0.5 release.

> >> what could be done to speed up this process ?

Apart from not doing pre stitches, nothing at the moment. Is enblend really
that much slower when running on complete set of images?

You could keep all images in the project and remap them, and then use enblend on
the output images of the upper row, then on the lower row, and finally on the
last row, if that saves some time.

Also, enblend will work faster if it is told how much memory it can uses (there
is a commandline switch to do that).

> >> i checked finetune preferences and didn't change any parameters that
> >> were specified in pixels, only reduced 'search area width' (which
> >> didn't help a lot though, and it seems that it is constantly reset
> >> back to 10%)
> >> (rotation check is disabled)
> >
> > Ok, will check that. It shouldn't pop back to 10%.
>
> it seems that it reverts to 10% after each finetuning, at least for this
> pair of images (or maybe even _before_, as i can see no difference in
> performance when changing this quite radically)

this bug is fixed, but the problem is that hugin wasn't designed for input
images with more than 100 MB size..

ciao
  Pablo

----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.