[ptx] hugin has problems with particular imageset

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

Zitat von Rich <rich at hq.vsaa.lv>:

> Pablo d'Angelo wrote:
> > Rich schrieb:
> >
> >> manouchk wrote:
> ...
> >> exactly. this is a single output image from hugin - all other are just
> >> like this, all software chokes on them. hugin did not process all
> >> pictures, though, it just closed before all of them were finished. and
> >> it took a lot of ram for a single picture.
> >
> > Sounds like nona is running out of memory and crashed. How big (output
> > image size) is your panorama?
>
> panorama is 31570x15580 - size calculated by hugin.

This is a very big image, 491 Megapixels. A single remapped image of the entire
panorama requires
31570*15580 * 4 = 1.8 GB. Nona might need 2 times the output image size
in TIFF_m mode, so this is up to 2.6 GB. A future version will be much more
memory efficient. Remember that on a 32 bit processor, each process can only
access up to 2 or 3 GB, no matter how big your swapspace is.

> to make sure there is enough ram, i added a couple of gigs of swapspace
> and monitored memory usage during stiching process. it never went over a
> couple of gigs (that is, there were still 2gb of freem ram at any given
> moment).
>
> > One more thing you could try, is to use nona with the multilayer tiff
> > output and use tiffcp to extract the single images. Then you can use
> > enblend on these images. This should need less memory
>
> yep, i also can stich it by separating the panorama before the process,
> but i would like to find out what exactly caused the problem.

Seems that you are running out of virtual memory. I will add an error message
when that occurs.

Using the TIFF_multilayer option of nona is more memory efficent, since it will
only need memory for the cropped, remapped images. Can you try if that works?
Running enblend on these files should also be much faster. Do not forget to
tell enblend how much physical memory it is allowed to use. (using the -m
option)

> i have reproduced this problem on several machines with latest cvs
> (050602) and rc1
>
> > ciao
> >   Pablo
> --
>   Rich
>
>
>




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