[ptx] Nona problems on ubuntu/amd64

Christian Conkle conklech at grinnell.edu
Tue Aug 16 21:44:42 BST 2005


I could swear I actually wrote something in the last e-mail to the list!  Oh 
well.

I'm having a problem with nona and hugin on a Kubuntu system on amd64.  I 
followed Rob Park's guide to compiling libpano12, hugin, enblend, and 
autopano-sift, and everything seems to have gone well; I haven't gotten all 
the bits of Mono that autopano-sift's frontend needs, but other than that 
have had no compilation problems or that sort of business.

Hugin works just as it ought; I can open and edit and stitch 
previously-created panorama files.  However, I'm having problems with some 
newer images I'm trying to stitch: they're 2000x3008 pixel JPGs off a Nikon 
D70 I borrowed, and hugin appears to completely choke on them.  Stitches take 
much, much longer than they ought: the other day, I successfully stitched a 
13-image panorama (final resolution something like 11000x3000), but it took 
all morning, three or four hours, on an Athlon 64 3000.  The behavior was 
strange, too: it spent almost the whole time reporting that it was blending 
the first image, and once that finally completed, it went through the rest of 
the panorama hey-presto in a minute or two.  Through the whole time 'top' 
reported it pegged on CPU, and using about two hundred megs of memory, out of 
an available gigabyte.

I tried running gprof against a smaller stitching run, and have the results if 
anyone wants them, but really don't know anything about reading code profiler 
output and couldn't make much sense out of it.

I can't think of any good reason why marginally larger input files and a 
somewhat larger panorama ought to increase the stitching time more than a 
hundredfold; I suppose I'll try doing the same stitch under Windows or on a 
different system, but I'd like to figure out what's going on here.  Does 
anybody have debugging suggestions?  I can provide the material that's giving 
me trouble if wanted.

Thanks,

-Christian Conkle.


More information about the ptx mailing list