[ptx] hugin, nona, enblend on amd64

[Pablo d'Angelo]

On Thu, 26 Aug 2004, Mike Runge wrote:

> 
> Congrats, Pablo.
> 
> You should compare that against the PTStitcher with/without Fulvio Seniores
> fast pano12.dll. I never measured the times needed, but I think nona is still
> a lot faster :-)

Hmm, it depends. for full spherical panos, fulvio's code could be faster in
some cases, because nona still remaps all the black pixels for images that
overlap the left and right border. In the test case I used, it was twice as
fast for as nona. For partial pano's it's a different issue. however,
PTStitcher and nona save .tif files with different compression, this could
also lead to some differences.

For the test pano,
http://wurm.wh-wurm.uni-ulm.de/~redman/gallery/pano_testcases/ulm_muenster_view

PTStitcher, using latest source from panotools.sf.net, with "f0" option:
real    4m7.609s
user    4m0.910s
sys     0m3.355s

almost twice as fast as nona.

For a partial pano
http://wurm.wh-wurm.uni-ulm.de/~redman/gallery/panoramas/bollenwies

the results are similar however, Fulvio's acceleration beats nona
also for partial pano's..

nona:
real    1m47.447s
user    1m45.101s
sys     0m1.813s

PTStitcher (f0 acceleration):
real    0m57.642s
user    0m55.650s
sys     0m0.814s

PTStitcher (no acceleration):
real    7m8.921s
user    7m6.806s
sys     0m0.682s

looks like I should add the approximation algorithm to nona as well :)

ciao
  Pablo

> On 8/26/2004, "Pablo d'Angelo" <pablo.dangelo at web.de> wrote:
> 
> >I used nona to remap a 24 image pano ( 11152 x 5483) to multiple tiff and
> >enblend to blend it. here are the times, for both Suse 9.1 (AMD64) and
> >debian unstable (i386, also gcc 3.3).
> >
> >           AMD64               i386
> >nona
> > real    7m22.307s          8m13.001s
> > user    6m52.438s          7m56.454s
> > sys     0m7.081s           0m8.480s
> >
> >enblend
> > real    17m19.074s          21m22.067s
> > user    6m59.866s           7m17.790s
> > sys     1m36.806s           0m58.766s
> >