Hi Yuv and others, 2010/12/25 Yuval Levy <[email protected]>
> > I have uploaded your file to my server at > <http://www.photopla.net/hugin/namklim.zip> and strongly encourage > developers > to have a look / use it as a test case. > > This is the result of my testing on these files. cpfind was run on an > empty > pto file that is just listing the images (attached). panomatic was run on > the > images themselves. I only ran three tests. > > .... > these results are puzzling. they require more confirmation to see if this > is > an exceptional case, or if the issue affects more cases and we can only > find > this out if the code is out in real world use. > > In terms of speed, cpfind (default) is almost 20% faster than panomatic > (default), but what is speed if the quality is not there? > > What surprises me is that cpfind's results are worse at full scale than > resized. > > Is this a show stopper? I don't think so - shipping without cpfind is > worse > than shipping with cpfind as-is. But we definitely need more information > to > understand under which circumstances this problem happens. > > I have started a tracker ticket: > <https://bugs.launchpad.net/hugin/+bug/694329> > > Please continue discussion and testing there. > > Yuv > > > First of all: I'm a nitwit when it comes to fisheye panos as I don't do that (I only create wide angle panos like the Max Lyons ones on his gallery on tambaware. That's my cup of tea as well), but it is known that both cpfind and panomatic are not good at fisheye images. Pablo mentioned this at the time he was working on "pablomatic". And as we are talking about fisheye images here, the first thing I did was increase the "sieve1size" default value of 30 to 100 (--sieve1size 100), as that's recommended for fisheyes. Now this "--sieve1size 100" generates 20 CPs instead of your 4. However only a few images were connected. Then I used "--sieve1size 200" which generates 35 pairs of CPs, still 4 images not connected. Then with "--sieve1size 300" which generates 57 pairs of CPs. Still 2 images not connected. Then I used the "fullscale" option together with the "--sieve1size 200" option (still fisheye images, so I won't try without the increased size). CP pairs generated: 26. 5 images not connected (man, fullscale is slow!!). So indeed, fullscale delivers worse results. Then I did a final run with "--sieve1width 50 --sieve1height 50 --sieve1size 300" (no fullscale). This generated 64 CPs with only one image not connected: the 20101109_IMGP1465_KX.jpg, which is the nadir image. (From my limited knowledge this is always a hard image for the CPs, or not?) Optimization of the panos was weak to rubbish, be it that I don't have experience with fisheye panos, so maybe I did something not completely right. Finally: I would not qualify this as a showstopper. I myself never create fisheye 360x180 degree panos. cpfind always works for me! How many of the users are really creating 360x180 degree panos and who only create wide angle panos (like the Max Lyons ones I already mentioned.) Is there a way to make a poll to check the percentages of: - 360x180 degree only pano makers - 360 degree pano makers - wide angle pano makers. Harry -- You received this message because you are subscribed to the Google Groups "Hugin and other free panoramic software" group. A list of frequently asked questions is available at: http://wiki.panotools.org/Hugin_FAQ To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/hugin-ptx
