Page 1 of 1

[APG 2.5RC2 WinXP/32] Problem image set

PostPosted: Tue Jan 18, 2011 1:34 pm
by mediavets
I like to try and stitch image sets that others present as challenging.

This set of seven was supposedly shot with a Nikon D3X witha Samyang 14mm WA rectilinear lens. In an irregularly spaced single row.
http://www.chocolatehouse.pl/samplepanoforum.rar

The only way I can get APG 2.5RC2 to give me a decent stitch is to treat the images as if they were fullframe fisheye images, with Detection quality set to High, which doesn't make sense to me.

PostPosted: Tue Jan 18, 2011 6:08 pm
by hankkarl
What counts as "decent"? I got RMS under 6, but some images were severely warped.

Note that APP sees the lens as 12mm, not 14.

Did a detection for 12mm rectangular lens, detection quality set to high. Image 1 linked to both 6 and 7 but not 2. Cleared the link from 1 to 6, added link from 1 to 2, optimized, set multiple viewpoints on and reoptimized.

APP seems to have issues with very wide rectilinear lenses, or perhaps wide rectilinear lenses distort the image to make it rectilinear, and APP "undistorts" it, then optimizes it.

PostPosted: Tue Jan 18, 2011 6:49 pm
by mediavets
hankkarl wrote:APP seems to have issues with very wide rectilinear lenses, or perhaps wide rectilinear lenses distort the image to make it rectilinear, and APP "undistorts" it, then optimizes it.

Yes, I agree, APP/APG does seem to struggle bit with wide angle rectilinear lenses. I have read that such lenses tend to exhibit quite complex distortions.

Perhaps Alexandre can comment on that?

FWIW several PTGui users claimed they were able to stitch those images and didn't observe/report that it was that much of a challenge.

PostPosted: Wed Jan 19, 2011 8:46 am
by AlexandreJ
Really short rectilinear focal length can be a bit more complicated to detect ( not to optimize in fact ).
So, one solution that works quite well is to use ransac model 'homography' when having short focal length rectilinear lenses. It works better in detection.
The rest ( meaning lens model, optimization ) is working great.

Issue 539 opened ( to check the case and confirm the above hypothesis )

PostPosted: Mon Jan 24, 2011 5:18 pm
by AlexandreJ
I cannot access these images => download error.