On November 10, 2010 10:19:35 pm Robert Krawitz wrote:
> One more panorama from the trip last month:
> http://rlk.smugmug.com/Other/Landscapes/4851912_oeCNm#1086043156_p4b23

nice work, thanks for sharing.

 
> Hugin kept giving me FOV estimates of between 87 and 95 degrees HFOV

don't worry, as long as it is < 360° (i.e. fits on the panosphere) for partial 
(non 360° panos) accuracy of estimated HFOV is much less important than the 
visual result.


> The amount of error was truly extreme -- the worst CP was about 85
> units off and the average was 8 units.  However, there were only a
> couple of bad seams when I actually stitched it together, and they
> were all in locations that were fairly easy to fix.  Interestingly,
> when I added more CPs, the worst error went down to 70, but there were
> more seams.  I guess the thing to worry about is what happens when
> it's stitched, not what the optimization results are (Yuv, I guess
> that's what you were telling me :-) ).

yep, you guess right!  in this particular case, parallax affects the lower 1/3 
of the image more than the top 2/3.  If I was to add CPs manually, I would 
only put them on the far away mountains and maybe on the far away trees.  The 
CP generator does not know this.  It does not have a sense for depth, and it 
does not know that CPs at infinity suffer much less from parallax than CPs in 
proximity of the camera's viewpoint.  

points in the lower 1/3 will have higher error.  the cleaning/pruning function 
of CPs in Hugin may or may not catch them - depending on where the majority of 
the points are found.  running it multiple times (i.e. clicking the button 
multiple times) may help or may make things worse.

the random/regular pattern on the surface of the rock makes it very easy to 
deal with bad seams in your favorite image editor.

Yuv

Attachment: signature.asc
Description: This is a digitally signed message part.

Reply via email to