[ptx] Registration testcases

Pablo d'Angelo pablo at mathematik.uni-ulm.de
Wed Jan 21 07:16:11 GMT 2004


Hallo Sebastian!

Sebastian Nowozin schrieb am Mittwoch, den 21. Januar 2004:

> 
> 
> Hi Pablo & *,
> 
> 
> done the Muenster overview testcase tonight (30 minutes for keypoint
> generation, 4 hours for the correlation), and the result is here:
> 
> http://user.cs.tu-berlin.de/~nowozin/sift/muenster-auto-1300.jpg

Nice, result, as expected. I really think all these results are truely
great, and we're definately playing in the first league :)

It contains some seams, but that could also be due to local minima in the
optimisation.

> I changed no control points, but optimizing the panorama was non-trivial.
> Wouldn't it be possible to optimize the panorama incrementally?

Idealy, one would do a pairwise registration after the matching, and use
this to create an approximate initialisation for all images, before starting
a global optimisation.

What I usually do is (for big panos), set some of the pics behind my back to
yaw=180, and maybe some to 90,-90 then optimize for yaw only, then for yaw
and pitch and after that for yaw, pitch, roll.

ciao
  Pablo
-- 
http://wurm.wohnheim.uni-ulm.de/~redman/
Please use PGP


More information about the ptX mailing list