[ptx] Even more SIFT successes
Pablo d'Angelo
pablo at mathematik.uni-ulm.de
Mon Jan 19 16:59:48 GMT 2004
On Mon, 19 Jan 2004, Mike Runge wrote:
>
> I think one of the earliest goal for hugin was to create a user-friendly
> interface to pano-tools. Meanwhile it has become much more :-)
> But panosifter within hugin could be a good chance to help newbies having an
> easy start. So I would not hide it for advanced use. I would try to make it
> the default or a part of a wizard function?!
Well, I'm dependant on Dr. Lowe's sift implementation, which is only
available for research use. people other than the ones doing research with
it wouldn't be able to use it. But I hope that we can use Sebastian's
implementation in the future. Then I'll add an interface into hugin.
It should be possible to choose different matching strategies, since I guess
that they will have complementary properties.
Maybe It would be nice to have a test set of panoramic pictures, with
different reqirements, so that the matching algorithms performance could be
tested against each other. for example I have one set of images, where
panosifter fails completely, because it contains very few, but repetive
structures, I guess algorithms with homography constraints, or similar
feature supression could handle it better.
> Maybe another function in hugin would be helpful, that cleans up some extreme
> mismatching controlpoints for image pairs with a lot of controlpoints?! Or
> the wizard could recommend a list of actions (deleting specific points,
> etc.)
Jep, that would be nice.. a delete all control points with errors above x
pixels function. Or multiselection in the control point list, and a delete
button.
> Maybe the future use of hugin could be to start with panosifter and (for
> advanced users) clean up some problems and optimize by hand?!
hmm jep. I'd like to automate that stuff as soon as possible.
Actually there are some other issues that need to be solved as well, like
better seaming algorithms.
I've came across a paper that compares multiple seaming techniques:
http://leibniz.cs.huji.ac.il/tr/acc/2003/HUJI-CSE-LTR-2003-82_blending.pdf
ciao
Pablo
--
http://wurm.wohnheim.uni-ulm.de/~redman/
Please use PGP
More information about the ptX
mailing list