Hi Aron,

Aron H schrieb:
I've just been successful in building on Win XP with VS 2005, and
started playing with panomatic_nopatent :). With the default settings
from the readme, on a 2 row, 360 panorama, I'm finding a number of
false matches. I'd like to provide helpful input, or other help,
however I can.

It currently is more sensitive to distortions than the other descriptors. I hope to improve this (and add a better ransac model for outlier checking). Unfortunately, I have been rather busy with some other stuff lately.

What kinds of cases is this algorithm suited for? What would give it
problems? My problem panorama has varying white balance and exposure,
for example. I think you mentioned some research papers about the
algorithm - can you point me to them?

Hmm, the algorithm is not really based on a published approach, just on some ideas that are mentioned in a few of them, and its currently simpler than any of them:

http://www.eecs.berkeley.edu/~aberg/gb.html
http://cvlab.epfl.ch/~tola/daisy.html
http://cvlab.epfl.ch/~brown/learndesc/learndesc.html

The descriptor is basically just x and y gradient strength computed in a daisy like layout, where the radius of the circles define the scale of the gradient filter.

The code is relatively simple: localfeatures/CircularKeyPointDescriptor.cpp

I did some quick evaluation and parameter tuning using two panoramas, to learn the best radi and scales, but probably some kind of aggregation of the gradients is needed to really improve the performance.

I have also fixed the typo you mentioned in a later mail, and liblocalfeatures will be build as a static library on windows now.

ciao
 Pablo

--
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

Reply via email to