How many degrees of freedom in thier model for deformation? Dunno if they said, but I must have missed it if they did. I'm guessing it has to be a low number if they're running a ransac or ransac based algorithm for the matching.

Just to clarify, keypoints are single pixels, right? (not clusters of pixels?) If so, I am curious how robust the authors' method is when varying the resolution of the images.

hmm... I see.. The keypoint is one pixel, and for each keypoint, they choose a neighborhood of pixels around it. All these neighborhoods of keypoints from all the images help generate the view set. I wonder how they control for the size of neighborhood? Perhaps that does depends on the image resolution. (Perhaps [19] talks more about this.)

I think it is very interesting to see the keypoint matching problem formulated as a classification problem, in order to reduce the run time complexity. The synthesize view sets is a good idea to fight distortions and blur, but I guess they depend on the object to detect, since some objects may fire just few keypoints (or way to many). Adding white noise should help for recognize keypoints in cluttered backgrounds? This is still not very clear to me.
Using randomized trees it would indeed make fast matching, since we store a bunch of patch examples, but it is concerning that we have to deal with the issue of storing them on memory (It is still a lot just for one object). Building more than one tree seems to be a good approach for searching, but it is still not optimal (same size still).
The comparison made with SIFT keypoints may not be fair, since the software used could be the demo and not the licenced (or real) SIFT version. Moreover, other descriptors should have been good to include in the comparison.
This approach should work very efficiently for tracking objects, since the tree implementation uses simple and efficient tests.

And some more..
*

At last...
*

Hej!
Check this out!
,

And some more..
*

*

9j3jLV Hey, there is what you need.

Hello! great idea of color of this siyte!

If you want do delete your site from our spam bases - just email us with domain of your site:

abuse-here@inbox.ru

thank you!

97749c0399788455b22cf5837aabe4d4

Hello, nice site :)