On Wed, Aug 12, 2009 at 1:26 PM, David Doria <daviddoria@...> wrote:
> Does anyone use RPL's RGRL? I took their most basic example
> (registration_blackbox.cxx) and tried to make it as simple as possible. I
> changed everything from affine to rigid transformation. Instead of using a
> whole image worth of points I used two sets of 3 points each (two noisy
> lines translated 1 unit in x from each other). I set the initial
> transformation to identity. I had to move the example to 3d since the rigid
> transform class doesn't support 2d (why??). After debugging for several
> hours, I can't get past this error:
> rgrl_feature_based_registration:: Empty match set!!!
> I looked into it and it seems like 1 nearest neighbor match is performed
> with a "too far" threshold of -1, which tells it to find a neighbor no
> matter how far away it is - so there should be no way the match set can be
> Can anyone look at this example (attached: rigid.cpp) and see if I have
> done anything wrong or if there are actually some issues with rgrl?
Haha - it seems like this is as ill-maintained as I thought!