PDF version of this entire document
In the 2004 paper introducing GPCA [34], page 11 shows and explains an experiment reminiscent of Bronstein & Kimmel work (which was misrepresented somewhat by UWA, based on a toy experiment from an IEEE journal in 2006). It ought to be possible to implement a GPCA approach alongside the existing framework, i.e. without interfering with existing methods and thus facilitating performance comparisons.
Ah...agreed. I will try to get multidimensional scaling results within days.
