PDF version of this entire document

Experimental Framework

A lot of the necessary work for constructing an experiment is described in prior sections (loaders, GUI, etc.) and now that there is handling of data with aim of input alignment14 we look forward to some experimental work and results that support the main hypothesis, which is about GMDS having an upper hand over PCA, at least for faces. At the moment, until pairs of images are set up for the initial, proof-of-concept experiments which validate the approach by showing extraction of biologically sound eigenvectors (we have rudimentary code for eigenfaces too), the least we can do is conceptualise and write down expectations.

While we look forward to seeing something similar to the pictures in the IJCV paper, in which the authors attempt to synthesise expressions via the eigen-structures, it may be necessary to vary another set of parameters when dealing with GMDS.


Roy Schestowitz 2012-01-08