PDF version of this entire document
We now have an adapted image loader for the 90 gigabytes or so of GIP data, much of which (volume-wise) comprises video sequences. The usage of this raw data is unclear at this stage because a systematic set of experiments needs to be decided on based on the utility of this data. For instance, one application might involve determining who is imaged based on an image sequence rather than standalone 3-D images. In prior work from Hack and Taylor there was a Ph.D. thesis on modeling (in the PCA sense) the sequence of talking faces in 2-D, basically predicting the likely subsequent frame based on a training phase. For purposes of straightforward face recognition we have less grainy images that also constitute more individuals and a wider variety of environmental conditions.
The important thing is that GIP data is readily available for loading shall this be required and the data is mounted from a remote storage server rather than stored locally on a computational server.
Roy Schestowitz 2012-01-08
