PDF version of this document

next up previous
Next: Potential Development Up: Active Appearance Models Optimisation Previous: Current work

Alternatives

From the graphical description above it can be somehow assumed that pipelining is worth investigation so that each of the modules in the cycle is kept busy. This of course would require an architecture that supports parallelism and is therefore not quite usable in typical work environments. Also, due to really heavy dependency in this cyclic system, many guesses need to be made and effort is thus being going wasted.

Some specific points of weakness were often described in technical reports in the field. There are many alternatives that were not taken into consideration, often because they did not appear trivial or had not been used before. It is often concluded that such alternatives are not constructive, whereas more widespread algorithms show higher success rates. For instance, no apparent attempt has been made to assign dynamically changing weights to the different components in the objective function and no wide range of warps has been made available in the transformation ``arsenal''.



2004-07-19