Hand animation is definitely going to look better than a canned set of parametric functions, but I've wondered why no one has generated (fit by vector regression or monte carlo) a specific set of parametric functions for a given set of hand-mapped animation data. It would be a form of compaction, really, since the degree of lossiness would be dependent upon the complexity of the polynomials that you generated, ie the maximum degree of each function, and the number of orthogonal functions in your basis set.
Generating your parametric functions from existant lifelike animation data is a much more sensible approach, as its essentially learning by example. Of course, it would be more difficult to animate using these functions, as the functions in your basis set would behave differently than any set a human would choose, but once you got used to it would be much more lifelike, much more versatile (you could have multiple sets of parametric functions for different people and moods, since any two sets of data would generate unique basis sets), would be much more compact, and would by its design glean the unique sense of personality for that being you were animating, in a best-case implementation.
If this is already done, oops. If not, seems like a good project.
Slashdot Mirror
Hand animation is definitely going to look better than a canned set of parametric functions, but I've wondered why no one has generated (fit by vector regression or monte carlo) a specific set of parametric functions for a given set of hand-mapped animation data. It would be a form of compaction, really, since the degree of lossiness would be dependent upon the complexity of the polynomials that you generated, ie the maximum degree of each function, and the number of orthogonal functions in your basis set.
Generating your parametric functions from existant lifelike animation data is a much more sensible approach, as its essentially learning by example. Of course, it would be more difficult to animate using these functions, as the functions in your basis set would behave differently than any set a human would choose, but once you got used to it would be much more lifelike, much more versatile (you could have multiple sets of parametric functions for different people and moods, since any two sets of data would generate unique basis sets), would be much more compact, and would by its design glean the unique sense of personality for that being you were animating, in a best-case implementation.
If this is already done, oops. If not, seems like a good project.
Aaron Schoeffler