Style Components
 
Ari Shapiro
UCLA
ashapiro@cs.ucla.edu
http://www.cs.ucla.edu/~ashapiro
Yong Cao
UCLA
abingcao@cs.ucla.edu
Petros Faloutsos
UCLA
pfal@cs.ucla.edu
 
Download the paper
 
Abstract
We propose a novel method for interactive editing of motion data based on motion decomposition. Our method employs Independent Component Analysis (ICA) to separate motion data into visually meaningful components called style components . The user then interactively identifies suitable style components and manipulates them based on a proposed set of operations. In particular, the user can transfer style components from one motion to another in order to create new motions that retain desirable aspects of the style and expressiveness of the original motion. For example, a clumsy walking motion can be decomposed so as to separate the clumsy nature of the motion from the underlying walking pattern. The clumsy style component can then be applied to a running motion, which will then yield a clumsy-looking running motion. Our approach is simple, efficient and intuitive since the components are themselves motion data. We demonstrate that the proposed method can serve as an effective tool for interactive motion analysis and editing.
  
Walking and Sneaking
In this example, we transfer a style component between a walking motion and a sneaking motion. Joining motions and decomposing them into five style components allowed us to successfully identify one of the components that models the difference between the hunched posture of the sneaking motion and the upright stance of the walking motion. Applying this style component to both original motions produces two new stylized variations. The figure on the left shows a sneaky walk, while the figure on the right is a walk-like sneak. The latter motion appears to be the motion of a character tiptoeing in order to keep quiet, without the characteristic hunched posture of a sneaky motion.


original walk
original sneak
walk, sneaky-style
sneak, walk-style (tiptoeing)
 
Running and Sneaking
Here we combine a running motion with the previous sneaking motion. We find a similar style component that captures the hunched posture of the sneak, as in the previous example, and apply it to the run. The sneaky run is shown in figure on the bottom. Notice how the footsteps in the fast sneak are delicate and light the way the sneaking motion expresses, but with the speed of the run.

run and fast-sneak
  
Running and Walking
For this example we combine a running and a walking motion. A style component is found that captures the shrugged shoulders, the raised elbows and the bending of the knees of the running motion.The same style component captured the upright stance and relaxed arms of the walking motion. By applying the walking style to the run, our resulting motion resembles a jogging motion (top figure), while our run-like walk resembles a power walk (bottom figure).


run and jog
walk and powerwalk
 
Old Man's Walk and An Approaching Fighter
We combine an old man's walk with a threatening, fighter-like approach. The old man's arms shake while he walks, and his legs move in a bow-legged manner (top figure). The fighter moves with a steady upper body and a deliberate gait (middle figure). By finding a style component that captures the fighter's uprightness and raised hands, we synthesize a new motion showing an old fighter walking in an aggressive manner. The new motion retains the cadence, shaky arms and bow legged movement, but incorporates the raised hands and slightly raised head (bottom figure). /td>


<
old man
boxer
old man as boxer
Motion Interpolation The original and stylized motion retain very similar characteristics, including global translation and general movement speed. The alignment between these two motions eliminates problems such as foot-skating and phase differences when interpolating two different motions. Thus, the stylized motion can be linearly interpolated with the original motion in order to produce a continuum of motions that contain varying amount of style. The figure on the bottom shows an interpolation between the sneak and the walk-like sneak (tiptoeing).




interpolation between sneaking and tiptoeing
Visual Interface
Our interface is entirely visual. The user chooses style component that best represents the style difference between the two motions. The figure on the top shows visual representations of the style components. The original motion is reconstructed by adding together the components and is shown in white in the lower right.

The figure on the bottom shows an editing session where a run is combined with walk that raises the arms at the end of the walk cycle. Notice how the second style component (top, middle) captures the open arms of the walk.