Harmonic Coordinates for Character Articulation

How to read this

Category

Cage-based volume deformation method (rigging paper)

Contributions

• Introduces harmonic coordinates, generalized barycentric coordinates computed as solutions to Laplace's equation inside a cage, rather than in closed form.
• Proves harmonic coordinates are non-negative everywhere inside the cage, even for strongly concave cages, unlike mean value coordinates which can go negative and cause points to move opposite the cage handle.
• Shows the coordinates have interior locality, meaning influence falls off with distance measured inside the cage rather than in a straight line, so moving one limb's cage does not tug at a nearby but topologically distant limb.
• Allows optional interior vertices, edges and faces inside the cage to give animators finer local control without meshing the whole interior, unlike free-form deformation lattices.
• Demonstrates a working cage-based rig usable on production character models made of many separate primitives, such as skin, teeth, clothing and eyeballs, bound to a single cage.

Context

The paper sits between free-form deformation (Sederberg and Parry 1986) and Ju et al.'s 2005 mean value coordinates, which it directly improves on: mean value coordinates gave cage flexibility but leaked negative weights on concave shapes like bent legs, and harmonic coordinates fix that at the cost of needing a numerical solve instead of a closed form. The DeRose subdivision surfaces work in the lineage is the underlying Pixar modeling substrate that this cage-based articulation sits on top of. After this paper, the field moved toward bounded biharmonic weights (Jacobson et al. 2011) and later cage-free skinning methods that chased the same non-negativity and locality goals with cheaper precomputation.Builds on: Mean Value Coordinates for Closed Triangular Meshes · Subdivision Surfaces in Character Animation

Correctness

The paper proves its non-negativity and interior locality claims mathematically from properties of harmonic functions, and backs them with side by side renders against mean value coordinates on a bipedal character. The main limitation the authors are upfront about is cost, harmonic coordinates have no closed form and must be solved numerically as a precompute, which is fine for a bind pose but was a real engineering concern at 2007 hardware and mesh resolutions. That precompute cost is largely why the technique never became a runtime skinning method, it survives mainly as a cage-deformation tool and academic reference point.

Clarity

Well written and largely self-contained for anyone comfortable with barycentric coordinates and PDEs, the SIGGRAPH page-count format keeps the theory tight. A rigger without a numerical methods background can still follow the practical argument, Figure 2's mean value versus harmonic comparison does most of the persuading, even if the Laplace equation proofs in section 2 wash past them.

How to read it

First pass, read the abstract, look at Figure 2's side by side deformation comparison, and read the two lists of desired properties in the introduction, that alone tells you what problem is being solved and why mean value coordinates fall short. Second pass, read section 2's seven properties and skim the proofs to see how non-negativity and interior locality follow from harmonic functions achieving extrema on the boundary. Third pass, if you rig in a DCC that still ships a cage or lattice deformer, compare its weight computation against this paper's formulation to see whether it inherited harmonic coordinates, mean value coordinates, or a cheaper approximation.