Pose Space Deformation:A unified approach to shape interpolation

and skeleton-driven deformation

J.P.Lewis, Matt Cordner, Nickson FongCentropolis

Character animation

1) Body: 1) “skeleton subspace deformation”

2) Face animation: 1) Shape interpolation (Tony de Peltrie,

Star Wars, Dinosaur) 2) Custom procedural models (Parke,

Toy Story)3) Common practice for 10-15 years

Pose Space Deformation

Combine direct geometric shape interpolation and skeleton driven deformation

address the limitations of Shape Interpolation and Skeleton-driven deformation

Shape interpolation

“Blend shapes, multi-target morphs, ”Weighted sum of key shapes

Or delta shapes...

k

kk SwS

delta shapes: equivalent!

(combinations)

Shape interpolation (delta)

)( 01

0 SSwSSk

kk

11

0)1( kkk SwSw

Shape Interpolation

Advantages– direct manipulation– desired targets are achieved

Disadvantages– Superposition, not independent,

linear subspace

Skeleton Subspace Deformation (SSD)

“enveloping”, ... Symbolics, Softimage

Weighted sum of surface point as transformed by various coordinate frames.

Diagram

SSD problems: collapse

SSD problems

Collapses,No direct manipulation,You can’t tell when you’re done

Goal

Combine direct geometric shape interpolation and skeleton driven deformation– BODY: use shape interpolation to control

shoulder shape as it deforms according to the skeleton

– FACE: drive the face using jaw rotation rather than linear interpolation

address the limitations of SI, SSD

Related work

Combine skeletal and local deformation control – Burtnyk and Wein, 1976

Related work: Idea #1

Animation as scattered interpolation – Van Overveld, 1990; – Litwinowicz & Williams 1994

Apply wide literature of interpolation directly to our animation problem!

Related work: Idea #2

Animation, etc. as general functional mapping – Poggio & Brunelli 1992; – Lewis 1988, 91

Synthesis

attempt #1– Lewis 95: shape interpolation in the context

of SSD

Related work

Symbolics/Nichimen 91Maya

AdditionInte

rpol

atio

n

Pose space deformation

Left Upper Arm

R. Upper Arm

Nec

k Jo

int

Pose Space:N Dimensions

Surface CV movement:3 dimensions

Surface cvmovement

Skeleton driven PSD

elbow, f( rotation, load ) face is f( jaw rotation, emotion,

phoneme...) quadruped leg/chest/neck

Scattered Interpolation

ShepardWiener interpolation, KrigingOthers...Radial Basis functions

– universal approximation– smooth if smooth kernel– large literature

VIDEO: Skeleton driven PSD

use same set of sculpted key shapes/delta shapes

Shape Interpolation is superposition; PSD interpolates

SI A B C crossfade is piecewise linear; PSD is smooth if desired

PSD vs. Shape Interpolation (SI)

Shape Interpolation vs. PSD

•adds•one dimension per shape•not orthogonal•animator manually solves for the representation

•interpolates•shapes placed as desired in desired space

Adding smirk counteracts raiseDinosaur: 100+ face shapes

Shape Interpolation: reinforcement / fighting

raise smirk

+ =

PSD facial animation

Delighted ? SerenePleased

Distressed

<neutral>

<linear>

VIDEO: PSD facial animation

Sleepy

Aroused

PleasureDispleasure

J.A.Russell, 1980 (simplified) delighted

serenetired

frustrated

alarmed

Conclusion

Unification / simplification / modest improvement to existing common practice

How many poses? Like shape interpolation, more is better, up to a point...

Future: Learn the space? Example code: www.idiom.com/~zilla/PSD

(september)

SSD Schematic

Robot arm, two cylinders bent at the elbow

SSD Schematic

Related work

Combine skeletal and local deformation control (Burtnyk and Wein, 1976)

Animation as scattered interpolation (Van Overveld, 1990; Litwinowicz & Williams 1994)

Animation as general functional mapping (Poggio & Brunelli 1992; Lewis 1988, 91)

Lewis 95

PSD algorithm steps

Definitions: pose, pose spaceSculptDefine delta(pose)SolveSynthesize, evaluate, adjust,

repeat

PSD vs. Shape Interpolation

SI: adds, not orthogonal,one dimension per shape

PSD: interpolates; shapesplaced as desired in space

PSD/SI facial animation

SI shapes define independent axes, PSD shapes are positioned as desired in desired space

PSD: animator plans interpolation path. SI: animator also has to solve for this path in a non-orthogonal basis

PSD is smooth (if desired)

Related Documents

Marketing plan– Location or contact name/phone

Budget– Location or contact name/phone

Post mortem– Location or contact name/phone

Submit questions– Location or contact name/phone

Project Goals

Ultimate goal of projectRelationship to other projectsHigh-level timing goals

Description

Describe the project in non-technical terms.

Use following slides for discussing status, schedules, budget, etc.

FOR MORE INFO...

List location or contact for specification (or other related documents) here

Competitive Analysis

Competitors – (You may want to allocate one slide

per competitor)Strengths

– Your strengths relative to competitorsWeaknesses

– Your weaknesses relative to competitor

Competitive Analysis, Cont.

CompetitorsStrengthsWeaknesses

FOR MORE INFO...

List location or contact for competitive analysis (or other related documents) here

Title:plot.psCreator:gnuplot 3.7 patchlevel 0Preview:This EPS picture was not savedwith a preview included in it.Comment:This EPS picture will print to aPostScript printer, but not toother types of printers.

Technology

New technology being used– Benefits

Standards being adopted– Benefits

Standards specifically being ignored– Drawbacks & benefits

DYA: define your acronyms!