HelpingHand : Example-based Stroke Stylization

HelpingHand: Example-based Stroke Stylization

Presenter: Feilong yan

Jingwan LuPrinceton University

Fisher YuPrinceton University

Adam FinkelsteinPrinceton University

Stephen DiVerdiAdobe Systems Inc.

Painting Device

A high quality stylus has 6 degrees offreedom (DOFs: 2D position, pressure, 2D tilt, and 1D rotation)(Note , at every point of trajectory)

Examples of 6DOF

2D trajectory alone 6DoF

A Painting with 6DoF

Current Problems

• Such devices are difficult for novices to control

• The devices with 6DoFs are relatively expensive

Is it possible to achieve 6DoFs with low quality devices and by novices?

This Paper

• Present a data driven method for synthesizing plausible pose data, optionally for synthesizing the stroke trajectory style.

This Paper

• Present a data driven method for synthesizing plausible pose data, optionally for synthesizing the stroke trajectory style.

• Two underpinning assumptions observed:

This Paper

• Present a data driven method for synthesizing plausible pose data, optionally for synthesizing the stroke trajectory style.

• Two underpinning assumptions observed:– Strokes of the same artist with similar trajectory

are drawn with similar hand pose (for Pose Synthesis)

This Paper

• Present a data driven method for synthesizing plausible pose data, optionally for synthesizing the stroke trajectory style.

• Two underpinning assumptions observed:– Strokes of the same artist with similar trajectory

are drawn with similar hand pose (for Pose Synthesis)

– Shapes of strokes are part of the “style” of a given artist (for Trajectory synthesis)

Connection

• With these two assumptions , we can use trajectory of stroke to query the library, whether pose or trajectory synthesis, the trajectory may be used to query

Data Driven Approach

Library of 6DoFs Strokes (By Artists)

Query (2DoFs)

Offline

Online Pose & Trajectory Synthesis Output

Pose Synthesis

Pose and TrajectorySynthesis

Offline

• Collect a library of strokes with 6D data by trained artists

• Compute the shape Feature Vectors (such as shape context) for every sample of the strokes, as the descriptor for comparison

Online

• Stroke by stroke• Compute the feature vectors for samples of the

current query stroke• With feature vector, search for the K-NN for

every sample.• For the current stroke, get the optimal sample

sequence from library using Dynamic programming

• Pose synthesis or trajectory synthesis

Result

Paper Stylization

Library of Beautiful Sentences

Query (poor exposition)

Offline

Online Sentence Synthesis

Desired Paper

Thank You