Simulation-Ready Hair Capture

How to read this

Category

research paper, hair capture and simulation parameter estimation

Contributions

• Presents the first method to capture dynamic hair from multi view video and automatically estimate physical simulation parameters that reproduce the observed motion, not just the observed static shape
• Formulates parameter estimation as a black box optimization, particle swarm optimization, over an energy function combining strand orientation, optical flow, and silhouette terms measured against the multi view video, agnostic to which hair simulator is used, demonstrated on both a super helices model and discrete elastic rods
• Shows that static only parameter inversion, the prior approach of Derouet-Jourdan et al. 2013, is insufficient: very different physical parameter sets can match the same static hairstyle under gravity yet diverge sharply once the hair is set in motion
• Produces a fully simulation ready hairstyle, captured geometry plus physical parameters, that can then be re-simulated under novel head motion, added forces like wind, or artistic edits, while staying faithful to the captured hairstyle's real behavior

Context

The paper builds on the multi view static hair digitization line, including Hu et al.'s own 2014 robust hair capture work, and directly extends Derouet-Jourdan et al.'s 2013 static parameter inversion into the dynamic domain. It sits inside Disney Research and Thabo Beeler's broader capture driven CFX research program, and it anticipates later data driven and learned approaches to hair parameter estimation that arrived once neural methods displaced particle swarm style black box optimization.Builds on: Robust Hair Capture Using Simulated Examples

Correctness

Validated by comparing re-simulated results against held out captured motion and by testing generalization to novel head animations, but only on a small number of captured hairstyles and only two underlying hair simulation models, not a large benchmark. The ten camera capture rig and per actor calibration make the pipeline heavy, and the results are only as faithful as the chosen simulator's ability to represent real hair-hair and hair-head interactions.

Clarity

A genuinely technical Eurographics paper with a real optimization formulation, energy terms and particle swarm optimization, but the writing is clear and the core motivating example, Figure 2's demonstration that static capture under gravity is ambiguous, is intuitive even without following every equation.

How to read it

First pass: abstract, Figure 2, and the end of the introduction, that trio motivates the entire paper by showing why static inversion fails. Second pass: Section 4's energy formulation, worth following closely if implementing something similar, otherwise skim it. Section 2's related work is a useful map of the hair capture literature up to 2017 even for readers who do not need the main method.