Vid2Haircut: 3D Strand-Based Hairstyle Reconstruction from Video

How to read this

Category

Eurographics short paper on strand based hair reconstruction from video

Contributions

• Recovers a strand level three dimensional hairstyle from ordinary video rather than from a single photograph or a controlled multi view capture
• Targets simulation ready geometry, so the output is meant to be groomed and dynamically simulated rather than only rendered
• Presented in the short paper track, so the scope is a focused contribution rather than a full system

Context

This sits in the strand based hair reconstruction line that Vanessa Sklyarova and Justus Thies have been building out, from Neural Haircut, which worked from monocular video and multi view images, to Im2Haircut, which recovered hair from a single photograph. Vid2Haircut returns to video input with the benefit of those stronger hair priors.

Correctness

This is a two page short paper, so expect a focused result rather than an exhaustive evaluation against every hair reconstruction baseline. At the time of writing no open PDF was available, so the claims here rest on the title, the author group's published lineage and the Eurographics program listing rather than on a read of the method section.

Clarity

Short papers in this track are compressed by design: dense, with the method compressed into a page or two and the detail pushed into figures.

How to read it

First pass, read the abstract and figures to see what the video input actually buys over the single image case. Second pass, compare the pipeline against Im2Haircut, since the interesting question is which parts of the prior carry over and which had to change for temporal input. A third pass only makes sense if you are building hair capture yourself, in which case go straight to whatever the paper says about temporal consistency across frames.