CT2Hair: High-Fidelity 3D Hair Modeling Using Computed Tomography

How to read this

Category

Hair capture and reconstruction method (data-driven groom acquisition)

Contributions

• Scans real-world hair wigs with computed tomography (CT) to recover volumetric hair density fields that see through occluding strands, something camera-based capture cannot do.
• Extracts guide strands from the CT density volume by estimating 3D fiber orientation fields, then fills in dense strand coverage using neural interpolation between guides.
• Refines the generated strands so they conform back to the input density volume, closing the loop between the captured data and the final strand geometry.
• Demonstrates the pipeline across a range of hairstyles from straight to tightly curled, released as open source with example scan data.

Context

CT2Hair follows a long line of camera and structured-light hair capture work but changes the sensing modality to CT, trading camera-rig accessibility for the ability to see hair strands that self-occlude, which is most curly and dense hairstyles. It has no listed predecessors in this archive, reflecting that the CT based approach is a departure from the image-based capture lineage rather than a direct extension of it, though its downstream use of the resulting strands connects to the artistic and physical hair simulation work also in this archive, such as Iben et al.'s curly hair simulation.

Correctness

The method's claims rest on reconstruction quality against the source wig scans and visual match to the original hairstyle, as reported in the ACM TOG publication and demonstrated in the released code and sample data. As with most CT-based capture work, practical adoption is limited by needing an actual CT scanner and a physical wig or hairpiece rather than a photo or video, a much higher access barrier than the multi-camera rigs used by prior hair capture papers, so it is best understood as a high fidelity reference method rather than a widely deployable production tool as of this writing.

Clarity

Reads as a standard SIGGRAPH TOG paper aimed at a hair and fiber capture research audience, comfortable for anyone with a graphics research background. The open source repository and sample data make it unusually easy to actually try compared to most capture papers in this archive.

How to read it

First pass, read the abstract and look at the strand reconstruction figures against source photographs of the wigs, since visual fidelity is the paper's main selling point. Second pass, read the guide strand extraction and neural interpolation sections to understand how sparse CT-derived guides become a full dense groom. Third pass, if evaluating it for adoption, check the GitHub repository for what CT hardware and preprocessing the pipeline actually assumes, since that access requirement is the main practical gate on using this method outside a research lab.