GroomCap: High-Fidelity Prior-Free Hair Capture

How to read this

Category

Capture system: multi-view strand-level hair

Contributions

• A prior-free multi-view hair capture method reconstructing high-fidelity strand-level geometry without external data priors.
• A neural implicit hair-volume representation encoding high-resolution 3D orientation and occupancy, trained with a volumetric 3D orientation rendering algorithm and 2D orientation distribution supervision to avoid blending-induced loss of structure.
• Strands traced in the volume then refined by a chained-Gaussian optimization with direct photometric supervision, yielding dense scalp-rooted strands across diverse hairstyles with the same parameters.

Context

Advances strand-accurate multi-view hair capture (Nam et al., 'Strand-Accurate Multi-View Hair Capture') and relates to generative hair modeling (Zhou et al., 'GroomGen'), here pursuing fidelity without learned priors; best paper at SIGGRAPH Asia 2024.Builds on: Strand-Accurate Multi-View Hair Capture · GroomGen: A High-Quality Generative Hair Model Using Hierarchical Latent Representations

Correctness

Demonstrated across diverse hairstyles using one parameter set and supports re-rendering, physics animation, and editing; being prior-free it leans on multi-view input quality and the orientation-supervision scheme, which a reader should keep in mind for sparse or occluded captures.

Clarity

Dense but well-motivated; a first pass conveys the implicit-volume-then-trace-then-refine flow, a second pass for the orientation rendering and chained-Gaussian refinement.

How to read it

First pass for the two-stage pipeline (implicit volume to traced strands to Gaussian refinement) and why prior-free matters; second pass on the orientation supervision if reconstructing your own grooms.