← ArchivePaper2023
GroomGen: A High-Quality Generative Hair Model Using Hierarchical Latent Representations
Yuxiao Zhou, Menglei Chai, Alessandro Pepe, Markus Gross, Thabo Beeler
First generative model for dense-strand hair using a hierarchical strand-guide-complete VAE architecture; introduces Strands400, 400-subject reconstructed strand dataset.
Abstract
Despite recent successes in hair acquisition that fits a high-dimensional hair model to a specific input subject, generative hair models, which establish general embedding spaces for encoding, editing, and sampling diverse hairstyles, are way less explored. In this paper, we present GroomGen, the first generative model designed for hair geometry composed of highly-detailed dense strands. Our approach is motivated by two key ideas. First, we construct hair latent spaces covering both individual strands and hairstyles. The latent spaces are compact, expressive, and well-constrained for high-quality and diverse sampling. Second, we adopt a hierarchical hair representation that parameterizes a complete hair model to three levels: single strands, sparse guide hairs, and complete dense hairs. This representation is critical to the compactness of latent spaces, the robustness of training, and the efficiency of inference. Based on this hierarchical latent representation, our proposed pipeline consists of a strand-VAE and a hairstyle-VAE that encode an individual strand and a set of guide hairs to their respective latent spaces, and a hybrid densification step that populates sparse guide hairs to a dense hair model.
How to read this
- Category
- Method / model: generative model for dense-strand hair
- Contributions
- GroomGen, presented as the first generative model for dense-strand hair geometry
- Hierarchical latent spaces over single strands, sparse guide hairs, and complete dense hairs via a strand-VAE and hairstyle-VAE plus densification
- Strands400, a dataset of reconstructed strands across 400 subjects
- Context
- Moves from per-subject hair acquisition toward a general generative embedding space, building on volumetric VAE hair synthesis (saito-vae-hair-2018).Builds on: 3D Hair Synthesis Using Volumetric Variational Autoencoders
- Correctness
- The hierarchical representation is credited with compact latents, robust training, and efficient inference; as a generative model trained on reconstructed strands, output realism and diversity are bounded by the Strands400 data and the densification step that fills sparse guides into dense hair.
- Clarity
- Moderately accessible; a first pass conveys the three-level hierarchy and the encode-edit-sample goal, a second pass for the two VAEs and the densification.
- How to read it
- Focus on the strand-to-guide-to-dense hierarchy and why it makes the latent spaces tractable; a second pass pays off for the VAE designs and the hybrid densification.
Related work
- Perm: A Parametric Representation for Multi-Style 3D Hair Modeling 2024 / ICLR
- 3D Hair Synthesis Using Volumetric Variational Autoencoders 2018 / SIGGRAPH Asia
- HAAR: Text-Conditioned Generative Model of 3D Strand-Based Human Hairstyles 2023 / arXiv
- Neural Haircut: Prior-Guided Strand-Based Hair Reconstruction 2023 / ICCV
Keywords
This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →