← ArchivePaper2024
Perm: A Parametric Representation for Multi-Style 3D Hair Modeling
Chengan He, Xin Sun, Zhixin Shu, Fujun Luan, Sören Pirk, Jorge Alejandro Amador Herrera, Dominik L. Michels, Tuanfeng Y. Wang, Meng Zhang, Holly Rushmeier, Yi Zhou
PCA-based frequency-domain strand disentanglement separates global hair structure from local curl patterns for precise editing and reconstruction as a generic prior.
Abstract
We present Perm, a learned parametric representation of human 3D hair designed to facilitate various hair-related applications. Unlike previous work that jointly models the global hair structure and local curl patterns, we propose to disentangle them using a PCA-based strand representation in the frequency domain, thereby allowing more precise editing and output control. Specifically, we leverage our strand representation to fit and decompose hair geometry textures into low- to high-frequency hair structures, termed guide textures and residual textures, respectively. These decomposed textures are later parameterized with different generative models, emulating common stages in the hair grooming process. We conduct extensive experiments to validate the architecture design of Perm, and finally deploy the trained model as a generic prior to solve task-agnostic problems, further showcasing its flexibility and superiority in tasks such as single-view hair reconstruction, hairstyle editing, and hair-conditioned image generation. More details can be found on our project page: https://cs.yale.edu/homes/che/projects/perm/.
How to read this
- Category
- Method: a parametric representation for 3D hair
- Contributions
- Perm, a learned parametric hair representation that disentangles global structure from local curl via a PCA-based frequency-domain strand model
- Decomposes hair geometry textures into low-frequency guide textures and high-frequency residual textures, parameterized by separate generative models
- Deployed as a generic prior for single-view reconstruction, hairstyle editing, and hair-conditioned image generation
- Context
- Builds on generative hair modeling with hierarchical latents (e.g. Zhou et al.'s GroomGen), proposing frequency-domain disentanglement for more precise control.Builds on: GroomGen: A High-Quality Generative Hair Model Using Hierarchical Latent Representations
- Correctness
- Assumes PCA in the frequency domain cleanly separates structure from curl and that this prior transfers across tasks; validity rests on the architecture ablations and the demonstrated applications, so a reader should gauge per-task quality rather than assume uniform superiority.
- Clarity
- Fairly accessible; a first pass conveys the disentanglement idea, a second pass for the strand frequency representation and the staged generative models.
- How to read it
- Focus on the strand representation and the guide/residual texture split; a second pass is worth it for the generative-model stages and how the prior is applied per task.
Builds on
Built upon by
Nothing yet.
Related work
- GroomGen: A High-Quality Generative Hair Model Using Hierarchical Latent Representations 2023 / SIGGRAPH Asia
- HAAR: Text-Conditioned Generative Model of 3D Strand-Based Human Hairstyles 2023 / arXiv
- HairNet: Single-View Hair Reconstruction Using Convolutional Neural Networks 2018 / ECCV
- Motion Guided Deep Dynamic 3D Garments 2022 / SIGGRAPH Asia
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 →