← ArchivePaper2022

Real-Time Hair Simulation with Neural Interpolation

Qing Lyu, Menglei Chai, Xiang Chen, Kun Zhou

Neural interpolation between precomputed hair simulation snapshots achieves real-time performance while preserving physically accurate dynamic behavior.

Open the paper ↗Trace in the lineage graph

Abstract

Traditionally, reduced hair simulation methods are either restricted to heuristic approximations or bound to specific hairstyles. We introduce the first CNN-integrated framework for simulating various hairstyles. The approach produces visually realistic hairs with an interactive speed. To address the technical challenges, our hair simulation pipeline is designed as a two-stage process. First, we present a fully-convolutional neural interpolator as the backbone generator to compute dynamic weights for guide hair interpolation. Then, we adopt a second generator to produce fine-scale displacements to enhance the hair details. We train the neural interpolator with a dedicated loss function and the displacement generator with an adversarial discriminator. Experimental results demonstrate that our method is effective, efficient, and superior to the state-of-the-art on a wide variety of hairstyles. We further propose a performance-driven digital avatar system and an interactive hairstyle editing tool to illustrate the practical applications.

How to read this

Category: Method: real-time hair simulation via neural interpolation
Contributions: The first CNN-integrated framework presented for simulating a variety of hairstyles at interactive speed
A two-stage pipeline with a fully-convolutional neural interpolator for guide-hair interpolation weights and a second generator for fine-scale displacements
A performance-driven digital avatar system and an interactive hairstyle editing tool as applications
Context: Builds on reduced-model and interactive hair simulation such as A Reduced Model for Interactive Hairs (Chai et al. 2014), replacing heuristic reduction with learned interpolation.Builds on: A Reduced Model for Interactive Hairs
Correctness: Trained with a dedicated loss plus an adversarial discriminator and reported superior across many hairstyles, but as a learned interpolator its fidelity is bounded by the training simulations and chosen guide hairs rather than a from-scratch physical solve.
Clarity: Accessible; a first pass conveys the two-stage interpolate-then-detail design, a second pass covers the loss and discriminator.
How to read it: Read first for the two-stage architecture and what each generator produces; second pass on the training losses if you want to reproduce the real-time quality.

Builds on

A Reduced Model for Interactive Hairs 2014

Built upon by

Nothing yet.

Related work

A Pixel-Based Framework for Data-Driven Clothing 2020 / SCA
Stable Spaces for Real-time Clothing 2010 / SIGGRAPH
PBNS: Physically Based Neural Simulation for Unsupervised Garment Pose Space Deformation 2021 / SIGGRAPH Asia
Cloth and Skin Deformation with a Triangle Mesh Based Convolutional Neural Network 2020 / CGF

Keywords

real-time hair simulation neural interpolation convolutional neural network guide hair hair interpolation weights generative adversarial network displacement generator mass-spring model linear blend skinning hair-hair interactions data-driven motion capture hairstyle generalization reduced simulation fine-scale details gan cnn physics-based

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 →