GroomLight: Hybrid Inverse Rendering for Relightable Human Hair Appearance Modeling

TL;DR

GroomLight tackles the challenge of relightable human hair by marrying physics-based accuracy with neural-detail modeling. It introduces an extended hair BSDF to capture primary light transport, paired with a light-aware residual built from fixed-geometry 3D Gaussians and dual-level spherical harmonics to encode fine, view- and light-dependent details. The two-stage hybrid inverse rendering optimizes the BSDF parameters and then the residuals, enabling high-fidelity relighting, robust view synthesis, and practical material editing. Evaluations on real-world hair data show state-of-the-art performance in relighting and appearance reconstruction, with applications in relighting, editing, and dynamic rendering for CG workflows.

Abstract

We present GroomLight, a novel method for relightable hair appearance modeling from multi-view images. Existing hair capture methods struggle to balance photorealistic rendering with relighting capabilities. Analytical material models, while physically grounded, often fail to fully capture appearance details. Conversely, neural rendering approaches excel at view synthesis but generalize poorly to novel lighting conditions. GroomLight addresses this challenge by combining the strengths of both paradigms. It employs an extended hair BSDF model to capture primary light transport and a light-aware residual model to reconstruct the remaining details. We further propose a hybrid inverse rendering pipeline to optimize both components, enabling high-fidelity relighting, view synthesis, and material editing. Extensive evaluations on real-world hair data demonstrate state-of-the-art performance of our method.

Figures (11)

• Figure 1: GroomLight achieves high-fidelity reconstruction of human hair appearance from real-world images, enabling realistic rendering under diverse lighting conditions. Here, in each column, we present relighting results of a same subject under two different environments, using the appearance model reconstructed by GroomLight. One input view with the similar head pose is shown at the top right corner.
• Figure 2: Our hybrid inverse rendering pipeline employs a two-stage optimization scheme. Given input OLAT images and the reconstructed hair geometry (here we visualize the 1/10 downsampled version), we first estimate the physical parameters of the extended hair BSDF model (Stage 1, §\ref{['Sec3.2']}), and then leverage a light-aware residual model to capture fine-grained appearance details (Stage 2, §\ref{['Sec3.3']}).
• Figure 3: Impact of light scattering. We compare the effects of varying the maximum number of ray bounces during path tracing ($\infty$ means the integrator will not enforce any hard cutoff on the number of bounces). Our method employs 8 bounces to achieve high quality rendering at practical efficiency.
• Figure 4: Qualitative evaluation of rendering results at novel testing views under novel lighting conditions of our method and baselines.
• Figure 5: Ablation study of components in our method. Starting from Stage 1 only with original BSDF model (Stage1 origBSDF), we sequentially add spatially-varying albedo (Stage1 NoCorr), rotation correctives (Stage1 Full), and the residual model (Full).