HairLRM: Strand-based Hair Modeling via Large Reconstruction Models

2026-06-13 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

HairLRM is a novel method for strand-based hair modeling that addresses the fundamental limitation of inferring complex 3D hair fields from 2D imagery. Traditional approaches suffer from ill-posedness, leading to failures in resolving global occlusion, such as in ponytails, and local directionality, like in curls, often producing over-smoothed or incorrect geometries. HairLRM integrates the strong geometric priors of Large Reconstruction Models (LRMs) into its strand generation pipeline. It utilizes the LRM mesh as a structural anchor and employs a novel Dual Orientation AutoEncoder to elevate coarse geometry into high-fidelity strands. The method resolves vector field singularities through latent-space optimization and surface-guided refinement, effectively disentangling complex topological structures and establishing a new benchmark for robustness and accuracy in hair reconstruction.

Key takeaway

For Computer Vision Engineers or 3D artists tasked with generating realistic strand-based hair models from 2D imagery, HairLRM offers a robust solution. You should consider integrating this LRM-based approach to overcome the inherent ill-posedness of traditional methods, especially when dealing with complex topologies like ponytails or curls. This method significantly improves accuracy and robustness, ensuring your reconstructed hair geometries are high-fidelity and topologically correct.

Key insights

Integrating LRM geometric priors and a Dual Orientation AutoEncoder resolves 3D hair reconstruction's ill-posedness, achieving high-fidelity strands.

Principles

• Ill-posed 3D inference from 2D causes geometric failures.
• Strong geometric priors improve complex 3D reconstruction.
• Latent-space optimization resolves vector field singularities.

Method

HairLRM integrates LRM mesh priors as a structural anchor. It uses a Dual Orientation AutoEncoder to lift coarse geometry to strands, resolving vector field singularities via latent-space optimization and surface-guided refinement.

In practice

• Generate high-fidelity 3D hair strands from 2D images.
• Improve robustness in complex hair topology reconstruction.
• Enhance accuracy for global occlusion and local curls.

Topics

• HairLRM
• Strand-based Hair Modeling
• Large Reconstruction Models
• 3D Hair Reconstruction
• Dual Orientation AutoEncoder
• Computer Graphics

Best for: Research Scientist, AI Scientist, Computer Vision Engineer

Related on AIssential

• HairPort: In-context 3D-aware Hair Import and Transfer for Images — HairPort enables robust 3D-aware hairstyle transfer by separating hair removal, enforcing geometric consistency, and leveraging 3D reconstruction.
• Less Gaussians, Texture More: 4K Feed-Forward Textured Splatting — LGTM uses textured Gaussian primitives to enable 4K novel view synthesis with fewer primitives.
• S23DR 2026 Winning Solution — A flow-matching DiT conditioned on scene tokens effectively reconstructs 3D wireframes from sparse multi-modal inputs.
• Large-Scale High-Quality 3D Gaussian Head Reconstruction from Multi-View Captures — HeadsUp reconstructs high-quality 3D Gaussian heads from multi-view inputs using a UV-parameterized latent representation.
• ExMesh: EXplicit Mesh Reconstruction with Topology Adaptation — ExMesh directly optimizes explicit meshes by integrating differentiable optimization with adaptive topology and real-time UV maintenance.
• 👩‍🦰Deformable 3D Hair👩‍🦰 👉Xi’an Jiaotong University unveils a novel method... — A new method reconstructs decoupled 3D Gaussian head avatars from a single image, enabling dynamic hairstyle transfer.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.