Strand-level hair geometry reconstruction is a fundamental problem in virtual human modeling and the digitization of hairstyles. However, existing methods still suffer from a significant trade-off between accuracy and efficiency. Implicit neural representations can capture the global hair shape but often fail to preserve fine-grained strand details, while explicit optimization-based approaches achieve high-fidelity reconstructions at the cost of heavy computation and poor scalability. To address this issue, we propose EfficientMonoHair, a fast and accurate framework that combines the implicit neural network with multi-view geometric fusion for strand-level reconstruction from monocular video. Our method introduces a fusion-patch-based multi-view optimization that reduces the number of optimization iterations for point cloud direction, as well as a novel parallel hair-growing strategy that relaxes voxel occupancy constraints, allowing large-scale strand tracing to remain stable and robust even under inaccurate or noisy orientation fields. Extensive experiments on representative real-world hairstyles demonstrate that our method can robustly reconstruct high-fidelity strand geometries with accuracy. On synthetic benchmarks, our method achieves reconstruction quality comparable to state-of-the-art methods, while improving runtime efficiency by nearly an order of magnitude.
Figures & tables
Figure 1: Our method efficiently reconstructs strand-level hair geometry from monocular video input. The top part demonstrates that EfficientMonoHair accurately captures diverse hairstyles, including curly hair, faithfully preserving both the global silhouette and fine local curls. The bottom part illustrates the scalp-attached strand reconstruction, which can be seamlessly imported into existing graphics systems for photorealistic rendering and physically based hair simulation. The colorful strands are rendered in Blender Blender Foundation, 2025 and the hair simulation is conducted in Houdini SideFX, 2024 .
Figure 2: Our framework reconstructs strand-level hair geometry from a monocular video in three stages: (a) Outer Direction Optimization, where coarse hair geometry is reconstructed via Instant-NGP, followed by our FPMVO (detailed illustration in appendix) to obtain a stable outer point cloud with directions. (b) Inner Direction Inference, which leverages a View-Aware Transformer to infer invisible inner orientations from unidirectional strand maps; and (c) Hair Growing, where both inner and outer direction fields are fused into an Occupancy/Orientation Volume, from which our parallel strand growing (PHG) generates the final high-fidelity 3D hair model.
Figure 3: Visual comparison of the reconstruction of curly hair by multiple methods.
Occupation Accuracy (Voxel size in mm)
Orientation Accuracy (Voxel size in mm / deg)
Precision
Recall
F1
Precision
Recall
F1
Method
Speed
Time
2
3
4
2
3
4
2
3
4
2 / 20
3 / 30
4 / 40
2 / 20
3 / 30
4 / 40
2 / 20
3 / 30
4 / 40
MonoHair Wu et al., 2024 (P=5)
1 ×
136m
41.2
45.8
49.7
73.2
78.8
82.4
50.6
56.3
60.6
21.1
30.0
38.7
22.5
29.4
36.1
19.9
27.3
34.7
MonoHair Wu et al., 2024 (P=1)
1.2 ×
111m
37.3
41.7
45.4
63.0
69.7
74.2
44.8
50.4
54.7
18.1
26.8
35.2
15.2
21.3
26.9
15.3
22.2
28.7
Ours (full)
5.9 ×
23m
45.0
49.8
53.8
68.7
75.6
80.0
52.2
58.4
62.9
19.0
27.4
38.3
15.0
20.2
27.6
15.5
21.7
30.3
Ours (PHG only)
1.9 ×
72m
45.7
50.8
55.0
74.6
80.0
83.4
55.6
61.2
65.4
22.6
32.4
42.8
19.8
26.1
33.2
20.5
28.2
36.5
Table 1: Quantitative Results on Hair20K . Our method is competitive with MonoHair at patch size P=5 , while being 6× faster. MonoHair’s fast variant ( P=1 ) is still considerably slower than Ours, while singificanlty degrading the accuracy. Diffusion based methods like DiffLocks are much faster, but produce comparatively inaccurate results.
Figure 4: Quality vs. Speed using an average of the F1 scores for occupation and orientation on Hair20K.
Figure 5: Visual comparison of the reconstruction of big wavy, asymmetric, short and medium curly hair styles by multiple methods. At the third row, we failed in reproducing this short hair using the GaussianHaircut codebase, therefore we present the results of Figure 7 in their paper Zakharov et al., 2024 next to our reproduced result. For the last row, GaussianHaircut also failed in the reconstruction. This shows the low robustness of their method compared to ours.
Figure 6: Time Breakdown. We compare the timing of our method to MonoHair (MH) and GaussianHaircut (GH), split into the individual steps of the pipeline where applicable. Darkened regions on top indicate our step time. Results are averaged over seven sequences of the real captured data shown in this paper.
Reconstructing strand-level 3D hair from a single-view image is highly challenging, especially when preserving consistent and realistic attributes in unseen regions. Existing methods rely on limited frontal-view cues and small-scale/style-restricted synthetic data, often failing to produce satisfactory results in invisible regions. In this work, we propose a novel framework that leverages the strong 3D priors of video generation models to transform single-view hair reconstruction into a calibrated multi-view reconstruction task. To balance reconstruction quality and efficiency for the reformulated multi-view task, we further introduce a neural orientation extractor trained on sparse real-image annotations for better full-view orientation estimation. In addition, we design a two-stage strand-growing algorithm based on a hybrid implicit field to synthesize the 3D strand curves with fine-grained details at a relatively fast speed. Extensive experiments demonstrate that our method achieves state-of-the-art performance on single-view 3D hair strand reconstruction on a diverse range of hair portraits in both visible and invisible regions.
Leyang Jin, Yujian Zheng, Bingkui Tong +3
Mohamed bin Zayed University of Artificial Intelligence · SSE, The Chinese University of Hong Kong, Shenzhen · Pinscreen
The fundamental limitation of traditional strand-based modeling is not simply data scarcity, but the ill-posedness of inferring complex 3D fields from 2D imagery without structural constraints. This unconstrained regression leads to catastrophic failures in resolving both global occlusion (e.g., in ponytails) and local directionality (e.g., in curls), resulting in over-smoothed, plausible-but-incorrect geometries. To resolve this, we integrate the strong geometric priors of Large Reconstruction Models (LRMs) into the strand generation pipeline. Using the LRM mesh as a structural anchor, we employ a novel Dual Orientation AutoEncoder to lift coarse geometry into high-fidelity strands. By resolving vector field singularities through latent-space optimization and surface-guided refinement, our method effectively disentangles complex topological structures, setting a new benchmark for robustness and accuracy in hair reconstruction.
Yuefan Shen, Yican Dong, Xiufeng Huang +3
LIGHTSPEED, China · State Key Lab of CAD and CG, Zhejiang University, China · Hong Kong Baptist University, China +1
We propose a lightweight real-time method for reconstructing strand-based hair G-Buffers from severely undersampled rasterized inputs. Our pipeline first applies neural spatial reconstruction and temporal accumulation to recover hair coverage, i.e., fractional hair visibility within a pixel, and tangent. It then uses a tangent-guided reconstruction step to complete the position, which is subsequently used for physically based deferred hair shading. We evaluate our method across a diverse set of hairstyles, including straight, wavy, afro, and ponytail styles, under both static and dynamic scenarios. Our method achieves higher hair reconstruction quality than general industrial neural reconstruction solutions such as DLSS and FSR.
Chenghao Wu, Yuefan Shen, Tao Huang +3
University of Manchester, United Kingdom · LIGHTSPEED, China · LIGHTSPEED, USA