pith. sign in

arxiv: 2505.05376 · v3 · pith:74K3MFBCnew · submitted 2025-05-08 · 💻 cs.CV

GeomHair: Reconstruction of Hair Strands from Colorless 3D Scans

Rachmadio Noval Lazuardi , Artem Sevastopolsky , Egor Zakharov , Matthias Niessner , Vanessa Sklyarova This is my paper

Pith reviewed 2026-05-22 15:50 UTC · model grok-4.3

classification 💻 cs.CV
keywords hair strand reconstruction3D geometrycolorless scansdiffusion priororientation estimationdigital avatarsStrands400 dataset
0
0 comments X

The pith

Hair strands can be reconstructed accurately from colorless 3D scans alone by combining surface features, shading-based line detection, and a text-adapted diffusion prior.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to prove that individual hair strands can be recovered from raw geometry data without any color information. It achieves this by first locating sharp edges on the scan surface, then applying a neural line detector to shaded renderings to estimate orientations, and finally guiding a diffusion model that was pre-trained on synthetic hair data and tuned to each scan with a text prompt. A reader would care because existing approaches require color images or other cues, which are often unavailable in raw 3D captures used for avatars and animation. If the approach holds, strand-level detail becomes available directly from existing geometry scans and supports building large public datasets for further generative modeling.

Core claim

The central claim is that accurate strand reconstruction from colorless geometry is possible through multi-modal supervision signals: direct identification of sharp surface features on the scan, orientation estimation via a neural 2D line detector applied to shaded renderings of the scan, and a diffusion prior trained on synthetic hair scans that is refined with a noise schedule and adapted to each scan through a specific text prompt. This combination produces faithful strands for both simple and intricate hairstyles.

What carries the argument

Multi-modal hair orientation extraction that fuses sharp surface feature detection, neural line detection on shaded scan renderings, and a scan-specific text-adapted diffusion prior.

If this is right

  • Produces the Strands400 dataset containing strand reconstructions and surface geometry from 400 real subjects.
  • Supplies training data for image-to-strands and text-to-strands generative models.
  • Extends directly to artist-created mesh assets that require strand representations for simulation and rendering.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same pipeline could be tested on dynamic scans to extract time-varying strand motion.
  • Integrating the output strands with existing body meshes would allow end-to-end avatar pipelines that start from pure geometry.
  • The method's reliance on shading renderings suggests it may degrade under uniform lighting or low-resolution scans.

Load-bearing premise

The diffusion prior trained on synthetic hair scans generalizes to real-world colorless scans after adaptation via a scan-specific text prompt.

What would settle it

Reconstruction results on a set of real-world colorless scans where the output strands show large angular deviations from manually traced ground-truth orientations or visibly mismatch the input geometry's surface details.

Figures

Figures reproduced from arXiv: 2505.05376 by Artem Sevastopolsky, Egor Zakharov, Matthias Niessner, Rachmadio Noval Lazuardi, Vanessa Sklyarova.

Figure 1
Figure 1. Figure 1: We present GeomHair – a method for reconstructing complete hair strands representations from 3D scans that can be obtained from various sources, such as handheld 3D scanners, designer assets, and others. Our method extracts information about guiding sharp features directly from the scan geometry by employing a combination of 3D and 2D orientation detectors and fits strands with a diffusion prior conditione… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our GeomHair framework. Our method consists of two main stages: orientations extraction (left) and strands reconstruction (right). In the orientation extraction stage, we extract complementary orientation signals by combining 3D orientations from crest lines with 2D orientations obtained from TEED features applied to rendered shading of the scans. During reconstruction, we optimize a geometry t… view at source ↗
Figure 4
Figure 4. Figure 4: Results on 3D-designed assets. Here, we demonstrate the results of [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Ablation over the various components of the GeomHair pipeline. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of our method with state-of-the-art hair reconstruction methods across five different scenes. For Neural Haircut [Sklyarova et al [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 6
Figure 6. Figure 6: The distribution of the age, stratified by gender, (left) and of ethnicity (right), reported by the participants in the Strands400 dataset. Only the votes of [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Sample strands reconstructions from Strands400 dataset. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Sample scans from Strands400 dataset and the corresponding reconstructions. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The distribution of hair length in the Strands400 dataset. The captions are collected from the answers of a VQA model (LLaVA [Li et al [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The distribution of hair waviness in the Strands400 dataset. The captions are collected from the answers of a VQA model (LLaVA [Li et al [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
read the original abstract

We propose a novel method that reconstructs hair strands directly from colorless 3D scans by leveraging multi-modal hair orientation extraction. Hair strand reconstruction is a fundamental problem in computer vision and graphics, essential for high-fidelity digital avatar synthesis, animation, and AR/VR applications. However, accurately recovering hair strands from raw scan data remains challenging due to the complex and fine-grained structure of human hair, and none of the existing methods operate on colorless 3D geometry alone. To address this gap, our method directly identifies sharp surface features on the scan and estimates strand orientation using a neural 2D line detector applied to the renderings of scan shading. Additionally, we incorporate a diffusion prior trained on a diverse set of synthetic hair scans, refined with a noise schedule, and adapted to the reconstructed contents via a scan-specific text prompt. We demonstrate that this combination of supervision signals enables accurate reconstruction of both simple and intricate hairstyles from geometry alone. By enabling strand extraction from 3D scans, we compile Strands400, the largest publicly available dataset of hair strands with detailed surface geometry extracted from real-world data, comprising reconstructions from 400 subjects' scans. Strands400 enables training data-driven generative models for downstream tasks such as image-to-strands and text-to-strands. Moreover, our method applies to designer mesh assets, supporting a practical CG workflow where artists model hair as meshes and need strand-level representations for simulation and rendering. All code and data will be released for research purposes on https://seva100.github.io/GeomHair/.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes GeomHair, a pipeline to reconstruct individual hair strands from colorless 3D scans. It extracts orientation via sharp surface features on the scan geometry and a neural 2D line detector applied to renderings of scan shading, then fuses these signals with a diffusion prior trained on synthetic hair scans; the prior is refined via a noise schedule and adapted to each target scan through a scan-specific text prompt. The authors claim this multi-modal supervision enables accurate strand recovery for both simple and intricate hairstyles, release the Strands400 dataset of 400 real-scan reconstructions, and note applicability to designer mesh assets for CG workflows.

Significance. If the central reconstruction claim holds with quantitative validation, the work would meaningfully advance geometry-only hair strand extraction for digital avatars, animation, and AR/VR. The public release of Strands400 together with code constitutes a clear strength, as it supplies training data for downstream generative tasks such as image-to-strands and text-to-strands and supports practical mesh-to-strand conversion.

major comments (2)
  1. [Abstract] Abstract: the headline claim that 'this combination of supervision signals enables accurate reconstruction of both simple and intricate hairstyles from geometry alone' is presented without any quantitative metrics, error statistics, ablation tables, or comparisons against prior strand-reconstruction methods; this absence makes the central empirical assertion impossible to evaluate.
  2. [Abstract (diffusion prior description)] Diffusion-prior paragraph: the assertion that a diffusion model trained exclusively on synthetic hair scans, after noise-schedule refinement and scan-specific text-prompt adaptation, successfully transfers to real-world colorless 3D scans is load-bearing for the method yet unsupported by domain-gap measurements, cross-domain error metrics, or failure-case analysis.
minor comments (2)
  1. The description of the neural 2D line detector would benefit from explicit citation of the detector architecture and training data.
  2. Strands400 dataset statistics (hairstyle diversity, scan resolution distribution, reconstruction success rate) are missing and should be added.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment point by point below and are prepared to revise the paper accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the headline claim that 'this combination of supervision signals enables accurate reconstruction of both simple and intricate hairstyles from geometry alone' is presented without any quantitative metrics, error statistics, ablation tables, or comparisons against prior strand-reconstruction methods; this absence makes the central empirical assertion impossible to evaluate.

    Authors: We agree that the abstract, as a concise summary, does not include specific quantitative metrics. The full manuscript provides these details through error statistics, ablation studies, and comparisons with prior methods in the Experiments section. We will revise the abstract to incorporate key quantitative highlights supporting the claim. revision: yes

  2. Referee: [Abstract (diffusion prior description)] Diffusion-prior paragraph: the assertion that a diffusion model trained exclusively on synthetic hair scans, after noise-schedule refinement and scan-specific text-prompt adaptation, successfully transfers to real-world colorless 3D scans is load-bearing for the method yet unsupported by domain-gap measurements, cross-domain error metrics, or failure-case analysis.

    Authors: We acknowledge that explicit validation of the synthetic-to-real transfer strengthens the presentation. The manuscript demonstrates this transfer through results on real scans using the Strands400 dataset, along with the described adaptation techniques. We will add domain-gap measurements, cross-domain error metrics, and failure-case analysis to the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external synthetic prior

full rationale

The paper's core claim rests on combining surface feature detection, a neural 2D line detector on scan renderings, and a diffusion prior trained on separate synthetic hair scans then adapted via scan-specific text prompt. None of these reduce to the target scan geometry by construction or self-definition. The prior originates from external synthetic data rather than being fitted directly to real scans, and the adaptation step is presented as a transfer mechanism rather than a tautological fit. No load-bearing self-citations, uniqueness theorems, or ansatzes imported from the authors' prior work appear in the abstract or description. The downstream Strands400 dataset is an output of the method, not an input that forces the result. The derivation is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The method rests on the assumption that synthetic hair data plus a text prompt can adapt a diffusion model to real scans, and that 2D line detection on shaded renderings reliably recovers 3D strand orientations. No explicit free parameters or invented entities are named in the abstract.

axioms (2)
  • domain assumption Shaded renderings of the 3D scan contain sufficient 2D line features to recover 3D hair orientation via a neural detector.
    Invoked in the description of the neural 2D line detector applied to scan shading renderings.
  • domain assumption A diffusion prior trained on synthetic hair can be refined and adapted to real scans using a noise schedule and text prompt.
    Central to the diffusion prior component described in the abstract.

pith-pipeline@v0.9.0 · 5834 in / 1424 out tokens · 27671 ms · 2026-05-22T15:50:20.447159+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Wiley interdis- ciplinary reviews: computational statistics 2, 4 (2010), 433–459

    Principal component analysis. Wiley interdis- ciplinary reviews: computational statistics 2, 4 (2010), 433–459. Mohiuddin Ahmed, Raihan Seraj, and Syed Mohammed Shamsul Islam

    work page 2010

  2. [2]

    Electronics 9, 8 (2020),

    The k- means algorithm: A comprehensive survey and performance evaluation. Electronics 9, 8 (2020),

    work page 2020

  3. [3]

    IEEE Transactions on Pattern Analysis and Machine Intelligence PAMI-8 (1986), 679–698

    A Computational Approach to Edge Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence PAMI-8 (1986), 679–698. https://api. semanticscholar.org/CorpusID:13284142 Blender Online Community

    work page 1986

  4. [4]

    Advances in Neural Information Processing Systems (NeurIPS) (2022)

    Geo-Neus: Geometry-Consistent Neural Implicit Surfaces Learning for Multi-view Reconstruc- tion. Advances in Neural Information Processing Systems (NeurIPS) (2022). Guy Gafni, Justus Thies, Michael Zollhofer, and Matthias Nießner

    work page 2022

  5. [5]

    2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2020), 8645–8654

    Dynamic Neural Radiance Fields for Monocular 4D Facial Avatar Reconstruction. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2020), 8645–8654. Simon Giebenhain, Tobias Kirschstein, Markos Georgopoulos, Martin Rünz, Lourdes Agapito, and Matthias Nießner

    work page 2021

  6. [6]

    2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021), 18632–18643

    Neural Head Avatars from Monocular RGB Videos. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021), 18632–18643. Chengan He, Xin Sun, Zhixin Shu, Fujun Luan, Soren Pirk, Jorge Alejandro Amador Her- rera, Dominik Ludewig Michels, Tuanfeng Y. Wang, Meng Zhang, Holly Rushmeier, and Yi Zhou

    work page 2022

  7. [7]

    ArXiv abs/2407.19451 (2024)

    Perm: A Parametric Representation for Multi-Style 3D Hair Modeling. ArXiv abs/2407.19451 (2024). https://api.semanticscholar.org/CorpusID: 271534613 Liwen Hu, Chongyang Ma, Linjie Luo, and Hao Li

    work page arXiv 2024

  8. [8]

    ACM Transactions on Graphics (ToG) 34, 4 (2015), 1–9

    Single-view hair modeling using a hairstyle database. ACM Transactions on Graphics (ToG) 34, 4 (2015), 1–9. Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine

    work page 2015

  9. [9]

    In SIGGRAPH Asia 2022 Confer- ence Papers (Daegu, Republic of Korea) (SA ’22)

    Deep- MVSHair: Deep Hair Modeling from Sparse Views. In SIGGRAPH Asia 2022 Confer- ence Papers (Daegu, Republic of Korea) (SA ’22). Association for Computing Machin- ery, New York, NY, USA, Article 10, 8 pages. https://doi.org/10.1145/3550469.3555385 Chunyuan Li, Cliff Wong, Sheng Zhang, Naoto Usuyama, Haotian Liu, Jianwei Yang, Tristan Naumann, Hoifung P...

    work page doi:10.1145/3550469.3555385 2022

  10. [10]

    Advances in Neural Information Processing Systems 36 (2023), 28541–28564

    Llava-med: Training a large language-and-vision assistant for biomedicine in one day. Advances in Neural Information Processing Systems 36 (2023), 28541–28564. Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi

    work page 2023

  11. [11]

    ACM Transactions on Graphics, (Proc

    Learning a model of facial shape and expression from 4D scans. ACM Transactions on Graphics, (Proc. SIGGRAPH Asia) 36, 6 (2017), 194:1–194:17. https://doi.org/10.1145/3130800. 3130813 Haimin Luo, Min Ouyang, Zijun Zhao, Suyi Jiang, Longwen Zhang, Qixuan Zhang, Wei Yang, Lan Xu, and Jingyi Yu

    work page doi:10.1145/3130800 2017

  12. [12]

    arXiv preprint arXiv:2402.10483 (2024)

    GaussianHair: Hair Modeling and Rendering with Light-aware Gaussians. arXiv preprint arXiv:2402.10483 (2024). Morgan McGuire, Tiancheng Sun, Giljoo Nam, Carlos Aliaga, Christophe Hery, and Ravi Ramamoorthi

    work page arXiv 2024

  13. [13]

    2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2019), 155–164

    Strand-Accurate Multi-View Hair Capture. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2019), 155–164. Sylvain Paris, Héctor M. Briceño, and François X. Sillion

    work page 2019

  14. [14]

    ACM Transactions on Graphics (TOG) 23 (2004), 712 –

    Capture of hair geometry from multiple images. ACM Transactions on Graphics (TOG) 23 (2004), 712 –

    work page 2004

  15. [15]

    2021 IEEE/CVF International Conference on Computer Vision (ICCV) (2020), 5845–

    Nerfies: Deformable Neural Radiance Fields. 2021 IEEE/CVF International Conference on Computer Vision (ICCV) (2020), 5845–

    work page 2021

  16. [16]

    In Computer Vision – ECCV 2018: 15th European Conference

    Neural Strands: Learning Hair Geometry and Appearance from Multi-View Images. In Computer Vision – ECCV 2018: 15th European Conference . Yuefan Shen, Shunsuke Saito, Ziyan Wang, Olivier Maury, Chenglei Wu, Jessica Hodgins, Youyi Zheng, and Giljoo Nam

    work page 2018

  17. [17]

    42, 4, Article 75 (July 2023), 13 pages

    CT2Hair: High-Fidelity 3D Hair Modeling using Computed Tomography. 42, 4, Article 75 (July 2023), 13 pages. https://doi.org/10.1145/3592106 Vanessa Sklyarova, Jenya Chelishev, Andreea Dogaru, Igor Medvedev, Victor Lempit- sky, and Egor Zakharov. 2023a. Neural Haircut: Prior-Guided Strand-Based Hair Reconstruction. In Proceedings of IEEE International Conf...

    work page doi:10.1145/3592106 2023

  18. [18]

    Journal of machine learning research 9, 11 (2008)

    Visualizing data using t-SNE. Journal of machine learning research 9, 11 (2008). Ziyan Wang, Giljoo Nam, Tuur Stuyck, Stephen Lombardi, Chen Cao, Jason M. Saragih, Michael Zollhoefer, Jessica K. Hodgins, and Christoph Lassner

    work page 2008

  19. [19]

    ArXiv abs/2212.00613 (2022)

    NeuWigs: A Neural Dynamic Model for Volumetric Hair Capture and Animation. ArXiv abs/2212.00613 (2022). Ziyan Wang, Giljoo Nam, Tuur Stuyck, Stephen Lombardi, Michael Zollhoefer, Jessica K. Hodgins, and Christoph Lassner

    work page arXiv 2022

  20. [20]

    2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021), 6133–6144

    HVH: Learning a Hybrid Neural Volumetric Representation for Dynamic Hair Performance Capture. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021), 6133–6144. Keyu Wu, Lingchen Yang, Zhiyi Kuang, Yao Feng, Xutao Han, Yuefan Shen, Hongbo Fu, Kun Zhou, and Youyi Zheng

    work page 2022

  21. [21]

    In 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    MonoHair: High-Fidelity Hair Modeling from a Monocular Video. In 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Keyu Wu, Yifan Ye, Lingchen Yang, Hongbo Fu, Kun Zhou, and Youyi Zheng

    work page 2023

  22. [22]

    2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2022), 1516–1525

    NeuralHDHair: Automatic High-fidelity Hair Modeling from a Single Image Using Implicit Neural Representations. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2022), 1516–1525. https://api.semanticscholar.org/ CorpusID:248572051 Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M Alvarez, and Ping Luo

    work page 2022

  23. [23]

    Advances in neural information processing systems 34 (2021), 12077– 12090

    SegFormer: Simple and efficient design for semantic segmentation with transformers. Advances in neural information processing systems 34 (2021), 12077– 12090. Shin Yoshizawa, Alexander Belyaev, and Hans-Peter Seidel

    work page 2021

  24. [24]

    In Proceedings of the 2005 ACM symposium on Solid and physical modeling

    Fast and robust detection of crest lines on meshes. In Proceedings of the 2005 ACM symposium on Solid and physical modeling . 227–232. Cem Yuksel, Scott Schaefer, and John Keyser

    work page 2005

  25. [25]

    InACM SIGGRAPH Asia 2009 Papers (Yokohama, Japan) (SIGGRAPH Asia ’09)

    Hair meshes. InACM SIGGRAPH Asia 2009 Papers (Yokohama, Japan) (SIGGRAPH Asia ’09). Association for Computing Machinery, New York, NY, USA, Article 166, 7 pages. https://doi.org/10.1145/ 1661412.1618512 Egor Zakharov, Vanessa Sklyarova, Michael Black, Giljoo Nam, Justus Thies, and Otmar Hilliges

    work page arXiv 2009

  26. [26]

    In Computer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part XVI (Milan, Italy)

    Human Hair Reconstruction with Strand-Aligned 3D Gaussians. In Computer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part XVI (Milan, Italy). Springer-Verlag, Berlin, Heidelberg, 409–425. https://doi.org/10.1007/978-3-031-72640-8_23 Yufeng Zheng, Victoria Fernández Abrevaya, Xu Chen, Marcel C. Buhl...

    work page doi:10.1007/978-3-031-72640-8_23 2024

  27. [27]

    2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021), 13535–13545

    I M Avatar: Implicit Morphable Head Avatars from Videos. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021), 13535–13545. Yujian Zheng, Zirong Jin, Moran Li, Haibin Huang, Chongyang Ma, Shuguang Cui, and Xiaoguang Han

    work page 2022

  28. [28]

    2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2023), 12726–12735

    HairStep: Transfer Synthetic to Real Using Strand and Depth Maps for Single-View 3D Hair Modeling. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2023), 12726–12735. https: //api.semanticscholar.org/CorpusID:257365928 Yufeng Zheng, Yifan Wang, Gordon Wetzstein, Michael J. Black, and Otmar Hilliges

    work page 2023

  29. [29]

    ArXiv abs/2212.08377 (2022)

    PointAvatar: Deformable Point-based Head Avatars from Videos. ArXiv abs/2212.08377 (2022). Junsheng Zhou, Baorui Ma, Shujuan Li, Yu-Shen Liu, and Zhizhong Han. 2023b. Learn- ing a More Continuous Zero Level Set in Unsigned Distance Fields through Level Set Projection. In Proceedings of the IEEE/CVF international conference on computer vision. Yuxiao Zhou,...

    work page arXiv 2022

  30. [30]

    ACM Trans

    GroomCap: High-Fidelity Prior- Free Hair Capture. ACM Trans. Graph. 43, 6, Article 254 (Nov. 2024), 15 pages. https://doi.org/10.1145/3687768 Yi Zhou, Liwen Hu, Jun Xing, Weikai Chen, Han-Wei Kung, Xin Tong, and Hao Li

    work page doi:10.1145/3687768 2024

  31. [31]

    InEuropean Conference on Computer Vision

    Single-View Hair Reconstruction using Convolutional Neural Networks. InEuropean Conference on Computer Vision . https://api.semanticscholar.org/CorpusID:49666680 A STRANDS400 DATASET Legal notice. All participants in our dataset signed an agreement form compliant with GDPR. Please note that GDPR compliance includes the right for every participant to reque...

    work page 2023

  32. [32]

    Capture setup

    Each second sample has been shown in the plots to provide more space for the captions. Capture setup. Our dataset consists of two parts – the subset of the latest version of NPHM that contains 383 scans, and separately collected 17 scans with a setup similar to the NPHM setup. The latter setup consists of two handheld Artec Eva scanners, rotating over a 3...

    work page 2023

  33. [33]

    Only the votes of the participants who willingly disclosed that information were taken into account

    The distribution of the age, stratified by gender, (left) and of ethnicity (right), reported by the participants in the Strands400 dataset. Only the votes of the participants who willingly disclosed that information were taken into account. (3) Does the depicted hairstyle longer than the shoulders or shorter than the shoulder? (4) Does the depicted hairst...

    work page 2022

  34. [34]

    The captions are collected from the answers of a VQA model (LLaVA [Li et al

    The distribution of hair length in the Strands400 dataset. The captions are collected from the answers of a VQA model (LLaVA [Li et al. 2023]) after showing the rendered shading of the frontal and back views of the 3D scans in Strands400. The locations correspond to the t-SNE [Van der Maaten and Hinton 2008] over BLIP embeddings [Li et al. 2022] of the LL...

    work page 2023

  35. [35]

    The captions are collected from the answers of a VQA model (LLaVA [Li et al

    The distribution of hair waviness in the Strands400 dataset. The captions are collected from the answers of a VQA model (LLaVA [Li et al . 2023]) after showing the rendered shading of the frontal and back views of the 3D scans in Strands400. The locations correspond to the t-SNE [Van der Maaten and Hinton 2008] over BLIP embeddings [Li et al. 2022] of the...

    work page 2023