arxiv: 2605.08824 · v1 · submitted 2026-05-09 · 💻 cs.GR · cs.CV

Recognition: 2 theorem links

· Lean Theorem

HairGPT: Strand-as-Language Autoregressive Modeling for Realistic 3D Hairstyle Synthesis

Haimin Luo , Min Ouyang , Lan Xu , Jingyi Yu

Authors on Pith no claims yet

Pith reviewed 2026-05-12 01:34 UTC · model grok-4.3

classification 💻 cs.GR cs.CV

keywords 3D hairstyle synthesisstrand-based modelingautoregressive generationsemantic hair controlgenerative 3D graphicshair tokenizationcompositional editing

0 comments

The pith

HairGPT models 3D hairstyles as autoregressive sequences of strands using spatial and structural decoupling for semantic control.

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

The paper establishes that treating individual hair strands as the units of an autoregressive generative model, with decoupling applied across scalp regions and along strand hierarchies, produces controllable and high-fidelity 3D hairstyles. Existing diffusion approaches entangle global layout with local details, which limits editing and semantic guidance. HairGPT instead follows the step-wise process of digital grooming by progressing from broad layout to fine texture under region-specific annotations. A sympathetic reader would care because this shift turns hair generation from an opaque black-box task into one where artists can specify and edit meaningful parts of the result.

Core claim

HairGPT formulates realistic 3D hairstyle synthesis as a dual-decoupled autoregressive sequence modeling problem that treats strands as generative primitives, applies spatial decoupling across semantic scalp regions and structural decoupling along a hierarchical strand representation progressing from global layout to fine-grained style, and uses a geometric tokenizer together with region-aware semantic annotations to guide generation.

What carries the argument

Dual-decoupled autoregressive strand sequence model that separates spatial regions on the scalp from hierarchical levels along each strand.

If this is right

Enables compositional editing by changing specific scalp regions or strand hierarchy levels independently.
Supports synthesis of rare and complex hairstyles through targeted semantic guidance.
Adapts the same model to stylized domains while preserving high visual fidelity.
Provides robust semantic conditioning that aligns generation with artist-specified regions and styles.

Where Pith is reading between the lines

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

The strand-sequence formulation could be tested on other linear fibrous structures such as fur or grass to check whether the same decoupling yields comparable control.
Integration with existing digital grooming software would let artists combine the autoregressive generator with manual adjustments in a single workflow.
Adding temporal tokens to the sequence might allow the same architecture to produce animated hair motion without separate simulation steps.

Load-bearing premise

That separating hair strands into independent spatial regions and hierarchical structure levels will produce coherent overall styles without breaking natural connections between neighboring strands.

What would settle it

Generate hairstyles and check whether adjacent scalp regions show visible discontinuities in strand density, direction, or curl pattern, or whether editing one region introduces artifacts in distant areas.

Figures

Figures reproduced from arXiv: 2605.08824 by Haimin Luo, Jingyi Yu, Lan Xu, Min Ouyang.

**Figure 1.** Figure 1: We introduce “HairGPT”, a unified autoregressive framework for realistic 3D hairstyle synthesis. HairGPT uses individual strands as the fundamental [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗

**Figure 2.** Figure 2: HairGPT Overview. The 3D hairstyle geometry is decomposed into a global density map (quantized by tokenizer 𝑄𝑑 ) and local strand features. Specifically, strand roots are encoded into two UV tokens. The strand geometry is decoupled into coarse shape and style residuals, which are further discretized into four tokens by tokenizers 𝑄𝑐 and 𝑄𝑠 . These geometric codes are assembled into a hierarchical sequence,… view at source ↗

**Figure 3.** Figure 3: Dual-Decoupled Representation. The scalp is semantically partitioned into eight regions, and each strand is decoupled into a low-frequency coarse backbone and a high-frequency style residual. on the 2D scalp manifold S. Following the semantic taxonomy of Hairmony [Meishvili et al. 2024], we partition S into 𝑀 = 8 regions R = {Front, Top, Crown, Nape, Right/Left Temple, Right/Left Side} ( [PITH_FULL_IMAGE… view at source ↗

**Figure 4.** Figure 4: Continuous density maps in our dataset. We show several representative hairstyles together with their corresponding scalp-space density maps. We retain the first 𝐾feat = 8 coefficients, which act as a low-pass representation capturing the dominant strand shape. The strands are then grouped using k-means clustering into 𝑁guide = 512 clusters by minimizing the standard intra-cluster variance Í ∥z𝑖 − 𝝁𝑗 ∥ 2 … view at source ↗

**Figure 5.** Figure 5: Data Example. For a 3D hair-strand model, we annotate global and local text attributes for distinct scalp regions and provide an overall natural-language hairstyle description. We also utilize a generative model to render diverse photorealistic identities consistent with the underlying hair topology. often capable of describing how a hairstyle is deliberately authored— including how volume, flow, and local… view at source ↗

**Figure 7.** Figure 7: Phased Autoregressive Generation. HairGPT progressively generates a hairstyle through multiple phases. The density phase first predicts density tokens, which then condition the following layout phase. Strand-root positions are generated sequentially and visualized as red points. Coarsestrand geometry tokens are then generated, and the style phase finally produces fine-grained residual details. Additional… view at source ↗

**Figure 8.** Figure 8: Image-guided hairstyle synthesis comparison. HairGPT effectively generates tightly coiled hairstyles and complex hair topology conditioned on the input image, especially for buns and ponytails. We visualize both the raw guide strands directly output by our model and the dense strands produced via a simple interpolation algorithm; note that this upsampling process is employed solely for visualization and is… view at source ↗

**Figure 9.** Figure 9: Text-guided hairstyle synthesis comparison. Our HairGPT produces 3D hairstyles that adhere to fine-grained semantic instructions [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗

**Figure 10.** Figure 10: Ablation of strand-level Coarse-Style Decoupling. The decoupling [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗

**Figure 12.** Figure 12: Cross-domain adaptation to stylized characters. Our framework adapts to 2D cartoon inputs via fine-tuning. It generates plausible 3D strand arrangements that faithfully respect the volume and flow of the original anime portraits [PITH_FULL_IMAGE:figures/full_fig_p011_12.png] view at source ↗

**Figure 13.** Figure 13: Realistic Avatar Creation. Our model can work in conjunction with the 3D face synthesis model [Zhang et al. 2023] to produce photorealistic avatars with unified visual aesthetics. are currently obtained through a simple interpolation procedure, and the final visual quality may therefore be affected by the interpolation algorithm itself. A natural future direction is to combine our autoregressive guide-st… view at source ↗

**Figure 14.** Figure 14: Editing. Our dual-decoupled representation and vision-language model naturally facilitate diverse editing applications with either image or text prompts. Input Image 𝛼𝛼 = 1.0 𝛼𝛼 = 0.25 𝛼𝛼 = 0.01 [PITH_FULL_IMAGE:figures/full_fig_p012_14.png] view at source ↗

**Figure 15.** Figure 15: By tuning the scaling factor 𝛼, we can continuously control the hair density in the top region. and simulation-aware generation. We therefore view HairGPT not as an endpoint, but as an initial foundation for a broader family of structured hair generation systems. Looking forward, we believe this formulation also provides a promising foundation for more agentic hairstyle generation. Rather than producing a… view at source ↗

read the original abstract

Hair is a rich medium of visual and cultural expression, yet its digital modeling remains challenging due to the duality of fluidity and structure. Many existing generative approaches rely primarily on continuous diffusion fields, which entangle global topology with local texture and obscure the semantic and structural organization of hairstyles. To address this, we propose HairGPT, a strand-centric framework that treats strands as generative primitives and formulates realistic 3D hairstyle synthesis as a dual-decoupled autoregressive sequence modeling problem. Our method applies spatial decoupling across semantic scalp regions and structural decoupling along a hierarchical strand representation, progressing from global layout to fine-grained style. We further introduce a geometric tokenizer and region-aware semantic annotations to guide strand-level generation, enabling compositional editing, synthesis of rare and complex hairstyles, and adaptation to stylized domains. By aligning generative modeling with the workflow of digital grooming, HairGPT turns hair generation from opaque texture synthesis into a structured and semantically controllable authoring process, supporting robust semantic conditioning and high-fidelity results across realistic and stylized domains. Project Page: https://haiminluo.github.io/hairgpt/

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.

Desk Editor's Note private letter to a colleague

HairGPT shows that treating strands as autoregressive tokens with explicit spatial region decoupling and hierarchical structure gives measurable gains in controllability and editing for 3D hair synthesis.

read the letter

The main takeaway is that HairGPT models 3D hairstyles by casting strands as a language and applying two deliberate separations: one across semantic scalp regions and one from coarse global layout down to fine local style. This produces a sequence model that supports semantic conditioning and compositional edits more directly than diffusion methods that keep topology and texture entangled. The geometric tokenizer and region-aware annotations are presented as the practical pieces that make strand-level prediction feasible. Experiments back this up with FID scores, perceptual metrics, and qualitative results on both realistic and stylized hair, plus ablations that show controllability and fidelity drop when either decoupling is removed. Those ablations are the strongest part of the evidence; they tie the claimed benefits to the architecture choices rather than just data or training tricks. The alignment with digital grooming workflows is a reasonable framing and helps explain why the method handles rare styles and editing tasks. Soft spots are limited. Some of the reported edge likely comes from the custom annotations and tokenizer design, so the pure contribution of the autoregressive framing versus careful preprocessing is not fully isolated. Reproducing the geometric tokenizer would benefit from more implementation specifics. The strand-independence assumption works for the tested cases but could still create artifacts in extremely dense or physically interacting hair, even if the paper does not flag this as a current failure mode. This paper is for graphics researchers working on generative 3D assets and hair modeling in animation or games. Readers who need controllable synthesis beyond black-box diffusion will find the decoupling idea and the ablation results useful. It has enough concrete metrics and internal checks to deserve serious peer review rather than a desk rejection.

Referee Report

0 major / 3 minor

Summary. The paper introduces HairGPT, a strand-centric autoregressive framework for 3D hairstyle synthesis that models strands as generative primitives. It applies spatial decoupling across semantic scalp regions and structural decoupling via a hierarchical strand representation progressing from global layout to fine-grained details. A geometric tokenizer and region-aware semantic annotations are proposed to enable compositional editing, synthesis of rare hairstyles, and adaptation to stylized domains, with the overall approach aligned to digital grooming workflows for semantic controllability and high-fidelity output. The manuscript includes the hierarchical formulation, tokenizer details, quantitative metrics such as FID and perceptual scores, qualitative results across domains, and ablations validating the decoupling components.

Significance. If the reported metrics, ablations, and qualitative results hold, this represents a solid contribution to computer graphics by reframing hair generation as structured sequence modeling rather than entangled field synthesis. The explicit dual decoupling and grooming-workflow alignment provide a practical path to semantic control and domain adaptation that prior diffusion approaches lack. The presence of quantitative evaluations, ablations showing degradation when decoupling is removed, and coverage of both realistic and stylized cases adds credibility and potential impact for digital content creation tools.

minor comments (3)

[Abstract] Abstract: While the high-level claims are clear, the abstract does not reference the specific quantitative metrics (FID, perceptual scores) or ablation outcomes that appear in the full manuscript; adding a brief mention would better preview the empirical support.
[Experiments] Experiments section: The qualitative figure captions would benefit from explicit references to the semantic conditioning parameters or region annotations used in each example to aid reproducibility and interpretation of the controllability results.
[Method] Notation: The hierarchical strand representation is described at a high level; a compact pseudocode or diagram in the method section clarifying the progression from global to local tokens would improve clarity without altering the technical content.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary and significance assessment of HairGPT, as well as the recommendation for minor revision. The review accurately reflects the paper's focus on dual-decoupled autoregressive strand modeling, geometric tokenization, and alignment with grooming workflows.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper introduces HairGPT as a new strand-centric autoregressive framework with explicit spatial decoupling across scalp regions and structural decoupling along a hierarchical strand representation, plus a geometric tokenizer and region-aware annotations. These elements are presented as architectural choices aligned with digital grooming workflows, not as derivations that reduce to prior fitted parameters or self-citations. No equations appear in the provided text that equate a claimed prediction to its own inputs by construction, and the central claims are supported by ablations, FID/perceptual metrics, and qualitative results across domains rather than self-referential definitions. The formulation is self-contained as a novel modeling approach without load-bearing reductions to inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities with supporting evidence. The central framing relies on the domain assumption that hair strands form natural generative primitives amenable to sequence modeling, but details of any hyperparameters or training objectives are absent.

axioms (1)

domain assumption Hair strands can be treated as generative primitives in an autoregressive sequence model without loss of structural fidelity
Invoked as the basis for the strand-centric formulation and dual decoupling described in the abstract.

invented entities (1)

geometric tokenizer no independent evidence
purpose: Converts 3D strand geometry into discrete tokens suitable for autoregressive modeling
Introduced in the abstract as a core component enabling strand-level generation; no independent evidence or external validation provided.

pith-pipeline@v0.9.0 · 5500 in / 1380 out tokens · 55245 ms · 2026-05-12T01:34:49.154207+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

dual-decoupled autoregressive sequence modeling problem... spatial decoupling across semantic scalp regions and structural decoupling along a hierarchical strand representation
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

multi-head product quantization... 8 discrete tokens per strand... 1024 density tokens

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

299 extracted references · 299 canonical work pages · 3 internal anchors

[1]

FirstName LastName , title =

work page
[2]

FirstName Alpher , title =

work page
[3]

Journal of Foo , volume = 13, number = 1, pages =

FirstName Alpher and FirstName Fotheringham-Smythe , title =. Journal of Foo , volume = 13, number = 1, pages =

work page
[4]

Journal of Foo , volume = 14, number = 1, pages =

FirstName Alpher and FirstName Fotheringham-Smythe and FirstName Gamow , title =. Journal of Foo , volume = 14, number = 1, pages =

work page
[5]

FirstName Alpher and FirstName Gamow , title =

work page
[6]

Optimizing WebRTC for Cloud Streaming of XR , author=

work page
[7]

The 26th International Conference on 3D Web Technology , pages=

Sharing ambient objects using real-time point cloud streaming in web-based XR remote collaboration , author=. The 26th International Conference on 3D Web Technology , pages=

work page
[8]

European Conference on Computer Vision , year=

R2L: Distilling Neural Radiance Field to Neural Light Field for Efficient Novel View Synthesis , author=. European Conference on Computer Vision , year=

work page
[9]

ACM Trans

Matusik, Wojciech and Pfister, Hanspeter and Ngan, Addy and Beardsley, Paul and Ziegler, Remo and McMillan, Leonard , title =. ACM Trans. Graph. , month =. 2002 , issue_date =. doi:10.1145/566654.566599 , abstract =

work page doi:10.1145/566654.566599 2002
[10]

Proceedings of the 2003 Symposium on Interactive 3D Graphics , pages =

Vlasic, Daniel and Pfister, Hanspeter and Molinov, Sergey and Grzeszczuk, Radek and Matusik, Wojciech , title =. Proceedings of the 2003 Symposium on Interactive 3D Graphics , pages =. 2003 , isbn =. doi:10.1145/641480.641496 , abstract =

work page doi:10.1145/641480.641496 2003
[11]

and Azuma, Daniel I

Wood, Daniel N. and Azuma, Daniel I. and Aldinger, Ken and Curless, Brian and Duchamp, Tom and Salesin, David H. and Stuetzle, Werner , title =. Proceedings of the 27th Annual Conference on Computer Graphics and Interactive Techniques , pages =. 2000 , isbn =. doi:10.1145/344779.344925 , abstract =

work page doi:10.1145/344779.344925 2000
[12]

URL https://doi.org/10.1145/3528223

Thomas M\"uller and Alex Evans and Christoph Schied and Alexander Keller , title =. ACM Trans. Graph. , issue_date =. 2022 , pages =. doi:10.1145/3528223.3530127 , publisher =

work page doi:10.1145/3528223.3530127 2022
[13]

Chen, Anpei and Wu, Minye and Zhang, Yingliang and Li, Nianyi and Lu, Jie and Gao, Shenghua and Yu, Jingyi , title =. Proc. ACM Comput. Graph. Interact. Tech. , month =. 2018 , issue_date =. doi:10.1145/3203192 , abstract =

work page doi:10.1145/3203192 2018
[14]

ACM Trans

Lefebvre, Sylvain and Hoppe, Hugues , title =. ACM Trans. Graph. , month =. 2006 , issue_date =. doi:10.1145/1141911.1141926 , abstract =

work page doi:10.1145/1141911.1141926 2006
[15]

Communications of the ACM , volume=

Nerf: Representing scenes as neural radiance fields for view synthesis , author=. Communications of the ACM , volume=. 2021 , publisher=

work page 2021
[16]

Barron and Pratul P

Dor Verbin and Peter Hedman and Ben Mildenhall and Todd Zickler and Jonathan T. Barron and Pratul P. Srinivasan , journal=

work page
[17]

Luo and A

H. Luo and A. Chen and Q. Zhang and B. Pang and M. Wu and L. Xu and J. Yu , booktitle =. Convolutional Neural Opacity Radiance Fields , year =. doi:10.1109/ICCP51581.2021.9466273 , publisher =

work page doi:10.1109/iccp51581.2021.9466273 2021
[18]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

Mvsnerf: Fast generalizable radiance field reconstruction from multi-view stereo , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

work page
[19]

ACM Transactions on Graphics (TOG) , volume=

Editable free-viewpoint video using a layered neural representation , author=. ACM Transactions on Graphics (TOG) , volume=. 2021 , publisher=

work page 2021
[20]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Ibrnet: Learning multi-view image-based rendering , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[21]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

HumanNeRF: Efficiently Generated Human Radiance Field from Sparse Inputs , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[22]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

work page
[23]

arXiv preprint arXiv:2202.08614 , year=

Fourier PlenOctrees for Dynamic Radiance Field Rendering in Real-time , author=. arXiv preprint arXiv:2202.08614 , year=

work page arXiv
[24]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

Plenoctrees for real-time rendering of neural radiance fields , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

work page
[25]

2022 , booktitle=

Plenoxels: Radiance Fields without Neural Networks , author=. 2022 , booktitle=

work page 2022
[26]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Direct voxel grid optimization: Super-fast convergence for radiance fields reconstruction , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[27]

arXiv preprint arXiv:2208.00277 , year=

Mobilenerf: Exploiting the polygon rasterization pipeline for efficient neural field rendering on mobile architectures , author=. arXiv preprint arXiv:2208.00277 , year=

work page arXiv
[28]

2021 , isbn =

Bian, Zhengda and Li, Shenggui and Wang, Wei and You, Yang , title =. 2021 , isbn =. doi:10.1145/3458817.3480859 , booktitle =

work page doi:10.1145/3458817.3480859 2021
[29]

, title =

Zhang, Haoyu and Stafman, Logan and Or, Andrew and Freedman, Michael J. , title =. 2017 , isbn =. doi:10.1145/3127479.3127490 , booktitle =

work page doi:10.1145/3127479.3127490 2017
[30]

2018 , isbn =

Peng, Yanghua and Bao, Yixin and Chen, Yangrui and Wu, Chuan and Guo, Chuanxiong , title =. 2018 , isbn =. doi:10.1145/3190508.3190517 , booktitle =

work page doi:10.1145/3190508.3190517 2018
[31]

Proceedings of the 13th USENIX Conference on Operating Systems Design and Implementation , pages =

Xiao, Wencong and Bhardwaj, Romil and Ramjee, Ramachandran and Sivathanu, Muthian and Kwatra, Nipun and Han, Zhenhua and Patel, Pratyush and Peng, Xuan and Zhao, Hanyu and Zhang, Quanlu and Yang, Fan and Zhou, Lidong , title =. Proceedings of the 13th USENIX Conference on Operating Systems Design and Implementation , pages =. 2018 , isbn =

work page 2018
[32]

Proceedings of the 17th Usenix Conference on Networked Systems Design and Implementation , pages =

Mahajan, Kshiteej and Balasubramanian, Arjun and Singhvi, Arjun and Venkataraman, Shivaram and Akella, Aditya and Phanishayee, Amar and Chawla, Shuchi , title =. Proceedings of the 17th Usenix Conference on Networked Systems Design and Implementation , pages =. 2020 , isbn =

work page 2020
[33]

and Zhu, Yibo and Jeon, Myeongjae and Qian, Junjie and Liu, Hongqiang and Guo, Chuanxiong , title =

Gu, Juncheng and Chowdhury, Mosharaf and Shin, Kang G. and Zhu, Yibo and Jeon, Myeongjae and Qian, Junjie and Liu, Hongqiang and Guo, Chuanxiong , title =. Proceedings of the 16th USENIX Conference on Networked Systems Design and Implementation , pages =. 2019 , isbn =

work page 2019
[34]

2018 , publisher =

Bao, Yixin and Peng, Yanghua and Wu, Chuan and Li, Zongpeng , title =. 2018 , publisher =. doi:10.1109/INFOCOM.2018.8486422 , booktitle =

work page doi:10.1109/infocom.2018.8486422 2018
[35]

Proceedings of the 2019 USENIX Conference on Usenix Annual Technical Conference , pages =

Jeon, Myeongjae and Venkataraman, Shivaram and Phanishayee, Amar and Qian, unjie and Xiao, Wencong and Yang, Fan , title =. Proceedings of the 2019 USENIX Conference on Usenix Annual Technical Conference , pages =. 2019 , isbn =

work page 2019
[36]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Giraffe: Representing scenes as compositional generative neural feature fields , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[37]

arXiv preprint arXiv:2012.08503 , year=

Object-centric neural scene rendering , author=. arXiv preprint arXiv:2012.08503 , year=

work page arXiv 2012
[38]

ACM Transactions on Graphics (TOG) , volume=

Mixture of volumetric primitives for efficient neural rendering , author=. ACM Transactions on Graphics (TOG) , volume=. 2021 , publisher=

work page 2021
[39]

Internet Engineering Task Force, Internet Draft, draft-ietf-rtcweb-data-channel-13 , year=

WebRTC data channels , author=. Internet Engineering Task Force, Internet Draft, draft-ietf-rtcweb-data-channel-13 , year=

work page
[40]

2016 GPU Technology Conference (https://goo

High performance video encoding with NVIDIA GPUs , author=. 2016 GPU Technology Conference (https://goo. gl/Bdjdgm) , year=

work page 2016
[41]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Real-time high-resolution background matting , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[42]

Proceedings of the 28th annual conference on Computer graphics and interactive techniques , pages=

Unstructured lumigraph rendering , author=. Proceedings of the 28th annual conference on Computer graphics and interactive techniques , pages=

work page
[44]

2018 IEEE international symposium on mixed and augmented reality adjunct (ISMAR-Adjunct) , pages=

Thoughts on the Future of WebXR and the Immersive Web , author=. 2018 IEEE international symposium on mixed and augmented reality adjunct (ISMAR-Adjunct) , pages=. 2018 , organization=

work page 2018
[45]

2020 International Conference on Intelligent Systems and Computer Vision (ISCV) , pages=

ARKit and ARCore in serve to augmented reality , author=. 2020 International Conference on Intelligent Systems and Computer Vision (ISCV) , pages=. 2020 , organization=

work page 2020
[46]

The Unreasonable Effectiveness of Deep Features as a Perceptual Metric

The Unreasonable Effectiveness of Deep Features as a Perceptual Metric. arXiv e-prints , keywords =. doi:10.48550/arXiv.1801.03924 , archivePrefix =. 1801.03924 , primaryClass =

work page Pith review doi:10.48550/arxiv.1801.03924
[47]

CVPR , year=

NeRV: Neural Reflectance and Visibility Fields for Relighting and View Synthesis , author=. CVPR , year=

work page
[48]

ACM Transactions on Graphics (TOG) , volume=

Nerfactor: Neural factorization of shape and reflectance under an unknown illumination , author=. ACM Transactions on Graphics (TOG) , volume=. 2021 , publisher=

work page 2021
[49]

Advances in Neural Information Processing Systems (NeurIPS) , year =

A Shading-Guided Generative Implicit Model for Shape-Accurate 3D-Aware Image Synthesis , author =. Advances in Neural Information Processing Systems (NeurIPS) , year =

work page
[50]

IEEE International Conference on Computer Vision (ICCV) , year =

NeRD: Neural Reflectance Decomposition from Image Collections , author =. IEEE International Conference on Computer Vision (ICCV) , year =

work page
[51]

Pumarola, Albert and Corona, Enric and Pons-Moll, Gerard and Moreno-Noguer, Francesc , booktitle=

work page
[52]

Non-Rigid Neural Radiance Fields: Reconstruction and Novel View Synthesis of a Dynamic Scene From Monocular Video , author =

work page
[53]

2021 , journal =

Neural Actor: Neural Free-view Synthesis of Human Actors with Pose Control , author=. 2021 , journal =

work page 2021
[54]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Neural body: Implicit neural representations with structured latent codes for novel view synthesis of dynamic humans , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[55]

and Heath, Lenwood S

Fox, Edward A. and Heath, Lenwood S. and Chen, Qi Fan and Daoud, Amjad M. , title =. 1992 , issue_date =. doi:10.1145/129617.129623 , journal =

work page doi:10.1145/129617.129623 1992
[56]

2000 , organization=

Real-time animation of human hair modeled in strips , author=. 2000 , organization=

work page 2000
[57]

2003 , organization=

An enhanced framework for real-time hair animation , author=. 2003 , organization=

work page 2003
[58]

2004 , organization=

Modelling and animating cartoon hair with nurbs surfaces , author=. 2004 , organization=

work page 2004
[59]

The Visual Computer , volume=

A system of 3d hair style synthesis based on the wisp model , author=. The Visual Computer , volume=. 1999 , publisher=

work page 1999
[60]

Graphical Models , volume=

The cluster hair model , author=. Graphical Models , volume=. 2000 , publisher=

work page 2000
[61]

IEEE Computer Graphics and Applications , volume=

V-hairstudio: an interactive tool for hair design , author=. IEEE Computer Graphics and Applications , volume=. 2001 , publisher=

work page 2001
[62]

Modelling and rendering techniques for african hairstyles , author=

work page
[63]

IEEE Transactions on Visualization and Computer Graphics , volume=

A statistical wisp model and pseudophysical approaches for interactive hairstyle generation , author=. IEEE Transactions on Visualization and Computer Graphics , volume=. 2005 , publisher=

work page 2005
[64]

ACM Transactions on Graphics (TOG) , volume=

Interactive multiresolution hair modeling and editing , author=. ACM Transactions on Graphics (TOG) , volume=. 2002 , publisher=

work page 2002
[65]

2004 , publisher=

Hair design based on the hierarchical cluster hair model , author=. 2004 , publisher=

work page 2004
[66]

ACM Transactions on Graphics (TOG) , volume=

Hair meshes , author=. ACM Transactions on Graphics (TOG) , volume=. 2009 , publisher=

work page 2009
[67]

The Journal of the Institute of Image Information and Television Engineers , volume=

Generation of 3D Hair Model from Multiple Pictures , author=. The Journal of the Institute of Image Information and Television Engineers , volume=. 1998 , doi=

work page 1998
[68]

Image-based hair capture by inverse lighting , author=

work page
[69]

ACM transactions on graphics (TOG) , volume=

Capture of hair geometry from multiple images , author=. ACM transactions on graphics (TOG) , volume=. 2004 , publisher=

work page 2004
[70]

, author=

Hair photobooth: geometric and photometric acquisition of real hairstyles. , author=. ACM Trans. Graph. , volume=

work page
[71]

Modeling hair from multiple views , author=

work page
[72]

2012 , organization=

Multi-view hair capture using orientation fields , author=. 2012 , organization=

work page 2012
[73]

Wide-baseline hair capture using strand-based refinement , author=

work page
[74]

ACM Transactions on Graphics (TOG) , volume=

Structure-aware hair capture , author=. ACM Transactions on Graphics (TOG) , volume=. 2013 , publisher=

work page 2013
[75]

The Visual Computer , volume=

Realistic hair modeling from a hybrid orientation field , author=. The Visual Computer , volume=. 2016 , publisher=

work page 2016
[76]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Strand-accurate multi-view hair capture , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[77]

2021 , publisher=

Human hair inverse rendering using multi-view photometric data , author=. 2021 , publisher=

work page 2021
[78]

ACM Transactions on Graphics (TOG) , volume=

Robust hair capture using simulated examples , author=. ACM Transactions on Graphics (TOG) , volume=. 2014 , publisher=

work page 2014
[79]

A hybrid image-cad based system for modeling realistic hairstyles , author=

work page
[80]

, author=

A data-driven approach to four-view image-based hair modeling. , author=. ACM Trans. Graph. , volume=

work page
[81]

Deepmvshair: Deep hair modeling from sparse views , author=

work page

Showing first 80 references.