pith. sign in

arxiv: 2606.02573 · v1 · pith:QIMCGRX3new · submitted 2026-06-01 · 💻 cs.CV

HumanNOVA: Photorealistic, Universal and Rapid 3D Human Avatar Modeling from a Single Image

Hezhen Hu , Wangbo Zhao , Lanqing Guo , Hanwen Jiang , Jonathan C. Liu , Zhiwen Fan , Kai Wang , Zhangyang Wang
show 1 more author
Georgios Pavlakos
This is my paper

Pith reviewed 2026-06-28 15:12 UTC · model grok-4.3

classification 💻 cs.CV
keywords 3D human avatar modelingsingle-image reconstructiontriplane representationfeed-forward networkSMPL conditioningphotorealistic 3D generationdata scaling pipeline
0
0 comments X

The pith

A feed-forward model encodes a single RGB image and SMPL mesh into tokens, fuses them via cross-attention, and outputs a triplane 3D human avatar representation.

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

The paper establishes that scaling 3D human training data to 100k assets through animation of rigged models with daily poses and fitting of multi-camera captures enables a feed-forward network to produce photorealistic avatars from one image. The architecture encodes the image and a simplified mesh into compact tokens, then uses cross-attention to condition a triplane representation that supports inference in under one second without any test-time optimization. If correct, this removes the need for per-instance fitting or optimization loops that currently limit single-image avatar creation to slow, specialized pipelines. A sympathetic reader would care because it promises universal applicability across varied body types, clothing, and poses while remaining fast enough for interactive use.

Core claim

HumanNOVA is a feed-forward token-conditioned avatar modeling framework: an input image and estimated SMPL mesh are encoded into tokens that are fused through cross-attention to construct a triplane-based 3D avatar representation, trained on data scaled to 100k assets via two generation strategies and shown to outperform prior methods on multiple benchmarks with robustness to diverse inputs.

What carries the argument

token-conditioned triplane framework: image and SMPL tokens are fused by cross-attention to build the triplane representation that carries the 3D geometry and appearance

If this is right

  • Inference completes in less than one second on standard hardware
  • No per-instance optimization or fine-tuning is required at test time
  • Quantitative and qualitative results exceed prior single-image methods across multiple benchmarks
  • Performance remains stable under varied input conditions such as different poses, clothing, and backgrounds

Where Pith is reading between the lines

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

  • The same token-fusion approach could be applied to video sequences to produce temporally consistent dynamic avatars
  • The data-scaling pipeline itself might transfer to other domains that lack large 3D ground truth, such as animal or object modeling
  • If the triplane representation proves editable, downstream applications could allow users to modify generated avatars through simple mesh or image edits

Load-bearing premise

The two data-generation strategies produce training data whose quantity and diversity suffice for generalization to real single-image inputs without overfitting to synthetic artifacts.

What would settle it

Test the released model on a collection of real-world single photographs of humans in poses, clothing, and lighting absent from both the rigged-asset animations and the fitted multi-camera captures, then measure whether visual fidelity and geometric accuracy fall below the reported benchmark levels.

Figures

Figures reproduced from arXiv: 2606.02573 by Georgios Pavlakos, Hanwen Jiang, Hezhen Hu, Jonathan C. Liu, Kai Wang, Lanqing Guo, Wangbo Zhao, Zhangyang Wang, Zhiwen Fan.

Figure 1
Figure 1. Figure 1: Photorealistic, universal and rapid 3D human avatar modeling from a single image by the proposed approach, HumanNOVA. It benefits from both our generated large-scale data and feed-forward model design. Our data generation pipeline expands training data by 20 times (top-left for visualization). With this data, HumanNOVA achieves superior performance while maintaining rapid inference among existing methods (… view at source ↗
Figure 2
Figure 2. Figure 2: HumanNOVA network architecture. Given a real-world input image, we first estimate its corresponding simplified human mesh. Image and mesh are fed into the multi-modal encoder to extract features which are utilized as the condition for the following mapping network. After that, a Transformer-based mapping network directly maps the features to the 3D triplane representation. From this triplane representation… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison with state-of-the-art methods, including input from benchmarks (top) and in-the-wild images (bottom). The reconstructed human by our HumanNOVA method shows superior structure and appearance. (Best viewed in color.) often yield lower quality results when given a side-view in￾put, as they do not generalize well to such less common viewpoints. In contrast, our approach yields superior p… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative evaluation of our approach with in-the￾wild images as input. We also show some typical failure cases (bottom), e.g., inferring the plausible back texture of challenging clothes like dresses and overalls. (Best viewed in color.) the best performance across all metrics. Notably, on the side-view inputs, where occlusion affects performance, Hu￾manNOVA outperforms the human-specific method SiTH by … view at source ↗
Figure 5
Figure 5. Figure 5: Visual results on the effectiveness of the SMPL prior. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visual results on the robustness of HumanNOVA under [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of our training data. (Best viewed in color.) The first three rows correspond to real-world generated data, while the remaining rows are generated synthetic data [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Visualization of ablation studies on generated data type. (Best viewed in color.) Removing the generated data (multi￾cam or assets) negatively affects the model performance. GT HumanNOVA wo gen-data (multi-cam) wo gen-data (assets) GT HumanNOVA wo mesh small triplane size [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of ablation studies on model design. (Best viewed in color.) Excluding the mesh prior or reducing the triplane resolution reduces the quality of the output. j ∗ +r, :]. This strategy encourages sampling of patches rich in foreground content while preserving diversity. Evaluation. We utilize 50 assets from each dataset (Cus￾tomHuman, Thuman2, and 2K2K) for evaluation. Both Cus￾tomHuman and THu… view at source ↗
read the original abstract

In this paper, we present HumanNOVA, a photorealistic, universal, and rapid model for generating 3D human avatars from a single RGB image. Achieving both photorealism and generalization is challenging due to the scarcity of diverse, high-quality 3D human data. To address this, we build a scalable data generation pipeline that follows two strategies. The first one is to leverage existing rigged assets and animate them with extensive poses from daily life. The second strategy is to utilize existing multi-camera captures of humans and employ fitting to generate more diverse views for training. These two strategies enable us to scale up to 100k assets, significantly enhancing both the quantity and the diversity of data for robust model training. In terms of the architecture, HumanNOVA adopts a feed-forward, token-conditioned avatar modeling framework that allows fast inference in less than one second and requires no test-time optimization. Given an input image and an estimated simplified human mesh (SMPL) without detailed geometry or appearance, the model first encodes both inputs into compact token representations. These tokens then act as conditioning signals and are fused through cross-attention to construct a triplane-based 3D avatar representation. Extensive experiments on multiple benchmarks demonstrate the superiority of our approach, both quantitatively and qualitatively, as well as its robustness under diverse input image conditions. Project page at https://HumanNOVA.github.io .

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 presents HumanNOVA, a feed-forward token-conditioned triplane framework for photorealistic 3D human avatar modeling from a single RGB image. It introduces a scalable data-generation pipeline that combines animation of rigged assets with daily-life poses and fitting of multi-camera captures to produce 100k training assets, enabling inference in under one second with no test-time optimization. The manuscript asserts quantitative and qualitative superiority over prior methods on multiple benchmarks together with robustness across diverse input conditions.

Significance. If the performance and generalization claims are substantiated with rigorous evidence, the work would advance single-image 3D human reconstruction by demonstrating that a purely feed-forward architecture can achieve photorealism and universality when trained on a large-scale synthetic-plus-fitted dataset, potentially reducing reliance on per-instance optimization in graphics and vision applications.

major comments (2)
  1. [§3] Data-generation pipeline (abstract and §3): the claim that the two strategies produce training data whose quantity and diversity suffice for generalization to real single-image inputs is unsupported by any reported statistics on appearance/lighting/clothing coverage, domain-gap metrics, or validation against real photographs. The architecture description contains no explicit compensation mechanism, making this assumption load-bearing for the universality and photorealism assertions.
  2. [§5] Experiments section (abstract and §5): the abstract states that 'extensive experiments on multiple benchmarks demonstrate the superiority of our approach, both quantitatively and qualitatively,' yet supplies no numerical metrics, baseline comparisons, error bars, ablation tables, or statistical tests. This directly undermines the central performance claim.
minor comments (2)
  1. [Abstract] The term 'universal' is used repeatedly but never given an operational definition (e.g., range of body shapes, clothing types, lighting conditions, or ethnicities covered).
  2. [§4] The SMPL mesh is described as 'simplified... without detailed geometry or appearance,' yet the precise level of simplification and how it is obtained from the input image are not specified.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and propose revisions where the manuscript can be strengthened without misrepresenting our contributions.

read point-by-point responses

  1. Referee: [§3] Data-generation pipeline (abstract and §3): the claim that the two strategies produce training data whose quantity and diversity suffice for generalization to real single-image inputs is unsupported by any reported statistics on appearance/lighting/clothing coverage, domain-gap metrics, or validation against real photographs. The architecture description contains no explicit compensation mechanism, making this assumption load-bearing for the universality and photorealism assertions.

    Authors: We agree that explicit coverage statistics and domain-gap analysis would strengthen the support for the generalization claims. The two strategies (rigged-asset animation with daily-life poses and multi-camera fitting) are intended to increase diversity, but the manuscript does not quantify appearance, lighting, or clothing distributions. In revision we will add a table and short analysis in §3 reporting these statistics across the 100k assets and discuss how the token-conditioned triplane architecture provides implicit robustness via SMPL and image conditioning. We will also note any limitations in domain-gap validation. revision: yes

  2. Referee: [§5] Experiments section (abstract and §5): the abstract states that 'extensive experiments on multiple benchmarks demonstrate the superiority of our approach, both quantitatively and qualitatively,' yet supplies no numerical metrics, baseline comparisons, error bars, ablation tables, or statistical tests. This directly undermines the central performance claim.

    Authors: The abstract summarizes the results due to length constraints, but §5 presents quantitative metrics (PSNR, SSIM, LPIPS), baseline comparisons, and ablations on multiple benchmarks. If these elements were not sufficiently prominent or lacked error bars/statistical tests, we will revise §5 to include them explicitly and update the abstract with key numerical values where space permits. This addresses the concern directly. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical pipeline evaluated on external benchmarks.

full rationale

The manuscript presents a data-generation pipeline (rigged-asset animation plus multi-camera fitting) to produce 100k training assets, followed by a feed-forward token-conditioned triplane network. No equations, derivations, or first-principles results are supplied that reduce to the inputs by construction. Quantitative claims reference external benchmarks rather than quantities defined from the training distribution itself. No self-citation load-bearing steps, uniqueness theorems, or ansatz smuggling appear in the text. The central performance assertions therefore remain independent of the training data statistics.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies insufficient technical detail to enumerate specific free parameters, axioms, or invented entities; the approach relies on the pre-existing SMPL model and triplane representation from prior literature.

pith-pipeline@v0.9.1-grok · 5813 in / 1113 out tokens · 37721 ms · 2026-06-28T15:12:39.173969+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

82 extracted references

  1. [1]

    Single-image 3d hu- man digitization with shape-guided diffusion

    Badour AlBahar, Shunsuke Saito, Hung-Yu Tseng, Changil Kim, Johannes Kopf, and Jia-Bin Huang. Single-image 3d hu- man digitization with shape-guided diffusion. InSIGGRAPH Asia, pages 1–11, 2023. 2, 5

    2023

  2. [2]

    Video based reconstruction of 3d people models

    Thiemo Alldieck, Marcus Magnor, Weipeng Xu, Christian Theobalt, and Gerard Pons-Moll. Video based reconstruction of 3d people models. InCVPR, pages 8387–8397, 2018. 2

    2018

  3. [3]

    Learning to recon- struct people in clothing from a single rgb camera

    Thiemo Alldieck, Marcus Magnor, Bharat Lal Bhatnagar, Christian Theobalt, and Gerard Pons-Moll. Learning to recon- struct people in clothing from a single rgb camera. InCVPR, pages 1175–1186, 2019

    2019

  4. [4]

    Tex2shape: Detailed full human body geom- etry from a single image

    Thiemo Alldieck, Gerard Pons-Moll, Christian Theobalt, and Marcus Magnor. Tex2shape: Detailed full human body geom- etry from a single image. InICCV, pages 2293–2303, 2019. 2

    2019

  5. [5]

    Multi-garment net: Learning to dress 3d people from images

    Bharat Lal Bhatnagar, Garvita Tiwari, Christian Theobalt, and Gerard Pons-Moll. Multi-garment net: Learning to dress 3d people from images. InICCV, pages 5420–5430, 2019. 2

    2019

  6. [6]

    Sf3d: Stable fast 3d mesh reconstruction with uv- unwrapping and illumination disentanglement

    Mark Boss, Zixuan Huang, Aaryaman Vasishta, and Varun Jampani. Sf3d: Stable fast 3d mesh reconstruction with uv- unwrapping and illumination disentanglement. InCVPR, pages 1–12, 2025. 3, 6, 8

    2025

  7. [7]

    Generalizable human gaussians from single-view image

    Jinnan Chen, Chen Li, Jianfeng Zhang, Lingting Zhu, Buzhen Huang, Hanlin Chen, and Gim Hee Lee. Generalizable human gaussians from single-view image. InICLR, pages 1–16, 2025. 2

    2025

  8. [8]

    Dna-rendering: A diverse neural actor repository for high-fidelity human-centric rendering

    Wei Cheng, Ruixiang Chen, Siming Fan, Wanqi Yin, Keyu Chen, Zhongang Cai, Jingbo Wang, Yang Gao, Zhengming Yu, Zhengyu Lin, et al. Dna-rendering: A diverse neural actor repository for high-fidelity human-centric rendering. InICCV, pages 19982–19993, 2023. 4

    2023

  9. [9]

    Gpavatar: Generalizable and precise head avatar from image (s)

    Xuangeng Chu, Yu Li, Ailing Zeng, Tianyu Yang, Lijian Lin, Yunfei Liu, and Tatsuya Harada. Gpavatar: Generalizable and precise head avatar from image (s). InICLR, pages 1–20,

  10. [10]

    i just embodied you

    Alicia G Cork, Anca Salagean, Laura GE Smith, David A El- lis, Adam Joinson, and Danaë Stanton Fraser. “i just embodied you”: Psychological ownership of personalized photorealistic avatars. InCHI, pages 1–7, 2024. 2

    2024

  11. [11]

    Struc- tured 3d features for reconstructing controllable avatars

    Enric Corona, Mihai Zanfir, Thiemo Alldieck, Eduard Gabriel Bazavan, Andrei Zanfir, and Cristian Sminchisescu. Struc- tured 3d features for reconstructing controllable avatars. In CVPR, pages 16954–16964, 2023. 2

    2023

  12. [12]

    Objaverse: A universe of annotated 3d objects

    Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Ehsani, Aniruddha Kembhavi, and Ali Farhadi. Objaverse: A universe of annotated 3d objects. InCVPR, pages 13142–13153, 2023. 2

    2023

  13. [13]

    Moga: 3d generative avatar prior for monocular gaussian avatar reconstruction

    Zijian Dong, Longteng Duan, Jie Song, Michael J Black, and Andreas Geiger. Moga: 3d generative avatar prior for monocular gaussian avatar reconstruction. InICCV, pages 13304–13314, 2025. 2

    2025

  14. [14]

    Sigmoid- weighted linear units for neural network function approx- imation in reinforcement learning.Neural networks, 107: 3–11, 2018

    Stefan Elfwing, Eiji Uchibe, and Kenji Doya. Sigmoid- weighted linear units for neural network function approx- imation in reinforcement learning.Neural networks, 107: 3–11, 2018. 13

    2018

  15. [15]

    Learning disentangled avatars with hybrid 3d representations.arXiv, pages 1–16, 2023

    Yao Feng, Weiyang Liu, Timo Bolkart, Jinlong Yang, Marc Pollefeys, and Michael J Black. Learning disentangled avatars with hybrid 3d representations.arXiv, pages 1–16, 2023. 2

    2023

  16. [16]

    Humans in 4d: Recon- structing and tracking humans with transformers

    Shubham Goel, Georgios Pavlakos, Jathushan Rajasegaran, Angjoo Kanazawa, and Jitendra Malik. Humans in 4d: Recon- structing and tracking humans with transformers. InICCV, pages 14783–14794, 2023. 2, 3

    2023

  17. [17]

    High-fidelity 3d human digitization from single 2k resolution images

    Sang-Hun Han, Min-Gyu Park, Ju Hong Yoon, Ju-Mi Kang, Young-Jae Park, and Hae-Gon Jeon. High-fidelity 3d human digitization from single 2k resolution images. InCVPR, pages 12869–12879, 2023. 2, 5, 6

    2023

  18. [18]

    Arch++: Animation-ready clothed human re- construction revisited

    Tong He, Yuanlu Xu, Shunsuke Saito, Stefano Soatto, and Tony Tung. Arch++: Animation-ready clothed human re- construction revisited. InICCV, pages 11046–11056, 2021. 2

    2021

  19. [19]

    Magicman: Generative novel view synthesis of humans with 3d-aware diffusion and iterative refinement

    Xu He, Zhiyong Wu, Xiaoyu Li, Di Kang, Chaopeng Zhang, Jiangnan Ye, Liyang Chen, Xiangjun Gao, Han Zhang, and Haolin Zhuang. Magicman: Generative novel view synthesis of humans with 3d-aware diffusion and iterative refinement. InAAAI, pages 3437–3445, 2025. 5

    2025

  20. [20]

    Sith: Single-view textured human reconstruction with image-conditioned diffusion

    I Ho, Jie Song, and Otmar Hilliges. Sith: Single-view textured human reconstruction with image-conditioned diffusion. In CVPR, pages 538–549, 2024. 1, 2, 6, 8

    2024

  21. [21]

    Lrm: Large reconstruction model for single image to 3d

    Yicong Hong, Kai Zhang, Jiuxiang Gu, Sai Bi, Yang Zhou, Difan Liu, Feng Liu, Kalyan Sunkavalli, Trung Bui, and Hao Tan. Lrm: Large reconstruction model for single image to 3d. InICLR, pages 1–25, 2024. 2, 3

    2024

  22. [22]

    Sherf: Generalizable human nerf from a single image

    Shoukang Hu, Fangzhou Hong, Liang Pan, Haiyi Mei, Lei Yang, and Ziwei Liu. Sherf: Generalizable human nerf from a single image. InICCV, pages 9352–9364, 2023. 12

    2023

  23. [23]

    Tech: Text- guided reconstruction of lifelike clothed humans

    Yangyi Huang, Hongwei Yi, Yuliang Xiu, Tingting Liao, Jiaxiang Tang, Deng Cai, and Justus Thies. Tech: Text- guided reconstruction of lifelike clothed humans. In3DV, pages 1531–1542, 2024. 2

    2024

  24. [24]

    Pointinfinity: Resolution-invariant point diffusion models

    Zixuan Huang, Justin Johnson, Shoubhik Debnath, James M Rehg, and Chao-Yuan Wu. Pointinfinity: Resolution-invariant point diffusion models. InCVPR, pages 10050–10060, 2024. 3

    2024

  25. [25]

    Bcnet: Learning body and cloth shape from a single image

    Boyi Jiang, Juyong Zhang, Yang Hong, Jinhao Luo, Ligang Liu, and Hujun Bao. Bcnet: Learning body and cloth shape from a single image. InECCV, pages 18–35, 2020. 2

    2020

  26. [26]

    Leap: Liberate sparse-view 3d modeling from camera poses

    Hanwen Jiang, Zhenyu Jiang, Yue Zhao, and Qixing Huang. Leap: Liberate sparse-view 3d modeling from camera poses. InICLR, pages 1–19, 2023. 3

    2023

  27. [27]

    Real3d: Scaling up large reconstruction models with real- world images

    Hanwen Jiang, Qixing Huang, and Georgios Pavlakos. Real3d: Scaling up large reconstruction models with real- world images. InICCV, pages 1–19, 2025. 3, 5, 6

    2025

  28. [28]

    Learning category-specific mesh reconstruc- tion from image collections

    Angjoo Kanazawa, Shubham Tulsiani, Alexei A Efros, and Jitendra Malik. Learning category-specific mesh reconstruc- tion from image collections. InECCV, pages 371–386, 2018. 3

    2018

  29. [29]

    3d gaussian splatting for real-time radiance field rendering.ACM TOG, 42(4):1–14, 2023

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM TOG, 42(4):1–14, 2023. 2, 4, 13

    2023

  30. [30]

    Sapiens: Foundation for human vision mod- els

    Rawal Khirodkar, Timur Bagautdinov, Julieta Martinez, Su Zhaoen, Austin James, Peter Selednik, Stuart Anderson, and Shunsuke Saito. Sapiens: Foundation for human vision mod- els. InECCV, pages 206–228, 2024. 13

    2024

  31. [31]

    Neural human performer: Learning generalizable radiance fields for human performance rendering

    Youngjoong Kwon, Dahun Kim, Duygu Ceylan, and Henry Fuchs. Neural human performer: Learning generalizable radiance fields for human performance rendering. InNeurIPS, pages 24741–24752, 2021. 2

    2021

  32. [32]

    Pshuman: Photorealistic single-view human reconstruction using cross-scale diffusion

    Peng Li, Wangguandong Zheng, Yuan Liu, Tao Yu, Yang- guang Li, Xingqun Qi, Mengfei Li, Xiaowei Chi, Siyu Xia, Wei Xue, et al. Pshuman: Photorealistic single-view human reconstruction using cross-scale diffusion. InCVPR, pages 1–19, 2025. 2

    2025

  33. [33]

    High-fidelity clothed avatar reconstruction from a single image

    Tingting Liao, Xiaomei Zhang, Yuliang Xiu, Hongwei Yi, Xudong Liu, Guo-Jun Qi, Yong Zhang, Xuan Wang, Xiangyu Zhu, and Zhen Lei. High-fidelity clothed avatar reconstruction from a single image. InCVPR, pages 8662–8672, 2023. 2

    2023

  34. [34]

    Zero-1-to-3: Zero-shot one image to 3d object

    Ruoshi Liu, Rundi Wu, Basile Van Hoorick, Pavel Tokmakov, Sergey Zakharov, and Carl V ondrick. Zero-1-to-3: Zero-shot one image to 3d object. InICCV, pages 9298–9309, 2023. 3

    2023

  35. [35]

    Syncdreamer: Gener- ating multiview-consistent images from a single-view image

    Yuan Liu, Cheng Lin, Zijiao Zeng, Xiaoxiao Long, Lingjie Liu, Taku Komura, and Wenping Wang. Syncdreamer: Gener- ating multiview-consistent images from a single-view image. InICLR, pages 1–22, 2024. 3

    2024

  36. [36]

    Wonder3d: Single im- age to 3d using cross-domain diffusion

    Xiaoxiao Long, Yuan-Chen Guo, Cheng Lin, Yuan Liu, Zhiyang Dou, Lingjie Liu, Yuexin Ma, Song-Hai Zhang, Marc Habermann, Christian Theobalt, et al. Wonder3d: Single im- age to 3d using cross-domain diffusion. InCVPR, pages 9970–9980, 2024. 3

    2024

  37. [37]

    Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J. Black. Smpl: a skinned multi- person linear model.ACM TOG, 34(6):851–866, 2015. 2, 3, 4

    2015

  38. [38]

    Marching cubes: A high resolution 3d surface construction algorithm

    William E Lorensen and Harvey E Cline. Marching cubes: A high resolution 3d surface construction algorithm. InSeminal graphics: pioneering efforts that shaped the field, pages 347– 353, 1998. 15

    1998

  39. [39]

    Gas: Generative avatar syn- thesis from a single image

    Yixing Lu, Junting Dong, Youngjoong Kwon, Qin Zhao, Bo Dai, and Fernando De la Torre. Gas: Generative avatar syn- thesis from a single image. InICCV, pages 1–17, 2025. 2

    2025

  40. [40]

    Amass: Archive of motion capture as surface shapes

    Naureen Mahmood, Nima Ghorbani, Nikolaus F Troje, Ger- ard Pons-Moll, and Michael J Black. Amass: Archive of motion capture as surface shapes. InICCV, pages 5442–5451,

  41. [41]

    Age and realism of avatars in simulated augmented reality: Experimental evaluation of anticipated user experience

    Veronika Mikhailova, Christoph Gerhardt, Christian Kunert, Tobias Schwandt, Florian Weidner, Wolfgang Broll, and Nicola Döring. Age and realism of avatars in simulated augmented reality: Experimental evaluation of anticipated user experience. InVR, pages 83–93, 2024. 2

    2024

  42. [42]

    Nerf: Representing scenes as neural radiance fields for view synthe- sis.ECCV, pages 1–25, 2020

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthe- sis.ECCV, pages 1–25, 2020. 2, 13

    2020

  43. [43]

    Dinov2: Learning robust visual features without supervision.TMLR, pages 1–32, 2024

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision.TMLR, pages 1–32, 2024. 3

    2024

  44. [44]

    Humansplat: Generalizable single-image human gaussian splatting with structure priors

    Panwang Pan, Zhuo Su, Chenguo Lin, Zhen Fan, Yongjie Zhang, Zeming Li, Tingting Shen, Yadong Mu, and Yebin Liu. Humansplat: Generalizable single-image human gaussian splatting with structure priors. InNeurIPS, pages 1–22, 2024. 2, 5

    2024

  45. [45]

    Expressive body capture: 3d hands, face, and body from a single image

    Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed AA Osman, Dimitrios Tzionas, and Michael J Black. Expressive body capture: 3d hands, face, and body from a single image. InCVPR, pages 10975–10985,

  46. [46]

    Lhm: Large animatable human recon- struction model from a single image in seconds

    Lingteng Qiu, Xiaodong Gu, Peihao Li, Qi Zuo, Weichao Shen, Junfei Zhang, Kejie Qiu, Weihao Yuan, Guanying Chen, Zilong Dong, et al. Lhm: Large animatable human recon- struction model from a single image in seconds. InICCV, pages 1–15, 2025. 12

    2025

  47. [47]

    X-avatar: Ex- pressive human avatars

    Kaiyue Shen, Chen Guo, Manuel Kaufmann, Juan Jose Zarate, Julien Valentin, Jie Song, and Otmar Hilliges. X-avatar: Ex- pressive human avatars. InCVPR, 2023. 2, 5, 6

    2023

  48. [48]

    Mvdream: Multi-view diffusion for 3d gener- ation

    Yichun Shi, Peng Wang, Jianglong Ye, Mai Long, Kejie Li, and Xiao Yang. Mvdream: Multi-view diffusion for 3d gener- ation. InICLR, pages 1–22, 2024. 3

    2024

  49. [49]

    Canonicalfusion: Generating drivable 3d human avatars from multiple images

    Jisu Shin, Junmyeong Lee, Seongmin Lee, Min-Gyu Park, Ju- Mi Kang, Ju Hong Yoon, and Hae-Gon Jeon. Canonicalfusion: Generating drivable 3d human avatars from multiple images. InECCV, pages 38–56, 2024. 2

    2024

  50. [50]

    Recovering reflectance and illumination in a world of painted polyhedra

    Pawan Sinha and Edward Adelson. Recovering reflectance and illumination in a world of painted polyhedra. InICCV, pages 156–163, 1993. 2

    1993

  51. [51]

    Difu: Depth-guided implicit function for clothed human reconstruction

    Dae-Young Song, HeeKyung Lee, Jeongil Seo, and Donghyeon Cho. Difu: Depth-guided implicit function for clothed human reconstruction. InCVPR, pages 8738–8747,

  52. [52]

    Splatter image: Ultra-fast single-view 3d reconstruc- tion

    Stanislaw Szymanowicz, Chrisitian Rupprecht, and Andrea Vedaldi. Splatter image: Ultra-fast single-view 3d reconstruc- tion. InCVPR, pages 10208–10217, 2024. 3

    2024

  53. [53]

    Lgm: Large multi-view gaussian model for high-resolution 3d content creation

    Jiaxiang Tang, Zhaoxi Chen, Xiaokang Chen, Tengfei Wang, Gang Zeng, and Ziwei Liu. Lgm: Large multi-view gaussian model for high-resolution 3d content creation. InECCV, pages 1–18, 2024. 3

    2024

  54. [54]

    What do single-view 3d reconstruction networks learn? InCVPR, pages 3405– 3414, 2019

    Maxim Tatarchenko, Stephan R Richter, René Ranftl, Zhuwen Li, Vladlen Koltun, and Thomas Brox. What do single-view 3d reconstruction networks learn? InCVPR, pages 3405– 3414, 2019. 3

    2019

  55. [55]

    Tele-aloha: a low-budget and high- authenticity telepresence system using sparse rgb cameras

    Hanzhang Tu, Ruizhi Shao, Xue Dong, Shunyuan Zheng, Hao Zhang, Lili Chen, Meili Wang, Wenyu Li, Siyan Ma, Shengping Zhang, et al. Tele-aloha: a low-budget and high- authenticity telepresence system using sparse rgb cameras. In SIGGRAPH, pages 1–12, 2024. 2

    2024

  56. [56]

    Humandreamer-x: Photorealistic single- image human avatars reconstruction via gaussian restoration

    Boyuan Wang, Runqi Ouyang, Xiaofeng Wang, Zheng Zhu, Guosheng Zhao, Chaojun Ni, Guan Huang, Lihong Liu, and Xingang Wang. Humandreamer-x: Photorealistic single- image human avatars reconstruction via gaussian restoration. arXiv, 2025. 2

    2025

  57. [57]

    Geneman: Generalizable single-image 3d human reconstruction from multi-source human data

    Wentao Wang, Hang Ye, Fangzhou Hong, Xue Yang, Jianfu Zhang, Yizhou Wang, Ziwei Liu, and Liang Pan. Geneman: Generalizable single-image 3d human reconstruction from multi-source human data. InNeurIPS, pages 1–33, 2025. 2

    2025

  58. [58]

    Image quality assessment: from error visibility to structural similarity.IEEE TIP, 13:600–612, 2004

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE TIP, 13:600–612, 2004. 5

    2004

  59. [59]

    Crm: Single image to 3d textured mesh with convolu- tional reconstruction model

    Zhengyi Wang, Yikai Wang, Yifei Chen, Chendong Xiang, Shuo Chen, Dajiang Yu, Chongxuan Li, Hang Su, and Jun Zhu. Crm: Single image to 3d textured mesh with convolu- tional reconstruction model. InECCV, pages 57–74, 2024. 3

    2024

  60. [60]

    Meshlrm: Large reconstruction model for high- quality meshes.arXiv, pages 1–21, 2024

    Xinyue Wei, Kai Zhang, Sai Bi, Hao Tan, Fujun Luan, Valentin Deschaintre, Kalyan Sunkavalli, Hao Su, and Zex- iang Xu. Meshlrm: Large reconstruction model for high- quality meshes.arXiv, pages 1–21, 2024. 3

    2024

  61. [61]

    Template-free single-view 3d human digitalization with diffusion-guided lrm.arXiv, pages 1–21, 2024

    Zhenzhen Weng, Jingyuan Liu, Hao Tan, Zhan Xu, Yang Zhou, Serena Yeung-Levy, and Jimei Yang. Template-free single-view 3d human digitalization with diffusion-guided lrm.arXiv, pages 1–21, 2024. 2

    2024

  62. [62]

    Point transformer v3: Simpler faster stronger

    Xiaoyang Wu, Li Jiang, Peng-Shuai Wang, Zhijian Liu, Xihui Liu, Yu Qiao, Wanli Ouyang, Tong He, and Hengshuang Zhao. Point transformer v3: Simpler faster stronger. InCVPR, pages 4840–4851, 2024. 3

    2024

  63. [63]

    Structured 3d latents for scalable and versatile 3d generation

    Jianfeng Xiang, Zelong Lv, Sicheng Xu, Yu Deng, Ruicheng Wang, Bowen Zhang, Dong Chen, Xin Tong, and Jiaolong Yang. Structured 3d latents for scalable and versatile 3d generation. InCVPR, pages 1–26, 2025. 6, 8

    2025

  64. [64]

    Mvhumannet: A large- scale dataset of multi-view daily dressing human captures

    Zhangyang Xiong, Chenghong Li, Kenkun Liu, Hongjie Liao, Jianqiao Hu, Junyi Zhu, Shuliang Ning, Lingteng Qiu, Chongjie Wang, Shijie Wang, et al. Mvhumannet: A large- scale dataset of multi-view daily dressing human captures. In CVPR, pages 19801–19811, 2024. 4

    2024

  65. [65]

    ICON: Implicit clothed humans obtained from normals

    Yuliang Xiu, Jinlong Yang, Dimitrios Tzionas, and Michael J Black. ICON: Implicit clothed humans obtained from normals. InCVPR, pages 13286–13296, 2022. 2, 8

    2022

  66. [66]

    Econ: Explicit clothed humans optimized via normal integration

    Yuliang Xiu, Jinlong Yang, Xu Cao, Dimitrios Tzionas, and Michael J Black. Econ: Explicit clothed humans optimized via normal integration. InCVPR, pages 512–523, 2023. 2, 8

    2023

  67. [67]

    Instantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruction models.arXiv, pages 1–11, 2024

    Jiale Xu, Weihao Cheng, Yiming Gao, Xintao Wang, Shenghua Gao, and Ying Shan. Instantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruction models.arXiv, pages 1–11, 2024. 3

    2024

  68. [68]

    Disn: Deep implicit surface network for high-quality single-view 3d reconstruction

    Qiangeng Xu, Weiyue Wang, Duygu Ceylan, Radomir Mech, and Ulrich Neumann. Disn: Deep implicit surface network for high-quality single-view 3d reconstruction. InNeurIPS, pages 1–11, 2019. 3

    2019

  69. [69]

    Grm: Large gaussian reconstruction model for efficient 3d recon- struction and generation

    Yinghao Xu, Zifan Shi, Wang Yifan, Hansheng Chen, Ceyuan Yang, Sida Peng, Yujun Shen, and Gordon Wetzstein. Grm: Large gaussian reconstruction model for efficient 3d recon- struction and generation. InECCV, pages 1–20, 2024. 3

    2024

  70. [70]

    Human 3diffusion: Realistic avatar creation via explicit 3d consistent diffusion models

    Yuxuan Xue, Xianghui Xie, Riccardo Marin, and Gerard Pons-Moll. Human 3diffusion: Realistic avatar creation via explicit 3d consistent diffusion models. InNeurIPS, pages 1–45, 2024. 2

    2024

  71. [71]

    Infinihuman: Realistic 3d human creation with precise control

    Yuxuan Xue, Xianghui Xie, Margaret Kostyrko, and Gerard Pons-Moll. Infinihuman: Realistic 3d human creation with precise control. InSIGGRAPH Asia, pages 1–12, 2025. 2

    2025

  72. [72]

    Sigman: Scaling 3d human gaussian generation with millions of assets

    Yuhang Yang, Fengqi Liu, Yixing Lu, Qin Zhao, Pingyu Wu, Wei Zhai, Ran Yi, Yang Cao, Lizhuang Ma, Zheng-Jun Zha, et al. Sigman: Scaling 3d human gaussian generation with millions of assets. InICCV, pages 1–16, 2025. 2

    2025

  73. [73]

    Synbody: Synthetic dataset with layered human models for 3d human perception and modeling

    Zhitao Yang, Zhongang Cai, Haiyi Mei, Shuai Liu, Zhaoxi Chen, Weiye Xiao, Yukun Wei, Zhongfei Qing, Chen Wei, Bo Dai, et al. Synbody: Synthetic dataset with layered human models for 3d human perception and modeling. InICCV, pages 20282–20292, 2023. 12, 13

    2023

  74. [74]

    Function4d: Real-time human volumetric capture from very sparse consumer rgbd sensors

    Tao Yu, Zerong Zheng, Kaiwen Guo, Pengpeng Liu, Qionghai Dai, and Yebin Liu. Function4d: Real-time human volumetric capture from very sparse consumer rgbd sensors. InCVPR, pages 5746–5756, 2021. 2, 5, 6

    2021

  75. [75]

    Semantic human mesh reconstruction with textures

    Xiaoyu Zhan, Jianxin Yang, Yuanqi Li, Jie Guo, Yanwen Guo, and Wenping Wang. Semantic human mesh reconstruction with textures. InCVPR, pages 142–152, 2024. 2

    2024

  76. [76]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InCVPR, pages 586–595,

  77. [77]

    Global-correlated 3d-decoupling transformer for clothed avatar reconstruction.NeurIPS, 36, 2023

    Zechuan Zhang, Li Sun, Zongxin Yang, Ling Chen, and Yi Yang. Global-correlated 3d-decoupling transformer for clothed avatar reconstruction.NeurIPS, 36, 2023. 2

    2023

  78. [78]

    Sifu: Side-view conditioned implicit function for real-world usable clothed human reconstruction

    Zechuan Zhang, Zongxin Yang, and Yi Yang. Sifu: Side-view conditioned implicit function for real-world usable clothed human reconstruction. InCVPR, pages 9936–9947, 2024. 2, 6, 8

    2024

  79. [79]

    Hunyuan3d 2.0: Scaling diffusion mod- els for high resolution textured 3d assets generation.arXiv, pages 1–28, 2025

    Zibo Zhao, Zeqiang Lai, Qingxiang Lin, Yunfei Zhao, Haolin Liu, Shuhui Yang, Yifei Feng, Mingxin Yang, Sheng Zhang, Xianghui Yang, et al. Hunyuan3d 2.0: Scaling diffusion mod- els for high resolution textured 3d assets generation.arXiv, pages 1–28, 2025. 6, 8

    2025

  80. [80]

    Pamir: Parametric model-conditioned implicit representation for image-based human reconstruction.TPAMI, 44:3170–3184,

    Zerong Zheng, Tao Yu, Yebin Liu, and Qionghai Dai. Pamir: Parametric model-conditioned implicit representation for image-based human reconstruction.TPAMI, 44:3170–3184,

Showing first 80 references.