Pith. sign in

REVIEW 2 major objections 3 minor 96 references

Self-supervised expression learning allows Gaussian avatars from minimal data such as a single frame.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-06-28 01:59 UTC pith:7Y2LLMYR

load-bearing objection SAGE shows how to train animatable Gaussian avatars from single frames or rotations using self-supervised surfel-SDF constraints, and the full paper backs the data reduction without obvious breaks. the 2 major comments →

arxiv 2606.05912 v1 pith:7Y2LLMYR submitted 2026-06-04 cs.CV

Self-Learning Expression Deformations for Data-Efficient Gaussian Avatars

Jiahao Yang , Xiaohang Yang , Qing Wang , Yilan Dong , Gregory Slabaugh , Shanxin Yuan This is my paper
classification cs.CV
keywords 3D Gaussian splattingfacial animationexpression deformationself-superviseddata-efficientavatar reconstructionsigned distance field
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper proposes SAGE to self-learn expression deformations in 3D Gaussian face models. The method optimizes Gaussian surfels together with an SDF for surface alignment and replaces the need for long expression sequences with consistency constraints on geometry and appearance. It supports multiple low-data regimes including single-frame multiview, monocular head rotations only, and one-shot without priors. Experiments show it matches the quality of prior methods despite using far less data.

Core claim

By jointly optimizing 2D Gaussian surfels and a Signed Distance Field while enforcing geometric and appearance consistency in a self-supervised manner, the framework learns accurate expression-induced deformations, enabling animatable avatars from significantly reduced data across multiview, monocular, and one-shot settings.

What carries the argument

Self-Adaptive Gaussian Expression (SAGE) that uses SDF-enforced surfel optimization and self-supervised consistency constraints to learn deformations.

Load-bearing premise

Geometric and appearance consistency constraints suffice to learn accurate expression-induced Gaussian deformations without long training sequences or explicit labels.

What would settle it

Running the method on a new set of expressions and observing if the animated results match the fidelity of fully supervised training on extensive sequences.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Only one frame is needed for multiview reconstruction instead of thousands of timesteps.
  • Monocular settings require only head rotations, omitting expression sequences.
  • One-shot creation needs no pretraining or external priors.
  • Animation and reconstruction quality stays on par with methods that use much more data.

Where Pith is reading between the lines

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

  • The reduced data requirement may allow avatar generation in resource-limited environments.
  • Similar consistency-based learning could be tested on non-facial dynamic objects.
  • Integration with real-time capture systems might become feasible due to lower data demands.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 3 minor

Summary. The manuscript introduces SAGE, a framework for self-learning expression-induced deformations in 3D Gaussian avatars. It jointly optimizes 2D Gaussian surfels with an SDF for surface-aligned distributions and employs a self-supervised phase using geometric and appearance consistency constraints. This enables high-fidelity animatable avatars from minimal data across regimes: single-frame multiview, monocular head rotations only, or one-shot without pretraining/priors. The central claim is that reconstruction and animation quality matches state-of-the-art while reducing data requirements by several orders of magnitude.

Significance. If the results hold, this is a significant contribution to data-efficient dynamic facial modeling in computer vision and graphics. The self-supervised consistency approach and flexible multi-regime deployment directly tackle the data bottleneck of conventional Gaussian avatar pipelines. The manuscript supplies internal consistency checks, ablations, and quantitative/qualitative support that address the potential insufficiency of the constraints without long sequences, so the weakest assumption does not appear to be violated in the presented experiments.

major comments (2)
  1. [Experiments] Experiments section: the claim of 'several orders of magnitude' data reduction and 'comparable quality' requires explicit side-by-side tabulation of exact data volumes (frames, sequences, views) used by SAGE versus each baseline in the multiview, monocular, and one-shot regimes; without these numbers the central scalability claim cannot be fully evaluated.
  2. [§4.2] §4.2 (self-supervised expression learning): while ablations are reported, the geometric and appearance consistency losses should include an explicit test for trivial solutions (e.g., zero-deformation collapse) in the one-shot regime; the current formulation risks under-constraining the deformation field when input diversity is minimal.
minor comments (3)
  1. [Abstract] Abstract: the quantitative metrics and ablation results supporting the main claims are absent; moving a concise summary of key numbers (e.g., PSNR, LPIPS, data reduction factor) into the abstract would improve accessibility.
  2. [§3] Notation: the distinction between 2D Gaussian surfels and the underlying 3D representation is introduced without a dedicated equation or diagram; a small table or figure clarifying the mapping would aid readability.
  3. [Figures] Figure captions: several qualitative comparison figures lack explicit indication of which regime (multiview/single-frame, monocular, one-shot) each row corresponds to.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive review and positive assessment of our work on data-efficient Gaussian avatars. We address each major comment below and will incorporate the suggested changes in the revised manuscript.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: the claim of 'several orders of magnitude' data reduction and 'comparable quality' requires explicit side-by-side tabulation of exact data volumes (frames, sequences, views) used by SAGE versus each baseline in the multiview, monocular, and one-shot regimes; without these numbers the central scalability claim cannot be fully evaluated.

    Authors: We agree that an explicit side-by-side tabulation is necessary for full evaluation of the scalability claims. In the revised manuscript, we will add a dedicated table in the Experiments section that reports the exact number of frames, sequences, and views used by SAGE and each baseline method across the multiview, monocular, and one-shot regimes. This will directly support the data reduction statements with quantitative comparisons. revision: yes

  2. Referee: [§4.2] §4.2 (self-supervised expression learning): while ablations are reported, the geometric and appearance consistency losses should include an explicit test for trivial solutions (e.g., zero-deformation collapse) in the one-shot regime; the current formulation risks under-constraining the deformation field when input diversity is minimal.

    Authors: We acknowledge the potential concern about under-constraining in minimal-input settings. In the revised version of §4.2, we will include an additional ablation experiment that explicitly tests for zero-deformation collapse in the one-shot regime. This will involve comparing optimization outcomes with and without the geometric and appearance consistency losses to demonstrate that the constraints prevent trivial solutions. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's core contribution is a self-supervised optimization framework that jointly fits 2D Gaussian surfels and an SDF, then applies geometric/appearance consistency losses to learn expression deformations from minimal data. No equation reduces to its own inputs by construction (no fitted parameter renamed as prediction, no self-definitional loop). No load-bearing uniqueness theorem is imported via self-citation. The consistency constraints are standard self-supervision terms whose sufficiency is tested via ablations and cross-regime experiments rather than assumed. The derivation chain remains self-contained against external benchmarks and does not collapse to tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Review performed on abstract only; no equations, loss formulations, or implementation details available to enumerate free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5745 in / 1059 out tokens · 21597 ms · 2026-06-28T01:59:34.181092+00:00 · methodology

0 comments
read the original abstract

Modeling dynamic facial expressions using 3D Gaussian representations remains challenging due to their unstructured nature. Conventional Gaussian avatar pipelines require extensive multiview and sequential expression data, limiting scalability and accessibility. In this work, we introduce Self-Adaptive Gaussian Expression (SAGE), a framework for self-learning expression-induced Gaussian deformations that enables high-fidelity, animatable avatars from minimal input data. Our method jointly optimizes 2D Gaussian surfels and a Signed Distance Field (SDF) to enforce compact, surface-aligned Gaussian distributions, while a self-supervised expression learning phase replaces long training sequences with geometric and appearance consistency constraints. This design allows flexible deployment across multiple reconstruction regimes: in the multiview setting, only a single frame (timestep) is required instead of thousands; in the monocular setting, only head rotations are needed without expression sequences; and in the one-shot setting, no pretraining or priors are necessary. Experiments demonstrate that our approach achieves reconstruction and animation quality comparable to state-of-the-art methods, while reducing data requirements by several orders of magnitude. Our results highlight the potential of self-supervised Gaussian deformation learning as a step toward accessible, data-efficient avatar creation.

Figures

Figures reproduced from arXiv: 2606.05912 by Gregory Slabaugh, Jiahao Yang, Qing Wang, Shanxin Yuan, Xiaohang Yang, Yilan Dong.

Figure 1
Figure 1. Figure 1: We introduce SAGE, a method for creating animatable Gaussian avatars from minimal training data. Our pipeline [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Joint Optimization (top): SDF branch predicts Gaussian attributes and 2DGS will rasterize both depth and normal [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparisons for self-reenactment quality under Multiview input setting. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparisons for self-reenactment quality under different input types. Results (a) refers to monocular [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: We also conduct qualitative assessment of our pipeline’s self-reenactment performance on INSTA’s data which is a [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative evaluation of multi-view consistency of [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparison of geometry and rendered [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparisons for cross-reenactment. The [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: We visualize three pixel-level error maps used in [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Training log of distortion loss on the second training [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Static canonical SDF meshes and reconstructed dynamic meshes across identities and expressions. The canonical [PITH_FULL_IMAGE:figures/full_fig_p013_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Comparison of reconstructed normals across a range of facial expressions. Middle row is the monocular adaptation [PITH_FULL_IMAGE:figures/full_fig_p014_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Comparison of reconstructed normals across a range of identities. Middle row is the multiview adaptation of our [PITH_FULL_IMAGE:figures/full_fig_p014_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Experiment on baseline models’ self-reenactment quality as the number of frames used for training increases. [PITH_FULL_IMAGE:figures/full_fig_p014_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Qualitative comparison of novel-view synthesis under sparse-view input conditions. We compare RGBAvatar, LAM, [PITH_FULL_IMAGE:figures/full_fig_p015_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Visualization of the six expressions used in the expression-dependency study. From left to right, the selected [PITH_FULL_IMAGE:figures/full_fig_p015_17.png] view at source ↗

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

96 extracted references · 78 canonical work pages · 7 internal anchors

  1. [1]

    Abdal, W

    R. Abdal, W. Yifan, Z. Shi, Y . Xu, R. Po, Z. Kuang, Q. Chen, D.- Y . Yeung, and G. Wetzstein. Gaussian Shell Maps for Efficient 3D Human Generation, Nov. 2023. arXiv:2311.17857 [cs]

    work page arXiv 2023

  2. [2]

    Athar, Z

    S. Athar, Z. Xu, K. Sunkavalli, E. Shechtman, and Z. Shu. RigNeRF: Fully Controllable Neural 3D Portraits, June 2022. arXiv:2206.06481 [cs]

    work page arXiv 2022

  3. [3]

    Blanz and T

    V . Blanz and T. Vetter. A morphable model for the synthesis of 3D faces. InProceedings of the 26th annual conference on Computer graphics and interactive techniques, SIGGRAPH ’99, pages 187–194, USA, July 1999. ACM Press/Addison-Wesley Publishing Co

    1999

  4. [4]

    Bolkart, T

    T. Bolkart, T. Li, and M. J. Black. Instant Multi-View Head Capture through Learnable Registration, June 2023. arXiv:2306.07437 [cs]

    work page arXiv 2023

  5. [5]

    M. C. B ¨uhler, K. Sarkar, T. Shah, G. Li, D. Wang, L. Helminger, S. Orts-Escolano, D. Lagun, O. Hilliges, T. Beeler, and A. Meka. Preface: A Data-driven V olumetric Prior for Few-shot Ultra High- resolution Face Synthesis, Sept. 2023. arXiv:2309.16859 [cs]

    work page arXiv 2023

  6. [6]

    C. Cao, Y . Weng, S. Zhou, Y . Tong, and K. Zhou. FaceWarehouse: A 3D Facial Expression Database for Visual Computing.IEEE Transactions on Visualization and Computer Graphics, 20(3):413–425, Mar. 2014. Conference Name: IEEE Transactions on Visualization and Computer Graphics

    2014

  7. [7]

    Available

    D. Charatan, S. Li, A. Tagliasacchi, and V . Sitzmann. pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction, Apr. 2024. arXiv:2312.12337 [cs]

    work page arXiv 2024

  8. [8]

    H. Chen, C. Li, and G. H. Lee. NeuSG: Neural Implicit Surface Reconstruction with 3D Gaussian Splatting Guidance, Dec. 2023. arXiv:2312.00846 [cs]

    work page arXiv 2023

  9. [9]

    P. Chen, X. Wei, Q. Wuwu, X. Wang, X. Xiao, and M. Lu. Mixedgaus- sianavatar: Realistically and geometrically accurate head avatar via mixed 2d-3d gaussians, 2025

    2025

  10. [10]

    X. Chen, M. Mihajlovic, S. Wang, S. Prokudin, and S. Tang. Mor- phable Diffusion: 3D-Consistent Diffusion for Single-image Avatar Creation, Apr. 2024. arXiv:2401.04728 [cs]

    work page arXiv 2024

  11. [11]

    Y . Chen, L. Wang, Q. Li, H. Xiao, S. Zhang, H. Yao, and Y . Liu. MonoGaussianAvatar: Monocular Gaussian Point-based Head Avatar, Dec. 2023. arXiv:2312.04558 [cs]

    work page arXiv 2023

  12. [12]

    Y . Chen, H. Xu, C. Zheng, B. Zhuang, M. Pollefeys, A. Geiger, T.-J. Cham, and J. Cai. MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images. InComputer Vision – ECCV 2024, volume 15079, pages 370–386. Springer Nature Switzerland, 2025. arXiv:2403.14627 [cs]

    work page arXiv 2024

  13. [13]

    X. Chu, Y . Li, A. Zeng, T. Yang, L. Lin, Y . Liu, and T. Harada. GPAvatar: Generalizable and Precise Head Avatar from Image(s), Jan

  14. [14]

    arXiv:2401.10215 [cs]

    work page arXiv

  15. [15]

    Y . Deng, D. Wang, X. Ren, X. Chen, and B. Wang. Portrait4D: Learning One-Shot 4D Head Avatar Synthesis using Synthetic Data, June 2024. arXiv:2311.18729 [cs]

    work page arXiv 2024

  16. [16]

    Dhamo, Y

    H. Dhamo, Y . Nie, A. Moreau, J. Song, R. Shaw, Y . Zhou, and E. P ´erez-Pellitero. HeadGaS: Real-Time Animatable Head Avatars via 3D Gaussian Splatting, Dec. 2023. arXiv:2312.02902 [cs]

    work page arXiv 2023

  17. [17]

    Y . Feng, H. Feng, M. J. Black, and T. Bolkart. Learning an animatable detailed 3D face model from in-the-wild images.ACM Trans. Graph., 40(4):88:1–88:13, July 2021

    2021

  18. [18]

    Gafni, J

    G. Gafni, J. Thies, M. Zollh ¨ofer, and M. Nießner. Dynamic Neural Radiance Fields for Monocular 4D Facial Avatar Reconstruction, Dec

  19. [19]

    arXiv:2012.03065 [cs]

    work page arXiv 2012

  20. [20]

    J. Gao, C. Gu, Y . Lin, H. Zhu, X. Cao, L. Zhang, and Y . Yao. Relightable 3D Gaussian: Real-time Point Cloud Relighting with BRDF Decomposition and Ray Tracing, Nov. 2023. arXiv:2311.16043 [cs]

    work page arXiv 2023

  21. [21]

    X. Gao, C. Zhong, J. Xiang, Y . Hong, Y . Guo, and J. Zhang. Reconstructing Personalized Semantic Facial NeRF Models From Monocular Video.ACM Transactions on Graphics, 41(6):1–12, Dec

  22. [22]

    arXiv:2210.06108 [cs]

    work page arXiv

  23. [23]

    Giebenhain, T

    S. Giebenhain, T. Kirschstein, M. Georgopoulos, M. R ¨unz, L. Agapito, and M. Nießner. Learning Neural Parametric Head Models, Apr. 2023. arXiv:2212.02761 [cs]

    work page arXiv 2023

  24. [24]

    Giebenhain, T

    S. Giebenhain, T. Kirschstein, M. Georgopoulos, M. R ¨unz, L. Agapito, and M. Nießner. MonoNPHM: Dynamic Head Reconstruction from Monocular Videos, May 2024. arXiv:2312.06740 [cs]

    work page arXiv 2024

  25. [25]

    Giebenhain, T

    S. Giebenhain, T. Kirschstein, M. R ¨unz, L. Agapito, and M. Nießner. NPGA: Neural Parametric Gaussian Avatars, Sept. 2024. arXiv:2405.19331

    work page arXiv 2024

  26. [26]

    Grassal, M

    P.-W. Grassal, M. Prinzler, T. Leistner, C. Rother, M. Nießner, and J. Thies. Neural Head Avatars from Monocular RGB Videos, Mar

  27. [27]

    arXiv:2112.01554 [cs]

    work page arXiv

  28. [28]

    Gu ´edon and V

    A. Gu ´edon and V . Lepetit. SuGaR: Surface-Aligned Gaussian Splat- ting for Efficient 3D Mesh Reconstruction and High-Quality Mesh Rendering, Dec. 2023. arXiv:2311.12775 [cs]

    work page arXiv 2023

  29. [29]

    Y . He, X. Gu, X. Ye, C. Xu, Z. Zhao, Y . Dong, W. Yuan, Z. Dong, and L. Bo. LAM: Large Avatar Model for One-shot Animatable Gaussian Head, Apr. 2025. arXiv:2502.17796 [cs]

    work page arXiv 2025

  30. [30]

    Y . Hong, B. Peng, H. Xiao, L. Liu, and J. Zhang. Head- NeRF: A Real-time NeRF-based Parametric Head Model, Apr. 2022. arXiv:2112.05637 [cs]

    work page arXiv 2022

  31. [31]

    Gauhuman: Articulated gaussian splatting from monocular human videos,

    S. Hu and Z. Liu. GauHuman: Articulated Gaussian Splatting from Monocular Human Videos, Dec. 2023. arXiv:2312.02973 [cs] version: 1

    work page arXiv 2023

  32. [32]

    Huang, Z

    B. Huang, Z. Yu, A. Chen, A. Geiger, and S. Gao. 2D Gaussian Splat- ting for Geometrically Accurate Radiance Fields. InSpecial Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers ’24, pages 1–11, July 2024. arXiv:2403.17888 [cs]

    work page arXiv 2024

  33. [33]

    R. Jena, G. S. Iyer, S. Choudhary, B. Smith, P. Chaudhari, and J. Gee. SplatArmor: Articulated Gaussian splatting for animatable humans from monocular RGB videos, Nov. 2023. arXiv:2311.10812 [cs]

    work page arXiv 2023

  34. [34]

    Karras, S

    T. Karras, S. Laine, M. Aittala, J. Hellsten, J. Lehtinen, and T. Aila. Analyzing and Improving the Image Quality of StyleGAN, Mar. 2020. arXiv:1912.04958 [cs]

    work page arXiv 2020

  35. [35]

    arXiv preprint arXiv:2308.04079 (2023)

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis. 3D Gaus- sian Splatting for Real-Time Radiance Field Rendering, Aug. 2023. arXiv:2308.04079 [cs]

    work page arXiv 2023

  36. [36]

    H. Kim, P. Garrido, A. Tewari, W. Xu, J. Thies, M. Nießner, P. P ´erez, C. Richardt, M. Zollh ¨ofer, and C. Theobalt. Deep Video Portraits, May 2018. arXiv:1805.11714 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  37. [37]

    Kirschstein, S

    T. Kirschstein, S. Giebenhain, and M. Nießner. DiffusionAvatars: Deferred Diffusion for High-fidelity 3D Head Avatars, Apr. 2024. arXiv:2311.18635 [cs]

    work page arXiv 2024

  38. [38]

    Kirschstein, S

    T. Kirschstein, S. Giebenhain, J. Tang, M. Georgopoulos, and M. Nießner. GGHead: Fast and Generalizable 3D Gaussian Heads. InSIGGRAPH Asia 2024 Conference Papers, pages 1–11, Dec. 2024. arXiv:2406.09377 [cs]

    work page arXiv 2024

  39. [39]

    Kirschstein, S

    T. Kirschstein, S. Qian, S. Giebenhain, T. Walter, and M. Nießner. NeRSemble: Multi-view Radiance Field Reconstruction of Human Heads.ACM Transactions on Graphics, 42(4):1–14, Aug. 2023. arXiv:2305.03027 [cs]

    work page arXiv 2023

  40. [40]

    J. Lee, T. Kang, M. C. B ¨uhler, M.-J. Kim, S. Hwang, J. Hyung, H. Jang, and J. Choo. SurFhead: Affine Rig Blending for Geo- metrically Accurate 2D Gaussian Surfel Head Avatars, Apr. 2025. arXiv:2410.11682 [cs]

    work page arXiv 2025

  41. [41]

    L. Li, Y . Li, Y . Weng, Y . Zheng, and K. Zhou. RGBAvatar: Reduced Gaussian Blendshapes for Online Modeling of Head Avatars, Mar

  42. [42]

    arXiv:2503.12886 [cs] version: 1

    work page arXiv

  43. [43]

    T. Li, T. Bolkart, M. J. Black, H. Li, and J. Romero. Learning a model of facial shape and expression from 4D scans.ACM Transactions on Graphics, 36(6):1–17, Dec. 2017

    2017

  44. [44]

    Lombardi, T

    S. Lombardi, T. Simon, G. Schwartz, M. Zollhoefer, Y . Sheikh, and J. Saragih. Mixture of V olumetric Primitives for Efficient Neural Rendering, May 2021. arXiv:2103.01954 [cs]

    work page arXiv 2021

  45. [45]

    W. Lyu, Y . Zhou, M.-H. Yang, and Z. Shu. FaceLift: Learning Generalizable Single Image 3D Face Reconstruction from Synthetic Heads, Aug. 2025. arXiv:2412.17812 [cs]

    work page arXiv 2025

  46. [46]

    Lyu, Y .-T

    X. Lyu, Y .-T. Sun, Y .-H. Huang, X. Wu, Z. Yang, Y . Chen, J. Pang, and X. Qi. 3DGSR: Implicit Surface Reconstruction with 3D Gaussian Splatting, Mar. 2024. arXiv:2404.00409 [cs]

    work page arXiv 2024

  47. [47]

    S. Ma, T. Simon, J. Saragih, D. Wang, Y . Li, F. D. L. Torre, and Y . Sheikh. Pixel Codec Avatars, Apr. 2021. arXiv:2104.04638 [cs]

    work page arXiv 2021

  48. [48]

    S. Ma, Y . Weng, T. Shao, and K. Zhou. 3D Gaussian Blendshapes for Head Avatar Animation. InSpecial Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers ’24, pages 1–10, July 2024. arXiv:2404.19398 [cs]

    work page arXiv 2024

  49. [49]

    G. Moon, W. Xu, R. Joshi, C. Wu, and T. Shiratori. Authentic Hand Avatar from a Phone Scan via Universal Hand Model, May 2024. arXiv:2405.07933 [cs] version: 1

    work page arXiv 2024

  50. [50]

    M ¨uller

    T. M ¨uller. tiny-cuda-nn, Apr. 2021. original-date: 2021-04- 16T17:53:10Z

    2021

  51. [51]

    D. Pan, L. Zhuo, J. Piao, H. Luo, W. Cheng, Y . Wang, S. Fan, S. Liu, L. Yang, B. Dai, Z. Liu, C. C. Loy, C. Qian, W. Wu, D. Lin, and K.-Y . Lin. RenderMe-360: A Large Digital Asset Library and Benchmarks Towards High-fidelity Head Avatars, May 2023. arXiv:2305.13353 [cs]

    work page arXiv 2023

  52. [52]

    K. Park, U. Sinha, J. T. Barron, S. Bouaziz, D. B. Goldman, S. M. Seitz, and R. Martin-Brualla. Nerfies: Deformable Neural Radiance Fields, Sept. 2021. arXiv:2011.12948 [cs]. 9

    work page arXiv 2021

  53. [53]

    K. Park, U. Sinha, P. Hedman, J. T. Barron, S. Bouaziz, D. B. Goldman, R. Martin-Brualla, and S. M. Seitz. HyperNeRF: A Higher-Dimensional Representation for Topologically Varying Neural Radiance Fields, Sept. 2021. arXiv:2106.13228 [cs]

    work page arXiv 2021

  54. [54]

    Expressive Body Capture: 3D Hands, Face, and Body from a Single Image

    G. Pavlakos, V . Choutas, N. Ghorbani, T. Bolkart, A. A. A. Osman, D. Tzionas, and M. J. Black. Expressive Body Capture: 3D Hands, Face, and Body from a Single Image, Apr. 2019. arXiv:1904.05866 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  55. [55]

    Pumarola, E

    A. Pumarola, E. Corona, G. Pons-Moll, and F. Moreno-Noguer. D- NeRF: Neural Radiance Fields for Dynamic Scenes, Nov. 2020. arXiv:2011.13961 [cs]

    work page arXiv 2020

  56. [56]

    S. Qian. Vhap: Versatile head alignment with adaptive appearance priors, sep 2024

    2024

  57. [57]

    S. Qian, T. Kirschstein, L. Schoneveld, D. Davoli, S. Giebenhain, and M. Nießner. GaussianAvatars: Photorealistic Head Avatars with Rigged 3D Gaussians, Mar. 2024. arXiv:2312.02069 [cs]

    work page arXiv 2024

  58. [58]

    Generating 3D faces using Convolutional Mesh Autoencoders

    A. Ranjan, T. Bolkart, S. Sanyal, and M. J. Black. Generat- ing 3D faces using Convolutional Mesh Autoencoders, July 2018. arXiv:1807.10267 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  59. [59]

    Y . Ren, G. Li, Y . Chen, T. H. Li, and S. Liu. PIRenderer: Controllable Portrait Image Generation via Semantic Neural Rendering, Sept. 2021. arXiv:2109.08379 [cs]

    work page arXiv 2021

  60. [60]

    Saunders, C

    J. Saunders, C. Hewitt, Y . Jian, M. Kowalski, T. Baltrusaitis, Y . Chen, D. Cosker, V . Estellers, N. Gyde, V . P. Namboodiri, and B. E. Lundell. GASP: Gaussian Avatars with Synthetic Priors, Dec. 2024. arXiv:2412.07739 [cs]

    work page arXiv 2024

  61. [61]

    Schoneveld, Z

    L. Schoneveld, Z. Chen, D. Davoli, J. Tang, S. Terazawa, K. Nishino, and M. Nießner. Sheap: Self-supervised head geometry predictor learned via 2d gaussians, 2025

    2025

  62. [62]

    Siarohin, S

    A. Siarohin, S. Lathuili `ere, S. Tulyakov, E. Ricci, and N. Sebe. First Order Motion Model for Image Animation, Oct. 2020. arXiv:2003.00196 [cs]

    work page arXiv 2020

  63. [63]

    J. Tang, D. Davoli, T. Kirschstein, L. Schoneveld, and M. Niessner. GAF: Gaussian Avatar Reconstruction from Monocular Videos via Multi-view Diffusion, Apr. 2025. arXiv:2412.10209 [cs]

    work page arXiv 2025

  64. [64]

    Teotia, H

    K. Teotia, H. Kim, P. Garrido, M. Habermann, M. Elgharib, and C. Theobalt. GaussianHeads: End-to-End Learning of Drivable Gaus- sian Head Avatars from Coarse-to-fine Representations, Sept. 2024. arXiv:2409.11951 [cs]

    work page arXiv 2024

  65. [65]

    MoFA: Model-based Deep Convolutional Face Autoencoder for Unsupervised Monocular Reconstruction

    A. Tewari, M. Zollh ¨ofer, H. Kim, P. Garrido, F. Bernard, P. P ´erez, and C. Theobalt. MoFA: Model-based Deep Convolutional Face Autoencoder for Unsupervised Monocular Reconstruction, Dec. 2017. arXiv:1703.10580 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  66. [66]

    D. Wang, P. Chandran, G. Zoss, D. Bradley, and P. Gotardo. MoRF: Morphable Radiance Fields for Multiview Neural Head Modeling. In ACM SIGGRAPH 2022 Conference Proceedings, SIGGRAPH ’22, pages 1–9, New York, NY , USA, July 2022. Association for Com- puting Machinery

    2022

  67. [67]

    Wang, J.-C

    J. Wang, J.-C. Xie, X. Li, F. Xu, C.-M. Pun, and H. Gao. Gaussian- Head: High-fidelity Head Avatars with Learnable Gaussian Derivation, May 2024. arXiv:2312.01632 [cs]

    work page arXiv 2024

  68. [68]

    L. Wang, Z. Chen, T. Yu, C. Ma, L. Li, and Y . Liu. FaceVerse: a Fine-grained and Detail-controllable 3D Face Morphable Model from a Hybrid Dataset, May 2022. arXiv:2203.14057 [cs]

    work page arXiv 2022

  69. [69]

    P. Wang, L. Liu, Y . Liu, C. Theobalt, T. Komura, and W. Wang. NeuS: Learning Neural Implicit Surfaces by V olume Rendering for Multi- view Reconstruction, Feb. 2023. arXiv:2106.10689 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  70. [70]

    Y . Wang, Q. Han, M. Habermann, K. Daniilidis, C. Theobalt, and L. Liu. NeuS2: Fast Learning of Neural Implicit Surfaces for Multi- view Reconstruction, Nov. 2023. arXiv:2212.05231 [cs]

    work page arXiv 2023

  71. [71]

    X2Face: A network for controlling face generation by using images, audio, and pose codes

    O. Wiles, A. S. Koepke, and A. Zisserman. X2Face: A network for controlling face generation by using images, audio, and pose codes, July 2018. arXiv:1807.10550 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  72. [72]

    Xiang, X

    J. Xiang, X. Gao, Y . Guo, and J. Zhang. FlashAvatar: High- fidelity Head Avatar with Efficient Gaussian Embedding, Mar. 2024. arXiv:2312.02214 [cs]

    work page arXiv 2024

  73. [73]

    Y . Xu, B. Chen, Z. Li, H. Zhang, L. Wang, Z. Zheng, and Y . Liu. Gaussian Head Avatar: Ultra High-fidelity Head Avatar via Dynamic Gaussians, Mar. 2024. arXiv:2312.03029 [cs]

    work page arXiv 2024

  74. [74]

    Y . Xu, Z. Su, Q. Wu, and Y . Liu. GPHM: Gaussian Parametric Head Model for Monocular Head Avatar Reconstruction, Oct. 2024. arXiv:2407.15070 [cs]

    work page arXiv 2024

  75. [75]

    Y . Xu, L. Wang, X. Zhao, H. Zhang, and Y . Liu. AvatarMA V: Fast 3D Head Avatar Reconstruction Using Motion-Aware Neural V oxels. InSpecial Interest Group on Computer Graphics and Interactive Techniques Conference Conference Proceedings, pages 1–10, July

  76. [76]

    arXiv:2211.13206 [cs]

    work page arXiv

  77. [77]

    Y . Yan, Z. Zhou, Z. Wang, J. Gao, and X. Yang. DialogueNeRF: To- wards Realistic Avatar Face-to-Face Conversation Video Generation, Aug. 2023. arXiv:2203.07931 [cs]

    work page arXiv 2023

  78. [78]

    G. Yang, S. Belongie, B. Hariharan, and V . Koltun. Geometry Process- ing with Neural Fields. InAdvances in Neural Information Processing Systems, volume 34, pages 22483–22497. Curran Associates, Inc., 2021

    2021

  79. [79]

    H. Yang, M. Zheng, C. Ma, Y .-K. Lai, P. Wan, and H. Huang. VRMM: A V olumetric Relightable Morphable Head Model. InSpecial Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers ’24, pages 1–11, July 2024. arXiv:2402.04101 [cs]

    work page arXiv 2024

  80. [80]

    Yenamandra, A

    T. Yenamandra, A. Tewari, F. Bernard, H.-P. Seidel, M. Elgharib, D. Cremers, and C. Theobalt. i3DMM: Deep Implicit 3D Morphable Model of Human Heads, Nov. 2020. arXiv:2011.14143 [cs]

    work page arXiv 2020

Showing first 80 references.