Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

SVG-Head: Hybrid Surface-Volumetric Gaussians for High-Fidelity Head Reconstruction and Real-Time Editing

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read SVG-Head claims to be the first Gaussian-based head avatar with explicit texture images, enabling real-time appearance editing at about 0.014 seconds per edit while keeping rendering quality competitive with non-editable methods.

desk verdict A solid hybrid Gaussian head-avatar system with a real editing contribution, though the 'unique UV' guarantee is overstated and the editing eval is mostly qualitative. read the letter →

arxiv 2508.09597 v2 pith:Q66G2LWQ submitted 2025-08-13 cs.CV

Heyi Sun , Cong Wang , Tian-Xing Xu , Jingwei Huang , Di Kang , Chunchao Guo , Song-Hai Zhang This is my paper
classification cs.CV
keywords headavatar3DGaussiansplattingtextureeditingreal-timeFLAMEmeshhybridrepresentationUVmappingneuralrendering
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that a head avatar can be both photorealistic and truly editable if its global appearance is stored in explicit texture images that surface-bound Gaussians sample through a mesh-aware UV mapping, instead of being entangled in per-Gaussian colors. To do this it introduces surface Gaussians that are pinned to a FLAME mesh and fetch colors from two learnable texture images (a diffuse map and an expression-dependent dynamic map), plus volumetric Gaussians that are bound to the mesh but free to move, capturing non-Lambertian regions like hair and lips where surface-bound Gaussians underfit. A hierarchical optimization first trains the surface Gaussians alone to produce sharp, editable textures, then jointly optimizes both types with regularizers that keep the two roles disentangled. On the NeRSemble dataset the paper reports the best metrics among editable head-avatar methods (30.3 PSNR novel-view, 26.0 novel-expression) and a per-edit time of about 0.014 seconds, versus minutes to hours for neural-texture editing. If correct, this closes the gap between classic texture-mapped meshes and modern Gaussian splatting: avatars get a real paint-on texture while keeping high-fidelity rendering and animation.

What carries the argument

The mesh-aware Gaussian UV mapping: for a ray–Gaussian intersection $p$, the method projects $p$ onto the Gaussian's parent FLAME triangle along the face normal, then barycentrically interpolates UV coordinates, and condenses the whole projection-plus-interpolation into a single affine transform $T(\mu_i)$ per Gaussian center (Eq. 6). The load-bearing constraint is that surface Gaussian centers stay on the mesh and their rotations equal their face rotations (Eqs. 7–8); this guarantees that all intersections on a ray yield the same UV coordinate, giving sharp texture images and making paint-on editing possible. The appearance itself is carried by two learnable texture images—a diffuse map $T_

What would settle it

Take a trained SVG-Head model and paint a small patch on one texture location; render the same edited avatar from two cameras whose rays sample the same 3D surface point. If the patch appears at a different 3D location in the two views, or if rendering the same viewpoint with the edit shifts by more than a pixel, the unique-UV claim fails. Quantitatively, compute per-pixel the diameter of the set $\{\phi(I(r, G_i))\}$ of UV coordinates of all Gaussians contributing to that pixel; if it exceeds one texel anywhere in the face region, texture sharpness and edit fidelity are not guaranteed. The pa

Watch

Extended reading notes

Core claim

The central claim is that the entanglement of geometry and appearance in 3D Gaussian Splatting—the thing that blocks editing—can be broken by making a subset of Gaussians 'surface' Gaussians whose colors come from explicit texture images via a uniquely-defined UV coordinate. The trick is to constrain each surface Gaussian's center to lie on its parent FLAME triangle (learnable barycentric coordinates) and to set its rotation equal to the triangle's frame (Eqs. 7–8); then every ray–Gaussian intersection projects to the same UV coordinate, so the learned texture stays sharp rather than averaging conflicting samples. The mesh-aware UV mapping computes this as a single precomputed affine transfo

Load-bearing premise

Everything hinges on the assumption that every ray–Gaussian intersection projects to a single unambiguous UV coordinate on the FLAME mesh; wherever the tracked mesh deviates from the true head surface—acknowledged for hair and lips—texture blur and misapplied edits follow.

Editorial extensions

If this is right

  • Texture edits propagate immediately: painting on the diffuse map changes the rendered avatar in about 0.014 s, replacing an optimization loop that takes minutes to hours in MeGA and GaussianAvatar-Editor.
  • Because both Gaussian sets are bound to the FLAME mesh, appearance editing composes with animation: edits persist under new pose, shape, and expression parameters.
  • The unique-UV constraint is what keeps the texture perceptually sharp; the ablations show that relaxing it (optimizing rotations and positions) drops novel-expression PSNR from 30.9 to 29.6–30.0 and produces visibly blurrier texture images.
  • The hierarchical schedule—optimize surf-GS first, then add vol-GS with only opacity and dynamic texture unfrozen—prevents under-constrained joint optimization, which otherwise reduces PSNR to 28.0.
  • The explicit texture image opens standard image-editing operations (painting, inpainting, style transfer) to be applied directly to the avatar, which was previously possible only through slow per-edit optimization.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The uniqueness-of-UV condition suggests a simple quality metric for any editable-hybrid method: measure, per pixel, the spread of UV coordinates across all ray–Gaussian intersections; if the spread exceeds one texel, the texture is expected to blur and editing to smear. This metric can be computed without re-rendering.
  • The paper's acknowledged failure on non-Lambertian hair means the 'explicit texture' editing benefit currently applies to face regions only. A natural extension is a per-Gaussian edit-mask blending volumetric and surface colors, allowing users to paint edits that also stick to hair and lips.
  • Because the dynamic texture is expression-conditioned, user edits to the diffuse map will not, by default, affect wrinkles or expression-dependent shading. Explicitly providing a mask combining diffuse edits with dynamic texture output would give artists control over which appearance changes are permanent versus expression-linked.
  • The affine UV mapping's low cost suggests it could be extended to mipmapped or tiled textures for gigapixel-resolution avatars, or to bake per-Gaussian Jacobians to allow texture-space deformations during animation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes SVG-Head, a hybrid head-avatar representation combining surface-bound 3D Gaussians (surf-GS) that sample colors from two learnable texture images and volumetric Gaussians (vol-GS) that model residual non-Lambertian regions. The surf-GS are constrained to FLAME mesh faces and aligned with face normals, and a mesh-aware UV mapping is used to turn ray-Gaussian intersections into texture coordinates. A hierarchical optimization first trains surf-GS and then jointly optimizes them with vol-GS. On the NeRSemble dataset the method is reported to achieve rendering quality comparable to non-editable state-of-the-art methods (30.3 dB novel-view PSNR vs. 31.6 dB for GaussianAvatars) while being the first Gaussian head avatar with explicit texture editing at about 0.014 s per edit.

Significance. If the technical claims are correct, the paper makes a useful contribution: it demonstrates that a hybrid surface/volumetric Gaussian representation can support both high-quality head rendering and immediate texture-space editing. The use of a public dataset, comparisons with public baselines, and a series of ablations are strengths. The central novelty—real-time appearance editing through an explicit texture map for Gaussian head avatars—is credible and potentially valuable for AR/VR applications. However, the mathematical justification for the UV-consistency mechanism is not fully established, and the experimental evidence for the editing claim is largely qualitative, with point estimates and no variance reporting.

major comments (3)
  1. [3.2.1, Eqs. (7)-(8) and Fig. 2] The claim that constraining surf-GS centers to the mesh and aligning rotations with face normals 'ensures intersections lie on the surface, giving each pixel a unique UV coordinate' is not mathematically justified. For a ray that does not pass exactly through a Gaussian center c, the closest point p on the ray to c lies off the surface; projecting p along the face normal does not generally return c unless the ray direction has no tangent component. Since the normal-direction scale is a learnable parameter and L_scale (Eq. 12) only penalizes large scales, surf-GS are not zero-thickness. Consequently, Eq. (6) can map different Gaussians on the same ray to different UVs, and the claimed consistency argument is invalid. Please provide a rigorous derivation, modify the representation (e.g., use the Gaussian center as the texture-lookup point, or hard-constrain the normal scale), or explicitly
  2. [Tables 1-3] All reported metrics are point estimates without error bars or per-subject variance. Table 2 reports ablations only on subject 306, and Table 1 appears to aggregate over multiple subjects without stating how many or how stable the results are. The key comparison to GaussianAvatars (30.3 vs 31.6 PSNR) may be within run-to-run noise. Without standard deviations, per-subject breakdowns, or a significance test, the claim of being 'best among editable reconstruction methods' and 'comparable' to non-editable methods is not statistically supported. Please add variance information.
  3. [Section 4.2] The real-time editing evaluation is qualitative only: three edited examples and rough timings of 0.014 s vs. 30 min / 1.5 h / 4.5 h. The timings lack measurement details (hardware, protocol, whether the comparison methods used the same resolution and editing region). Since real-time editing is the paper's primary novelty, the evidence is thin. Please add quantitative editing metrics (e.g., edit-region fidelity, view consistency after editing) or at least a small user study; otherwise the central claim should be framed as a qualitative demonstration.
minor comments (5)
  1. [Eq. (6)] The transformation matrix T(µ_i) is not defined. Please specify how it is computed from the face's barycentric UV mapping and how it is precomputed before rendering.
  2. [Eq. (9)] There is a missing parenthesis in 'h(ϕ(I(p, Tdy + Tdiff)'; the intended function call is not clear.
  3. [Section 4, first paragraph] The image resolution is written as '802 × 550'; this looks like a typo for 800 × 550 or another standard resolution. Please correct.
  4. [Section 3.2.3] 'Differential hybrid rendering' should be 'differentiable hybrid rendering'.
  5. [Related Work] The abbreviation 'NPV A' in the PointAvatar paragraph is a formatting typo.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: central claims rest on external evaluation and novel design, not on fitted predictions or load-bearing self-citation.

full rationale

The paper's derivation is self-contained. The proposed hybrid representation is evaluated against external baselines (GaussianAvatars, MeGA, PointAvatar, etc.) on the public NeRSemble dataset, with reported metrics and ablations. The 'real-time editing' claim is a designed property of using explicit texture images with a differentiable mesh-aware UV mapping; it is not a quantity fitted to data and then re-predicted. Equation (6) is a first-order affine expansion of the UV projection, and the consistency assertion in Sec. 3.2.1 that surface-binding and rotation alignment ensure unique UV coordinates is a geometric assumption that may be questionable for off-surface ray–Gaussian intersections, but this is a correctness/precision concern, not circularity: the paper does not define 'sharp texture' in terms of the constraint, nor does it fit a parameter to force the claim. Several prior works by the same authors are cited (Texture-GS [33], MeGA [30], NPAVA [29]), but they are used as building blocks or as comparison baselines, not as the sole justification for the central claim. No uniqueness theorem is imported from these citations, and no ansatz is smuggled in via citation. The Limitations section honestly acknowledges that non-Lambertian regions (e.g., hair) are not accurately modeled by surface Gaussians with explicit textures, which further indicates the editing capability is not being asserted vacuously. Overall, no step reduces by construction to its own inputs.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the accuracy of FLAME tracking and the validity of the projection-based UV mapping. The loss weights and thresholds are hand-chosen hyperparameters. No new physical or conceptual entities are introduced beyond the distinction between surface and volumetric Gaussians, which is an algorithmic design rather than an invented mechanism.

free parameters (4)
  • lambda_diff_rgb = not reported
    Weight for the diffuse-only photometric loss in the surface Gaussian stage (Eq. 13). Chosen by hand; no sensitivity analysis.
  • lambda_scale, lambda_pos, lambda_a = not reported
    Weights for the scaling, position, and alpha regularization losses in Eq. 16. Chosen by hand.
  • epsilon_scale = 0.6
    Threshold in the scaling loss (Eq. 12) to avoid too large Gaussians. Set manually.
  • epsilon_pos = 1.0
    Threshold in the position loss (Eq. 14) to limit volumetric Gaussian drift. Set manually.
assumptions (4)
  • domain assumption FLAME mesh tracking provides a sufficiently accurate mesh for each training frame.
    All Gaussians are bound to the tracked FLAME mesh (Sec. 3.2.1, Fig. 1). If the mesh is wrong, the projection-based UV mapping and the animation-based editing fail.
  • domain assumption Projecting a ray-Gaussian intersection onto the mesh along the face normal yields a unique and correct UV coordinate.
    The mesh-aware UV mapping in Eq. 6 assumes this projection is a valid bijection. It holds only where the mesh is locally planar and close to the true surface.
  • standard math Depth-ordered alpha blending correctly composites colors from both surface and volumetric Gaussians.
    The hybrid rendering in Sec. 3.2.3 directly adopts the 3DGS alpha-blending model. This is a standard assumption in the splatting literature.
  • ad hoc to paper The hierarchical two-stage optimization with selective freezing avoids under-constrained solutions and is sufficient for disentanglement.
    The ablation studies (Tab. 2) compare the chosen strategy against alternatives on a single subject, but the strategy itself is a design choice without a general theoretical justification.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SVG-Head: Hybrid Surface-Volumetric Gaussians for High-Fidelity Head Reconstruction and Real-Time Editing." pith.science (2026). https://pith.science/paper/Q66G2LWQ

@misc{pith2026250809597,
  author       = {Pith},
  title        = {Pith review of: SVG-Head: Hybrid Surface-Volumetric Gaussians for High-Fidelity Head Reconstruction and Real-Time Editing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q66G2LWQ}},
  note         = {Machine review of arXiv:2508.09597}
}
read the original abstract

Creating high-fidelity and editable head avatars is a pivotal challenge in computer vision and graphics, boosting many AR/VR applications. While recent advancements have achieved photorealistic renderings and plausible animation, head editing, especially real-time appearance editing, remains challenging due to the implicit representation and entangled modeling of the geometry and global appearance. To address this, we propose Surface-Volumetric Gaussian Head Avatar (SVG-Head), a novel hybrid representation that explicitly models the geometry with 3D Gaussians bound on a FLAME mesh and leverages disentangled texture images to capture the global appearance. Technically, it contains two types of Gaussians, in which surface Gaussians explicitly model the appearance of head avatars using learnable texture images, facilitating real-time texture editing, while volumetric Gaussians enhance the reconstruction quality of non-Lambertian regions (e.g., lips and hair). To model the correspondence between 3D world and texture space, we provide a mesh-aware Gaussian UV mapping method, which leverages UV coordinates given by the FLAME mesh to obtain sharp texture images and real-time rendering speed. A hierarchical optimization strategy is further designed to pursue the optimal performance in both reconstruction quality and editing flexibility. Experiments on the NeRSemble dataset show that SVG-Head not only generates high-fidelity rendering results, but also is the first method to obtain explicit texture images for Gaussian head avatars and support real-time appearance editing.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. AvatarBack: Back-Head Generation for Complete 3D Avatars from Front-View Images

    cs.CV 2025-08 conditional novelty 5.0 of 10

    AvatarBack adds a generative back-head prior and a learned spatial alignment to Gaussian-splatting head avatars, improving rear geometry and texture while keeping frontal quality.

Reference graph

Works this paper leans on

38 extracted references · 37 canonical work pages · cited by 1 Pith paper

  1. [1]

    High- fidelity 3d digital human head creation from RGB-D selfies

    Linchao Bao, Xiangkai Lin, Yajing Chen, Haoxian Zhang, Sheng Wang, Xuefei Zhe, Di Kang, Haozhi Huang, Xinwei Jiang, Jue Wang, Dong Yu, and Zhengyou Zhang. High- fidelity 3d digital human head creation from RGB-D selfies. ACM Trans. Graph., 41(1):3:1–3:21, 2022. 1, 2

    work page 2022

  2. [2]

    Tim Brooks, Aleksander Holynski, and Alexei A. Efros. In- structpix2pix: Learning to follow image editing instructions. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023, Vancouver, BC, Canada, June 17- 24, 2023, pages 18392–18402. IEEE, 2023. 1

    work page 2023

  3. [3]

    HERA: hybrid explicit representation for ultra-realistic head avatars

    Hongrui Cai, Yuting Xiao, Xuan Wang, Jiafei Li, Yudong Guo, Yanbo Fan, Shenghua Gao, and Juyong Zhang. HERA: hybrid explicit representation for ultra-realistic head avatars. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2025, Nashville, TN, USA, June 11-15, 2025, pages 260–270. Computer Vision Foundation / IEEE,

    work page 2025

  4. [4]

    Digital humans–A new era of film technol- ogy

    Rachel Cogley. Digital humans–A new era of film technol- ogy. PhD thesis, Institute of Art, Design+ Technology, 2024. 1

    work page 2024

  5. [5]

    Huang, and Mike Danielsen

    Yun Fu, Renxiang Li, Thomas S. Huang, and Mike Danielsen. Real-time multimodal human-avatar interaction. IEEE Trans. Circuits Syst. Video Technol. , 18(4):467–477,

    work page

  6. [6]

    Dynamic neural radiance fields for monocular 4d facial avatar reconstruction

    Guy Gafni, Justus Thies, Michael Zollh ¨ofer, and Matthias Nießner. Dynamic neural radiance fields for monocular 4d facial avatar reconstruction. In IEEE Conference on Com- puter Vision and Pattern Recognition, CVPR 2021, virtual, June 19-25, 2021, pages 8649–8658. Computer Vision Foun- dation / IEEE, 2021. 2

    work page 2021

  7. [7]

    Reconstructing personalized se- mantic facial nerf models from monocular video

    Xuan Gao, Chenglai Zhong, Jun Xiang, Yang Hong, Yudong Guo, and Juyong Zhang. Reconstructing personalized se- mantic facial nerf models from monocular video. ACM Trans. Graph., 41(6):200:1–200:12, 2022. 1

    work page 2022

  8. [8]

    Neu- ral head avatars from monocular RGB videos

    Philip-William Grassal, Malte Prinzler, Titus Leistner, Carsten Rother, Matthias Nießner, and Justus Thies. Neu- ral head avatars from monocular RGB videos. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022 , pages 18632–18643. IEEE, 2022. 2

    work page 2022

Show all 38 references
  1. [9]

    Collabovr: A re- configurable framework for creative collaboration in virtual reality

    Zhenyi He, Ruofei Du, and Ken Perlin. Collabovr: A re- configurable framework for creative collaboration in virtual reality. In 2020 IEEE International Symposium on Mixed and Augmented Reality, ISMAR 2020, Recife/Porto de Gal- inhas, Brazil, November 9-13, 2020, pages 542–554, Brazil,

    2020

  2. [10]

    Automated avatar creation for 3d games

    Andrew Hogue, Sunbir Gill, and Michael Jenkin. Automated avatar creation for 3d games. In Proceedings of the 2007 conference on Future Play, Toronto, ON, Canada, November 15 - 17, 2007, pages 174–180. ACM, 2007. 2

    2007

  3. [11]

    Avatar digitization from a single image for real-time rendering

    Liwen Hu, Shunsuke Saito, Lingyu Wei, Koki Nagano, Jae- woo Seo, Jens Fursund, Iman Sadeghi, Carrie Sun, Yen- Chun Chen, and Hao Li. Avatar digitization from a single image for real-time rendering. ACM Trans. Graph., 36(6): 195:1–195:14, 2017. 1, 2

    2017

  4. [12]

    Dynamic 3d avatar creation from hand-held video input

    Alexandru Eugen Ichim, Sofien Bouaziz, and Mark Pauly. Dynamic 3d avatar creation from hand-held video input. ACM Trans. Graph., 34(4):45:1–45:14, 2015. 1, 2

    2015

  5. [13]

    Sphere face model: A 3d morphable model with hypersphere manifold latent space using joint 2d/3d training

    Diqiong Jiang, Yiwei Jin, Fang-Lue Zhang, Zhe Zhu, Yun Zhang, Ruofeng Tong, and Min Tang. Sphere face model: A 3d morphable model with hypersphere manifold latent space using joint 2d/3d training. Comput. Vis. Media , 9(2):279– 296, 2023. 2

    2023

  6. [14]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139:1– 139:14, 2023. 1, 2, 3, 5

    2023

  7. [15]

    Nersemble: Multi-view ra- diance field reconstruction of human heads

    Tobias Kirschstein, Shenhan Qian, Simon Giebenhain, Tim Walter, and Matthias Nießner. Nersemble: Multi-view ra- diance field reconstruction of human heads. ACM Trans. Graph., 42(4):161:1–161:14, 2023. 5

    2023

  8. [16]

    Tseng, Zon-Yin Shae, Ferdinand Hendriks, and Tsuhan Chen

    Wing Ho Leung, Belle L. Tseng, Zon-Yin Shae, Ferdinand Hendriks, and Tsuhan Chen. Realistic video avatar. In 2000 IEEE International Conference on Multimedia and Expo, ICME 2000, New York, NY, USA, July 30 - August 2, 2000 , pages 631–634. IEEE Computer Society, 2000. 2

    2000

  9. [17]

    Black, Hao Li, and Javier Romero

    Tianye Li, Timo Bolkart, Michael J. Black, Hao Li, and Javier Romero. Learning a model of facial shape and ex- pression from 4d scans. ACM Trans. Graph., 36(6):194:1– 194:17, 2017. 1, 2

    2017

  10. [18]

    Gaussianavatar-editor: Pho- torealistic animatable gaussian head avatar editor

    Xiangyue Liu, Kunming Luo, Heng Li, Qi Zhang, Yuan Liu, Li Yi, and Ping Tan. Gaussianavatar-editor: Pho- torealistic animatable gaussian head avatar editor. CoRR, abs/2501.09978, 2025. 1, 4, 6

    2025 arXiv

  11. [19]

    Saragih, Tomas Simon, and Yaser Sheikh

    Stephen Lombardi, Jason M. Saragih, Tomas Simon, and Yaser Sheikh. Deep appearance models for face rendering. ACM Trans. Graph., 37(4):68, 2018. 1, 2

    2018

  12. [20]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis. In Computer Vision - ECCV 2020 - 16th European Conference, Glasgow, UK, August 23-28, 2020, Proceed...

    2020

  13. [21]

    Davidson, Sameh Khamis, Ming- song Dou, Vladimir Tankovich, Charles T

    Sergio Orts-Escolano, Christoph Rhemann, Sean Ryan Fanello, Wayne Chang, Adarsh Kowdle, Yury Degtyarev, David Kim, Philip L. Davidson, Sameh Khamis, Ming- song Dou, Vladimir Tankovich, Charles T. Loop, Qin Cai, Philip A. Chou, Sarah Mennicken, Julien P. C. Valentin, Vivek Prad...

    2016

  14. [22]

    Vhap: Versatile head alignment with adaptive appearance priors, 2024

    Shenhan Qian. Vhap: Versatile head alignment with adaptive appearance priors, 2024. 2, 7

    2024

  15. [23]

    Gaus- sianavatars: Photorealistic head avatars with rigged 3d gaus- sians

    Shenhan Qian, Tobias Kirschstein, Liam Schoneveld, Davide Davoli, Simon Giebenhain, and Matthias Nießner. Gaus- sianavatars: Photorealistic head avatars with rigged 3d gaus- sians. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024, Seattle, WA, USA, ...

    2024

  16. [24]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022 , pages 10674–...

    2022

  17. [25]

    HAHA: highly articulated gaussian hu- man avatars with textured mesh prior

    David Svitov, Pietro Morerio, Lourdes Agapito, and Alessio Del Bue. HAHA: highly articulated gaussian hu- man avatars with textured mesh prior. In Computer Vision - ACCV 2024 - 17th Asian Conference on Computer Vision, Hanoi, Vietnam, December 8-12, 2024, Proceedings, Part IX,...

    2024

  18. [26]

    De- ferred neural rendering: image synthesis using neural tex- tures

    Justus Thies, Michael Zollh ¨ofer, and Matthias Nießner. De- ferred neural rendering: image synthesis using neural tex- tures. ACM Trans. Graph., 38(4):66:1–66:12, 2019. 1

    2019

  19. [27]

    My avatar, my self: Identity in video role- playing games

    Zach Waggoner. My avatar, my self: Identity in video role- playing games. McFarland, ., 2009. 1

    2009

  20. [28]

    Binglun Wang, Niladri Shekhar Dutt, and Niloy J. Mitra. Proteusnerf: Fast lightweight nerf editing using 3d-aware image context. Proc. ACM Comput. Graph. Interact. Tech., 7(1):22:1–22:17, 2024. 2

    2024

  21. [29]

    Neural point-based volumetric avatar: Surface-guided neural points for efficient and photo- realistic volumetric head avatar

    Cong Wang, Di Kang, Yan-Pei Cao, Linchao Bao, Ying Shan, and Song-Hai Zhang. Neural point-based volumetric avatar: Surface-guided neural points for efficient and photo- realistic volumetric head avatar. In SIGGRAPH Asia 2023 Conference Papers, SA 2023, Sydney, NSW, Australia, ...

    2023

  22. [30]

    Mega: Hybrid mesh-gaussian head avatar for high-fidelity rendering and head editing

    Cong Wang, Di Kang, Heyi Sun, Shen-Han Qian, Zi-Xuan Wang, Linchao Bao, and Song-Hai Zhang. Mega: Hybrid mesh-gaussian head avatar for high-fidelity rendering and head editing. CoRR, abs/2404.19026, 2024. 1, 2, 5, 6

    2024

  23. [31]

    Recent advances in 3d gaussian splatting

    Tong Wu, Yu-Jie Yuan, Ling-Xiao Zhang, Jie Yang, Yan- Pei Cao, Ling-Qi Yan, and Lin Gao. Recent advances in 3d gaussian splatting. Comput. Vis. Media, 10(4):613–642,

  24. [32]

    Neutex: Neural texture mapping for volumetric neural rendering

    Fanbo Xiang, Zexiang Xu, Milos Hasan, Yannick Hold- Geoffroy, Kalyan Sunkavalli, and Hao Su. Neutex: Neural texture mapping for volumetric neural rendering. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2021, virtual, June 19-25, 2021 , pages 7119–7128. C...

    2021

  25. [33]

    Texture-gs: Disentangling the geome- try and texture for 3d gaussian splatting editing

    Tian-Xing Xu, Wenbo Hu, Yu-Kun Lai, Ying Shan, and Song-Hai Zhang. Texture-gs: Disentangling the geome- try and texture for 3d gaussian splatting editing. In Com- puter Vision - ECCV 2024 - 18th European Conference, Mi- lan, Italy, September 29-October 4, 2024, Proceedings, Pa...

    2024

  26. [34]

    Gaussian head avatar: Ultra high-fidelity head avatar via dynamic gaussians

    Yuelang Xu, Bengwang Chen, Zhe Li, Hongwen Zhang, Lizhen Wang, Zerong Zheng, and Yebin Liu. Gaussian head avatar: Ultra high-fidelity head avatar via dynamic gaussians. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024, Seattle, WA, USA, June 16-22, ...

    2024

  27. [35]

    General neural gauge fields

    Fangneng Zhan, Lingjie Liu, Adam Kortylewski, and Chris- tian Theobalt. General neural gauge fields. In The Eleventh International Conference on Learning Representa- tions, ICLR 2023, Kigali, Rwanda, May 1-5, 2023. OpenRe- view.net, 2023. 2

    2023

  28. [36]

    Film and television animation pro- duction technology based on expression transfer and virtual digital human

    Ning Zhang and Belei Pu. Film and television animation pro- duction technology based on expression transfer and virtual digital human. Scalable Comput. Pract. Exp., 25(6), 2024. 1

    2024

  29. [37]

    B¨uhler, Xu Chen, Michael J

    Yufeng Zheng, Victoria Fern ´andez Abrevaya, Marcel C. B¨uhler, Xu Chen, Michael J. Black, and Otmar Hilliges. I M avatar: Implicit morphable head avatars from videos. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18- ...

    2022

  30. [38]

    Black, and Otmar Hilliges

    Yufeng Zheng, Wang Yifan, Gordon Wetzstein, Michael J. Black, and Otmar Hilliges. Pointavatar: Deformable point- based head avatars from videos. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023, Vancouver, BC, Canada, June 17-24, 2023 , pages 21057–...

    2023

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.