Pith. sign in

REVIEW 3 major objections 7 minor 51 references

GASP: Gaussian Avatars with Synthetic Priors

T0 review · 3 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read By training a Gaussian avatar prior entirely on synthetic faces, GASP fits a single webcam photo or short monocular video into a high-quality, animatable avatar that renders 360 degrees in real time at 70fps.

desk verdict Solid system paper with a real new capability; the 360-degree claim is softer than the metrics prove. read the letter →

arxiv 2412.07739 v1 pith:FBI2EN4M submitted 2024-12-10 cs.CV cs.AIcs.GR

Jack Saunders , Charlie Hewitt , Yanan Jian , Marek Kowalski , Tadas Baltrusaitis , Yiye Chen , Darren Cosker , Virginia Estellers
show 3 more authors
Nicholas Gyde Vinay P. Namboodiri Benjamin E Lundell
This is my paper · ORCID
classification cs.CVcs.AIcs.GR
keywords 3DGaussianSplattinganimatableavatarssyntheticdatapriorsingle-imageavatarfittingfree-viewpointrenderingauto-decodergenerativemodelmonocularvideoper-Gaussianfeatures
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

This paper aims to show that the ill-posed task of building a photorealistic, fully rotatable head avatar from a single photo or a short monocular video can be solved with a generative prior trained entirely on synthetic data. The authors train an auto-decoder over mesh-attached Gaussian avatars on one thousand synthetic identities with pixel-perfect annotations, then fit this prior to a new user in three stages: latent inversion, decoder fine-tuning, and final Gaussian refinement. The key mechanism is per-Gaussian semantic feature vectors that, when frozen during fitting, propagate observed appearance attributes such as hair colour from visible regions to unseen ones like the back of the head. The paper reports high-quality, animatable avatars that render at 70 fps on a consumer GPU and outperform existing single-camera and few-shot methods on the NeRSemble benchmark. The reason to care is that it makes consumer-grade 360-degree avatars practical from the data a webcam or smartphone can provide.

What carries the argument

The central object is an auto-decoder prior over mesh-attached 3D Gaussian avatars. Each Gaussian carries a learnable semantic feature vector $f_i \in \mathbb{R}^8$, and a decoder $D$ maps this feature together with a per-subject identity code $z_j \in \mathbb{R}^{512}$ to offsets from a learned Canonical Gaussian Template $C$, giving $A_{i,j} = C_i + D(f_i, z_j)$. The 8-dimensional features are the load-bearing mechanism: because the decoder learns to associate semantically similar Gaussians with similar attributes, freezing the features during fitting makes a change observed at a visible Gaussian (for example a blond front hairstyle) propagate automatically to unseen Gaussians at the back of the head. The three-stage fitting procedure, with an $L_{\text{prior}}$ regularizer during the later stages, keeps this propagation within a plausible regime while still adapting to the real person's identity. Rendering uses standard 3D Gaussian Splatting, so the fitted avatar is just a static set of Gaussian attributes plus mesh bindings and needs no network at inference time.

What would settle it

Fit avatars to single frontal images of 100 real identities never used in training, render each from a directly behind viewpoint, and ask human raters to classify each back-of-head rendering as a real capture or a synthetic render; the central claim of 360-degree quality holds only if the synthetic-detection rate stays near chance.

Watch

Extended reading notes

Core claim

The central claim is that a Gaussian avatar prior trained purely on synthetic data is enough to bridge the synthetic-to-real domain gap when combined with semantic per-Gaussian features and a staged fitting procedure. GASP defines a canonical Gaussian template plus a decoder that maps an 8-dimensional per-Gaussian semantic feature and a 512-dimensional identity code to per-person offsets, producing a mesh-attached Gaussian avatar that can be posed with a 3D morphable model. Fitting a new identity proceeds in three stages: optimizing only the identity code to stay inside the prior, fine-tuning the decoder while freezing the semantic features so that observed attributes propagate to unseen Gaussians, and refining all Gaussians under a prior-regularization loss. The paper argues that this yields high-quality 360-degree renderings from a single image or monocular video, with inference requiring no neural networks at all.

Load-bearing premise

The prior trained on 1000 synthetic identities rendered under uniform white lighting is representative enough of real human heads that, when fit to a single real photo, it can correctly fill in never-seen regions such as the back of the head without looking fake.

Editorial extensions

If this is right

  • With only a webcam or smartphone photo or short monocular video, a user can obtain a photorealistic, animatable avatar that renders from any viewpoint in real time, removing the need for a multi-camera capture rig.
  • The fitted avatar is stored as an approximately 15MB file of Gaussian attributes; no neural networks are needed at inference, and rendering runs at 70fps on a consumer GPU while posing can run at 67fps on a CPU.
  • Because the prior is trained exclusively on synthetic data, user enrollment carries no risk of dataset-distillation privacy attacks that could expose real identities from the prior.
  • The prior's latent space is semantically controllable: linear directions found by an SVM can edit attributes such as age, facial hair, and hair length.
  • When 16 synchronized cameras are available, the model stays competitive with state-of-the-art multi-camera avatars while converging in fewer steps, so the prior does not hurt as data increases.

Reading between the lines

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

  • The ablation trend (a one-subject prior hurts, a 1000-subject prior helps) suggests the method's quality scales with synthetic dataset size; a direct test would be training the same prior on tens of thousands of synthetic identities and measuring whether the residual 'synthetic-looking' back of the head disappears.
  • Because all evaluations use frontal frames from the NeRSemble rig, the boldest claim—that a casual in-the-wild webcam selfie works—remains untested; fitting on diverse, cluttered, differently lit real photos would be a sharp stress test of the domain-gap bridge.
  • Freezing the semantic features implies that any attribute with a learned semantic correlate (skin tone, hairstyle, facial hair, accessories) should transfer from visible to hidden regions; this could be verified by changing one visible attribute and checking that only semantically matching unseen Gaussians change.
  • The uniform white lighting used in synthetic training means the fitted avatar cannot generalize to novel illumination; incorporating varied lighting into the synthetic pipeline, which the paper names as future work, would be the natural step toward relightable single-image avatars.
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 / 7 minor

Summary. The paper proposes GASP, a method for creating animatable Gaussian avatars from a single image or short monocular video by leveraging a generative prior trained on 1000 synthetic identities. The prior is an auto-decoder over per-Gaussian semantic features and a canonical template; fitting proceeds in three stages (latent inversion, MLP fine-tuning, and Gaussian refinement). The method is evaluated on NeRSemble in monocular, single-image, and multi-camera settings against FlashAvatar, GaussianAvatars, ROME, and DiffusionRig, with ablations on the prior size, number of Gaussians, and fitting stages. The authors claim high-quality 360-degree rendering at 70 fps on commercial hardware with no neural network required at inference.

Significance. The main contribution is a coherent and practical pipeline: a synthetic-data prior over Gaussian avatar parameters, learnable per-Gaussian semantic features, and a three-stage fitting process that is well motivated and ablated. The paper's strengths include a quantitative evaluation across several settings, ablations that confirm the prior's role, an explicit user study, and a real-time rendering result (70 fps, no network at inference). If the 360-degree rendering claim were fully supported, this would be a meaningful step toward consumer-grade avatars. However, the central claim of high-quality 360-degree rendering is not yet quantitatively established, because the evaluation does not measure performance on the truly unseen back-of-head region that motivates the prior.

major comments (3)
  1. [Sec. 6, Tables 1–3; Sec. 7] The central claim of high-quality 360-degree rendering from limited data is not quantitatively tested on genuinely unseen back-of-head regions. All numeric metrics are computed on the four 'most extreme view' cameras selected from the NeRSemble ring (Table 4), and the paper states in the supplementary (Fig. 11) that the back of the head is never included in the fitting data. No ground-truth back-of-head view appears in Tables 1–3 or in the user study (App. F). The limitation in Sec. 7 concedes that the back of the head produces synthetic-looking results. Consequently, the current evidence establishes improved novel-view synthesis within the observed frontal hemisphere, not the advertised full 360-degree quality. The authors should add quantitative evaluation on true back-of-head views if the NeRSemble rig provides them, or add an equivalent transfer experiment with held-out synthetic subjects with known back-of-head ground truth, or they should revise the central claim to something like 'high-quality rendering within the observed hemisphere plus plausible hallucination of the back of the head.'
  2. [Sec. 6, Table 4] The evaluation cameras are chosen as 'the four most extreme view cameras, as determined by manual inspection.' With only eight test subjects, this ad hoc selection rule raises a risk of selection bias and makes the aggregate metrics hard to reproduce. The paper should either define an automatic angular-distance criterion for selecting test cameras or report metrics on all available test cameras, and it should provide per-subject results with confidence intervals.
  3. [App. F, Tables 1–2] The user study is underpowered for the claims made: 40 Mechanical Turk users each rated a single randomly assigned subject, yielding roughly five observations per subject, and only mean QUAL scores are reported without variance or significance tests. The statement in Sec. 6.2 that the method 'significantly outperforms' on user-perceived quality is therefore not statistically supported. Please report per-subject QUAL scores, confidence intervals, and a paired significance test, or justify why the existing sample size is sufficient for the claim.
minor comments (7)
  1. [Eq. (6)] The regularizer is written as Lreg = λσ||max(0.6, σ′)||2 + λµ||µ′||2; as written this penalizes all scale values, not only large ones. Please clarify whether a hinge term such as max(0, σ′ − 0.6) was intended, and specify whether the max is applied elementwise.
  2. [Sec. 5, App. G] The number of Gaussians is reported as 187,779 in Sec. 5 and 187,776 in App. G; please make these numbers consistent.
  3. [App. C] The text says 'f and ‡ are concatenated'; the symbol '‡' appears to be a typo for the identity code z_j.
  4. [References] Reference [5] has a garbled author string ('Ding, Cecilia, Zheng ans Zhang, Zhihao Xia' should likely be 'Ding, Zheng, and Zhang'); please correct the author list.
  5. [App. H] The sentence 'There is no chance of a subject withdrawing consent and requiring our prior to be retrained or detained' contains the non-word 'detained'; this should likely be 'deleted' or 'removed.'
  6. [Table 1 caption] The caption reads 'for the Monocular scenario we on a single camera'; a verb appears to be missing, likely 'we train on a single camera.'
  7. [Sec. 4] The virtual camera is sampled with azimuth in [-180, +180] degrees and elevation in [-20, +45] degrees, which describes a spherical band or cap rather than a hemisphere; please clarify the intended coverage and how it relates to the 360-degree rendering claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the synthetic prior is independently trained, fitted on enrollment views, and evaluated against external baselines on held-out real views; back-of-head limitations are coverage gaps, not circular reductions.

full rationale

The paper's derivation chain is: (i) train an autodecoder prior D on 1000 synthetic identities using Eqs. (4)-(6); (ii) fit a user's identity latent z with D frozen, then fine-tune D, then refine Gaussians against the enrollment camera(s) (Sec. 3.5); (iii) evaluate on NeRSemble at held-out extreme camera views, reporting PSNR/SSIM/LPIPS/FID/QUAL against external baselines (Tab. 1). No step defines a predicted quantity in terms of the fitted quantity: the prior never sees the real enrollment images during training, the evaluation views are different cameras from the training view, and the metrics compare against ground-truth frames from those views. The self-citations (Hewitt et al. [12], Wood et al. [38]) provide the synthetic data generator and 3DMM fitting tool, not the paper's conclusions, and the central result is benchmarked on an external real dataset. The paper's own limitation that 'For some regions, such as the back of the head, the model produces synthetic-looking results' (Sec. 7) and the supplementary statement that 'the back of the head is never included in the fitting data' identify an evaluation-coverage gap, not an equivalence between output and input. Therefore no circular step is present; the score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 1 invented entities

The claims rest on several hand-chosen hyperparameters and domain assumptions. The central domain assumption is the transferability of a synthetic prior to real humans; the evaluation also depends on manually selected test cameras. No new physical entities are introduced; the per-Gaussian feature vector is a learned latent code with supporting ablations.

free parameters (5)
  • Loss weights = λpix, λα, λpercep, λσ, λμ, λ1, λSSIM, λprior (values not reported)
    The final loss is a weighted sum of pixel, alpha, perceptual, and regularization terms (Eq. 5-6); the weights are hand-chosen and not listed in the paper.
  • Scalp regularization factor = λμ reduced by factor of 100 for scalp-region Gaussians
    A manually defined scalp region and a hand-chosen adjustment to let Gaussians model coarse hair (Sec. 3.3).
  • Latent and feature dimensions = z ∈ R512, f ∈ R8
    Architectural choices not derived from the data; they affect model capacity and are not ablated in the paper.
  • UV map resolution / number of Gaussians = 512×512 UV map -> 187,779 Gaussians
    Chosen resolution; ablation shows higher resolution is better, but the exact choice is a free parameter.
  • Number of training identities = 1000 synthetic subjects
    Ablation shows quality improves with more subjects; the specific number is a resource and diversity choice.
assumptions (4)
  • domain assumption The synthetic data distribution (1000 identities, uniform white lighting) is sufficiently close to real human heads that a prior trained on it transfers to real subjects.
    Sec. 3.3 and Sec. 4: the entire method relies on crossing the synthetic-to-real domain gap; the paper's limitations admit the back of the head looks synthetic (Sec. 7).
  • domain assumption The 3D morphable model (FLAME / Wood et al.) and the mesh-attached Gaussian formulation can represent hair and fine appearance, given the reduced scalp regularization.
    Sec. 3.3-3.4: the model binds Gaussians to a 3DMM mesh and must model coarse hair with those Gaussians; the manual scalp regularization is an adjustment to make this work.
  • ad hoc to paper The four 'most extreme view' cameras chosen by manual inspection are representative of true novel viewpoints for evaluation.
    Sec. 6 and Appendix D, Table 4: the evaluation cameras are selected by hand; this choice is not justified quantitatively and could bias the novel-view results.
  • domain assumption The FREE sequence is an unseen and representative test of expression and view generalization.
    Sec. 6 and Appendix D: only one test sequence is used; its representativeness is assumed rather than demonstrated.
invented entities (1)
  • Per-Gaussian semantic feature vector f_i independent evidence
    purpose: Learned latent code for each Gaussian that encodes semantic correlations, so that updating visible Gaussians during fitting also updates unseen ones (e.g., hair at the back of the head).
    The paper provides supporting evidence: PCA decomposition shows semantic structure (Fig. 4), and ablations of fitting stages show the features contribute to quality (Tab. 3). It is a learned representation, not an independently measurable physical entity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GASP: Gaussian Avatars with Synthetic Priors." pith.science (2026). https://pith.science/paper/FBI2EN4M

@misc{pith2026241207739,
  author       = {Pith},
  title        = {Pith review of: GASP: Gaussian Avatars with Synthetic Priors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FBI2EN4M}},
  note         = {Machine review of arXiv:2412.07739}
}
abstract

Gaussian Splatting has changed the game for real-time photo-realistic rendering. One of the most popular applications of Gaussian Splatting is to create animatable avatars, known as Gaussian Avatars. Recent works have pushed the boundaries of quality and rendering efficiency but suffer from two main limitations. Either they require expensive multi-camera rigs to produce avatars with free-view rendering, or they can be trained with a single camera but only rendered at high quality from this fixed viewpoint. An ideal model would be trained using a short monocular video or image from available hardware, such as a webcam, and rendered from any view. To this end, we propose GASP: Gaussian Avatars with Synthetic Priors. To overcome the limitations of existing datasets, we exploit the pixel-perfect nature of synthetic data to train a Gaussian Avatar prior. By fitting this prior model to a single photo or video and fine-tuning it, we get a high-quality Gaussian Avatar, which supports 360$^\circ$ rendering. Our prior is only required for fitting, not inference, enabling real-time application. Through our method, we obtain high-quality, animatable Avatars from limited data which can be animated and rendered at 70fps on commercial hardware. See our project page (https://microsoft.github.io/GASP/) for results.

Figures

Figures reproduced from arXiv: 2412.07739 by the authors.

Figure 1
Figure 1. We propose GASP, a novel model for creating photorealistic, realtime, animatable, 360◦ avatars from easily-captured data. We train a generative prior model of Gaussian Avatars on Synthetic data. The prior allows our model to be fit using a single image or a short video with the prior accounting for the unseen views. This lets users create their avatar with only a webcam or smartphone. Abstract Gaussian Splatting has… view at source ↗
Figure 2
Figure 2. The overview of our model. In the first stage, we train an autodecoder prior model on Synthetic data to predict the parameters of [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The architecture of our prior model. A latent vector [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Visualization of the first three components of a PCA [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Examples showing how the three stages in our fitting [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 7
Figure 7. Figure 7: Qualitative comparisons of our method with existing state-of-the-art in the [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Self/Cross Reenactment: We show examples of our model for self-reenactment (top) and cross-identity (bottom). The model is fit using a frontal view video only (frame with a gray background). Despite never seeing the back of a real person’s head, we still obtain good-qu…
Figure 9
Figure 9. Figure 9: Qualitative comparisons of our method with existing [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Additional Cross Reenactment Results. We show several more examples of cross-reenactment. We use the input image on the left to drive the avatars on the right. Each Avatar is trained in the Monocular Setting. 4-lips, EXP-5-mouth, EXP-6-tongue-1, EXP-7-tongue-2, EXP-8-…
Figure 11
Figure 11. Figure 11: Additional Self Reenactment Results. We show several more examples of self-reenactment with 360◦ rendering. We show models fit to a single image (Top), a monocular video (Middle) and multiple views (Bottom). In each case, the back of the head is never included in the …
Figure 12
Figure 12. Figure 12: We demonstrate that the latent space learned by our prior model is controllable by finding directions in it that correspond to [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]
Figure 13
Figure 13. Figure 13: The architecture of our MLP decoder D. f and ‡ are concatenated and passed through 6 linear layers with output size 256. The network then splits into per-attribute branches. Each block represents a linear layer followed by ReLU and using weight normalization. ing sett…
Figure 14
Figure 14. Figure 14: Ablations: We show the qualitative effect of using differing numbers of subjects to train the prior (top) and different numbers of Gaussians (bottom) [PITH_FULL_IMAGE:figures/full_fig_p015_14.png]
Figure 16
Figure 16. Figure 16: The training loss curves for λpixLpix + λpercepLpercep with (blue) and without (orange) the canonical Gaussians. Note the improved training stability and better overall loss. allowing a single account to operate an avatar. Before de￾ploying any avatar system using our…
Figure 15
Figure 15. Figure 15: Qualitative comparisons of our method with existing [PITH_FULL_IMAGE:figures/full_fig_p015_15.png]
Figure 17
Figure 17. Figure 17: A comparison of our method (Right) compared to [PITH_FULL_IMAGE:figures/full_fig_p016_17.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 43 canonical work pages

  1. [1]

    A Morphable Model For The Synthesis Of 3D Faces

    V olker Blanz and Thomas Vetter. A Morphable Model For The Synthesis Of 3D Faces. Association for Computing Ma- chinery, New York, NY , USA, 1 edition, 2023. 2

    work page 2023

  2. [2]

    Marcel C. Buehler, Gengyan Li, Erroll Wood, Leonhard Helminger, Xu Chen, Tanmay Shah, Daoye Wang, Stephan Garbin, Sergio Orts-Escolano, Otmar Hilliges, Dmitry La- gun, J´er´emy Riviere, Paulo Gotardo, Thabo Beeler, Abhim- itra Meka, and Kripasindhu Sarkar. Cafca: High-quality novel view synthesis of expressive faces from casual few- shot captures. In ACM ...

    work page 2024

  3. [3]

    Preface: A data-driven volumetric prior for few-shot ultra high-resolution face synthesis

    Marcel C B ¨uhler, Kripasindhu Sarkar, Tanmay Shah, Gengyan Li, Daoye Wang, Leonhard Helminger, Ser- gio Orts-Escolano, Dmitry Lagun, Otmar Hilliges, Thabo Beeler, et al. Preface: A data-driven volumetric prior for few-shot ultra high-resolution face synthesis. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 3402–3413, 2...

    work page 2023

  4. [4]

    Monogaus- sianavatar: Monocular gaussian point-based head avatar

    Yufan Chen, Lizhen Wang, Qijing Li, Hongjiang Xiao, Shengping Zhang, Hongxun Yao, and Yebin Liu. Monogaus- sianavatar: Monocular gaussian point-based head avatar. arXiv, 2023. 1, 2

    work page 2023

  5. [5]

    Diffusionrig: Learning personalized priors for facial appearance editing

    Cecilia Ding, Zheng ans Zhang, Zhihao Xia, Lars Jebe, Zhuowen Tu, and Xiuming Zhang. Diffusionrig: Learning personalized priors for facial appearance editing. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023. 3, 6, 7, 8

    work page 2023

  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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8649–8658, 2021. 1

    work page 2021

  7. [7]

    Learning neural parametric head models

    Simon Giebenhain, Tobias Kirschstein, Markos Georgopou- los, Martin R ¨unz, Lourdes Agapito, and Matthias Nießner. Learning neural parametric head models. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR),

  8. [8]

    Npga: Neural paramet- ric gaussian avatars

    Simon Giebenhain, Tobias Kirschstein, Martin R ¨unz, Lour- des Agapito, and Matthias Nießner. Npga: Neural paramet- ric gaussian avatars. In SIGGRAPH Asia 2024 Conference Papers (SA Conference Papers ’24), December 3-6, Tokyo, Japan, 2024. 1, 2

    work page 2024

Show all 51 references
  1. [9]

    Ad-nerf: Audio driven neural ra- diance fields for talking head synthesis

    Yudong Guo, Keyu Chen, Sen Liang, Yongjin Liu, Hujun Bao, and Juyong Zhang. Ad-nerf: Audio driven neural ra- diance fields for talking head synthesis. In IEEE/CVF Inter- national Conference on Computer Vision (ICCV) , 2021. 1, 2

    2021

  2. [10]

    Hearst, Susan T Dumais, Edgar Osuna, John Platt, and Bernhard Scholkopf

    Marti A. Hearst, Susan T Dumais, Edgar Osuna, John Platt, and Bernhard Scholkopf. Support vector machines.IEEE In- telligent Systems and their applications, 13(4):18–28, 1998. 1

    1998

  3. [11]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium. In Proceedings of the 31st International Conference on Neural Information Processing Systems , page 6...

    2017

  4. [12]

    Look ma, no markers: holistic perfor- mance capture without the hassle

    Charlie Hewitt, Fatemeh Saleh, Sadegh Aliakbarian, Lohit Petikam, Shideh Rezaeifar, Louis Florentin, Zafiirah Hose- nie, Thomas J Cashman, Julien Valentin, Darren Cosker, and Tadas Baltruˇsaitis. Look ma, no markers: holistic perfor- mance capture without the hassle. ACM Trans...

    2024

  5. [13]

    Denoising diffu- sion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models. arXiv preprint arxiv:2006.11239,

    2006 arXiv

  6. [14]

    Analyzing and improving the image quality of StyleGAN

    Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Analyzing and improving the image quality of StyleGAN. In Proc. CVPR, 2020. 3

    2020

  7. [15]

    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 Transactions on Graphics, 42 (4), 2023. 1, 2, 3, 4, 5

    2023

  8. [16]

    Realistic one-shot mesh-based head avatars

    Taras Khakhulin, Vanessa Sklyarova, Victor Lempitsky, and Egor Zakharov. Realistic one-shot mesh-based head avatars. In European Conference of Computer vision (ECCV), 2022. 3, 7, 8, 4

    2022

  9. [17]

    Deep video portraits

    Hyeongwoo Kim, Pablo Garrido, Ayush Tewari, Weipeng Xu, Justus Thies, Matthias Nießner, Patrick P´erez, Christian Richardt, Michael Zoll ¨ofer, and Christian Theobalt. Deep video portraits. ACM Transactions on Graphics (TOG) , 37 (4):163, 2018. 2

    2018

  10. [18]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. CoRR, abs/1412.6980, 2014. 5

    2014 arXiv

  11. [19]

    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), 2023. 1, 2, 5

    2023

  12. [20]

    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 Transactions on Graphics, (Proc. SIGGRAPH Asia), 36(6):194:1–194:17, 2017. 2

    2017

  13. [21]

    Robust high-resolution video matting with tempo- ral guidance, 2021

    Shanchuan Lin, Linjie Yang, Imran Saleemi, and Soumyadip Sengupta. Robust high-resolution video matting with tempo- ral guidance, 2021. 5

    2021

  14. [22]

    One-2-3-45++: Fast single image to 3d objects with consistent multi-view generation and 3d diffusion

    Minghua Liu, Ruoxi Shi, Linghao Chen, Zhuoyang Zhang, Chao Xu, Xinyue Wei, Hansheng Chen, Chong Zeng, Ji- ayuan Gu, and Hao Su. One-2-3-45++: Fast single image to 3d objects with consistent multi-view generation and 3d diffusion. arXiv preprint arXiv:2311.07885, 2023. 8

    2023 arXiv

  15. [23]

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

    Ruoshi Liu, Rundi Wu, Basile Van Hoorick, Pavel Tok- makov, Sergey Zakharov, and Carl V ondrick. Zero-1-to-3: Zero-shot one image to 3d object. 2023 IEEE/CVF Interna- tional Conference on Computer Vision (ICCV), pages 9264– 9275, 2023. 8

    2023

  16. [24]

    S. Ma, T. Simon, J. Saragih, D. Wang, Y . Li, F. La Torre, and Y . Sheikh. Pixel codec avatars. In 2021 IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR) , pages 64–73, Los Alamitos, CA, USA, 2021. IEEE Com- puter Society. 2

    2021

  17. [25]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. NeRF: 9 Representing scenes as neural radiance fields for view syn- thesis. In The European Conference on Computer Vision (ECCV), 2020. 1

    2020

  18. [26]

    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, 2020. 2

    2020

  19. [27]

    Deepsdf: Learning con- tinuous signed distance functions for shape representation

    Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. Deepsdf: Learning con- tinuous signed distance functions for shape representation. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019. 4

    2019

  20. [28]

    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. arXiv preprint arXiv:2312.02069, 2023. 1, 2, 3, 6, 7, 8

    2023 arXiv

  21. [29]

    V orf: V olumetric relightable faces

    Pramod Rao, Mallikarjun B R, Gereon Fox, Tim Weyrich, Bernd Bickel, Hanspeter Pfister, Wojciech Matusik, Ayush Tewari, Christian Theobalt, and Mohamed Elgharib. V orf: V olumetric relightable faces. In33rd British Machine Vision Conference 2022, BMVC 2022, London, UK, November...

    2022

  22. [30]

    High-resolution image syn- thesis with latent diffusion models, 2021

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models, 2021. 8

    2021

  23. [31]

    Dubbing for every- one: Data-efficient visual dubbing using neural rendering priors

    Jack Saunders and Vinay Namboodiri. Dubbing for every- one: Data-efficient visual dubbing using neural rendering priors. arxiv, 2024. 2, 3

    2024

  24. [32]

    Namboodiri

    Jack Saunders and Vinay P. Namboodiri. Read avatars: Re- alistic emotion-controllable audio driven avatars. In arxiv,

  25. [33]

    SplattingAvatar: Realistic Real-Time Human Avatars with Mesh-Embedded Gaussian Splatting

    Zhijing Shao, Zhaolong Wang, Zhuang Li, Duotun Wang, Xiangru Lin, Yu Zhang, Mingming Fan, and Zeyu Wang. SplattingAvatar: Realistic Real-Time Human Avatars with Mesh-Embedded Gaussian Splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...

    2024

  26. [34]

    Stylerig: Rigging style- gan for 3d control over portrait images

    Ayush Tewari, Mohamed Elgharib, Gaurav Bharaj, Flo- rian Bernard, Hans-Peter Seidel, Patrick P ´erez, Michael Zollh¨ofer, and Christian Theobalt. Stylerig: Rigging style- gan for 3d control over portrait images. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recogn...

    2020

  27. [35]

    Face2face: real-time face capture and reenactment of rgb videos

    Justus Thies, Michael Zollh ¨ofer, Marc Stamminger, Chris- tian Theobalt, and Matthias Nießner. Face2face: real-time face capture and reenactment of rgb videos. Commun. ACM, 62(1):96–104, 2018. 2

    2018

  28. [36]

    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), 2019. 2

    2019

  29. [37]

    Neural voice puppetry: Audio-driven facial reenactment

    Justus Thies, Mohamed Elgharib, Ayush Tewari, Christian Theobalt, and Matthias Nießner. Neural voice puppetry: Audio-driven facial reenactment. ECCV 2020, 2020. 2

    2020

  30. [38]

    Fake it till you make it: face analysis in the wild using synthetic data alone

    Erroll Wood, Tadas Baltru ˇsaitis, Charlie Hewitt, Sebastian Dziadzio, Thomas J Cashman, and Jamie Shotton. Fake it till you make it: face analysis in the wild using synthetic data alone. In Proceedings of the IEEE/CVF international conference on computer vision, pages 3681–36...

    2021

  31. [39]

    Multiface: A dataset for neural face rendering

    Cheng-hsin Wuu, Ningyuan Zheng, Scott Ardisson, Rohan Bali, Danielle Belko, Eric Brockmeyer, Lucas Evans, Tim- othy Godisart, Hyowon Ha, Xuhua Huang, Alexander Hy- pes, Taylor Koska, Steven Krenn, Stephen Lombardi, Xi- aomin Luo, Kevyn McPhail, Laura Millerschoen, Michal Perdo...

    2022

  32. [40]

    Flashavatar: High-fidelity head avatar with efficient gaussian embedding

    Jun Xiang, Xuan Gao, Yudong Guo, and Juyong Zhang. Flashavatar: High-fidelity head avatar with efficient gaussian embedding. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2024. 1, 2, 4, 6, 7

    2024

  33. [41]

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

    Yuelang Xu, Benwang Chen, Zhe Li, Hongwen Zhang, Lizhen Wang, Zerong Zheng, and Yebin Liu. Gaussian head avatar: Ultra high-fidelity head avatar via dynamic gaussians. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. 1, 2

    2024

  34. [42]

    3d gaussian parametric head model

    Yuelang Xu, Lizhen Wang, Zerong Zheng, Zhaoqi Su, and Yebin Liu. 3d gaussian parametric head model. In Pro- ceedings of the European Conference on Computer Vision (ECCV), 2024. 2, 4

    2024

  35. [43]

    Geneface: Generalized and high- fidelity audio-driven 3d talking face synthesis.arXiv preprint arXiv:2301.13430, 2023

    Zhenhui Ye, Ziyue Jiang, Yi Ren, Jinglin Liu, Jinzheng He, and Zhou Zhao. Geneface: Generalized and high- fidelity audio-driven 3d talking face synthesis.arXiv preprint arXiv:2301.13430, 2023. 2

    2023 arXiv

  36. [44]

    Bisenet v2: Bilateral net- work with guided aggregation for real-time semantic seg- mentation

    Changqian Yu, Changxin Gao, Jingbo Wang, Gang Yu, Chunhua Shen, and Nong Sang. Bisenet v2: Bilateral net- work with guided aggregation for real-time semantic seg- mentation. Int. J. Comput. Vision, 129(11):3051–3068, 2021. 5

    2021

  37. [45]

    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. In CVPR, 2018. 4, 7

    2018

  38. [46]

    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 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 2

    2023

  39. [47]

    In- stant volumetric head avatars

    Wojciech Zielonka, Timo Bolkart, and Justus Thies. In- stant volumetric head avatars. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4574–4584, 2022. 1, 2 10 GASP: Gaussian A vatars with Synthetic Priors Supplementary Material A. Further Resu...

    2023

  40. [48]

    The SVM here was learned to classify age ≥ 45 into a sepa- rate group from age < 45

    Age - this corresponds to the age of the person whose facial texture was used in the training data sample. The SVM here was learned to classify age ≥ 45 into a sepa- rate group from age < 45

  41. [49]

    Facial hair - here, the SVM was learned to classify sam- ples with facial hair separately from samples with no fa- cial hair

  42. [50]

    The results of the evaluation are shown in the supplemen- tary video as well as in Figure 12, where each column demonstrates one of the features we control

    Head hair - here, the SVM separated samples with long hair from samples with short hair. The results of the evaluation are shown in the supplemen- tary video as well as in Figure 12, where each column demonstrates one of the features we control. Subject Test Cameras Subject Te...

  43. [256]

    Each block represents a linear layer followed by ReLU and using weight normalization

    The network then splits into per-attribute branches. Each block represents a linear layer followed by ReLU and using weight normalization. ing setting (Monocular, Single Frame and Multi-Camera) for an individual subject, meaning a total of 13 images per user (Four methods time...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.