Pith. sign in

REVIEW 4 major objections 7 minor 1 cited by

AvatarMakeup: Realistic Makeup Transfer for 3D Animatable Head Avatars

T0 review · 4 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A single reference photo can transfer makeup onto an animated 3D head avatar without breaking identity or consistency.

desk verdict A sensible new-task paper — makeup transfer to animatable 3D Gaussian avatars — with a plausible pipeline, but the evaluation is too thin to support the state-of-the-art claim. read the letter →

arxiv 2507.02419 v2 pith:Y45SLDTF submitted 2025-07-03 cs.CV

Yiming Zhong , Xiaolin Zhang , Ligang Liu , Yao Zhao , Yunchao Wei This is my paper
classification cs.CV
keywords makeuptransfer3DGaussiansplattinganimatableheadavatarsdiffusionmodelsUVmapidentitypreservationavatarediting
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 to solve makeup transfer for drivable 3D head avatars by using a pretrained 2D makeup diffusion model as supervision and stabilizing its inconsistent outputs. The central idea is a coarse-to-fine pipeline: first, a global UV map, filled by averaging many canonical-expression makeup images generated from different views, is queried to synthesize coherent makeup guidance for any expression and viewpoint; then a refinement stage injects small timestamps into the diffusion model to restore fine detail. If the claim is right, any single reference photo of makeup can be applied to a rigged 3D Gaussian avatar while keeping the makeup fixed to the face during animation and preserving the avatar's identity. This matters because existing 3D editing tools are mostly text-driven, static, or identity-deforming, so they cannot serve as practical makeup customization for virtual avatars.

What carries the argument

Coherent Duplication is the load-bearing mechanism: a global UV map whose color at each UV coordinate is the average of pixel colors from N canonical-expression guidance images produced by Stable-Makeup, mapped through the FLAME mesh into UV space. Because the UV map is constant, querying it with a mesh renderer yields coherent makeup guidance from arbitrary views and expressions, so the avatar can be optimized without inheriting per-view diffusion artifacts. A Detail Refinement module then runs the diffusion model with small timestamps to sharpen details while keeping the coarse base makeup structurally fixed.

What would settle it

Render the target avatar in canonical expression, feed it and a single asymmetric makeup reference (for example, eyeshadow on only one eye) through Stable-Makeup, and inspect the UV map after averaging 16 views: if the asymmetric detail is blurred, displaced, or duplicated onto the wrong eye, the coherent base cannot support the claimed fine-detail preservation.

Watch

Extended reading notes

Core claim

AvatarMakeup establishes that image-conditioned makeup transfer to animatable 3D Gaussian head avatars can be achieved by decoupling consistency from detail: a coherent coarse pass built on a FLAME-bound UV map, followed by a diffusion refinement pass. The paper argues that directly optimizing against per-view diffusion guidance fails because generated images disagree and deform facial interior structure such as teeth; averaging guidance colors in UV space removes that inconsistency, and the refinement module, using small forward-diffusion timestamps, then injects high-frequency makeup detail without breaking the coherent base. The reported result is consistent makeup across multi-view and animation, with the original identity preserved, outperforming the tested baselines.

Load-bearing premise

The pipeline assumes that Stable-Makeup's transferred makeup aligns correctly with the avatar's facial region when the avatar is in its canonical expression, because any misalignment is baked into the UV average and inherited by every later stage.

Editorial extensions

If this is right

  • Makeup customization for avatars becomes reference-image-driven rather than prompt-driven, so a user can specify a look with one photo instead of wording it.
  • Because the coarse stage averages inconsistent diffusion outputs in UV space, even imperfect per-view generation can produce a stable base makeup for animation.
  • The optimization freezes Gaussian geometry and only updates opacity and appearance features, so the avatar's expression rig and geometric identity stay intact during makeup transfer.
  • The two-stage design separates consistency from detail quality, meaning each stage can be improved or replaced independently without collapsing the other.
  • The method provides a practical alternative to text-based 3D head editing for applications like virtual reality, gaming, and personalized avatars.

Reading between the lines

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

  • The UV-averaging principle could plausibly serve as a general consistency adapter for lifting other 2D image editors to 3D head avatars, not just Stable-Makeup; the paper only demonstrates it for makeup transfer.
  • Because the coherent UV map is resolution-limited and the refinement stage re-enters the diffusion model, the final quality may depend on how aggressively the identity-preserving mask hides non-makeup regions; extreme or highly asymmetric makeup would stress the averaging assumption.
  • A natural extension would be editing the UV map directly in UV space using text or reference images, allowing users to modify the coherent layer without regenerating hundreds of per-view diffusion images.
  • The method's reliance on many generated guidance images suggests a testable scalability question: how few views can fill the UV map before consistency degrades, especially for occluded facial regions.
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

4 major / 7 minor

Summary. The paper presents AvatarMakeup, a coarse-to-fine method for transferring makeup from a single reference photo to animatable FLAME-rigged 3D Gaussian avatars. In the coarse stage, called Coherent Duplication, the method renders the avatar at canonical expressions from multiple views, uses Stable-Makeup to generate makeup images for those views, and averages the generated pixels into a global UV map via the mesh renderer's mapping. The UV map is then queried to synthesize consistent guidance images for arbitrary views and expressions, which supervise optimization of the Gaussian features and opacities while the geometry is frozen. In the fine stage, a Detail Refinement module injects small-timestamp noise into Stable-Makeup to generate refined guidance images, and the avatar is further optimized with L1/LPIPS losses on masked makeup regions plus an identity-preservation loss on non-makeup regions. Experiments compare with ClipFace quantitatively and GaussianEditor/TIP-Editor qualitatively on NeRSemble avatars with LADN reference makeups, reporting DINO-I, FID/KID, and a proposed GPT-4o-based AIME metric.

Significance. If the method works as claimed, it would be a practical contribution: it offers an image-conditioned alternative to text-based 3D head editing and appears to achieve plausible makeup transfer with animation consistency. The coarse-to-fine pipeline is intuitive and the use of a UV map to stabilize diffusion-generated guidance is a reasonable idea. However, the evidence provided is not yet sufficient to support the state-of-the-art claim. The evaluation relies on a single baseline, an unvalidated LLM-based metric, and omits quantitative identity-preservation metrics in the main comparison. The central assumption that Stable-Makeup aligns makeup with the canonical facial region is asserted but never measured. These issues are addressable and do not invalidate the core approach, but they must be fixed before the claims can be accepted.

major comments (4)
  1. [Section IV-A, Eq. (7)] The statement 'We experimentally find that Stable-Makeup generates detailed makeup images and the makeup aligns well with the facial region when target avatars are under canonical expressions' is a load-bearing empirical assumption, but no quantitative alignment test is reported. Since Eq. (7) maps pixels of Stable-Makeup-generated canonical images to UV coordinates using a mapping computed from the original rendered mesh, any misalignment (shifted facial features, misplaced eyeshadow, altered mouth/teeth) is averaged into the global UV map and propagates through Coherent Duplication and the refinement stage. Please report an alignment metric, such as face-parsing IoU or landmark reprojection error between I_cano_theta and the rendered canonical images, across the 50 reference images and 9 avatars, or at least characterize failure cases.
  2. [Section V-B, Table I] The state-of-the-art claim rests on a comparison with a single baseline (ClipFace). Table I reports no error bars, no significance tests, and no identity-preservation metric in the main comparison (CLIP-I appears only in the ablation table). Additionally, FID/KID compare rendered avatar images against real reference makeup images, a distribution mismatch that makes the numbers hard to interpret. Please add more baselines (e.g., an adapted Geneavatar), report identity preservation against the original avatar (e.g., DINO-I/CLIP-I between output and original), and provide error bars over the 50 reference/9 avatar combinations.
  3. [Section V-B, AIME metric] The proposed GPT-4o-based AIME metric is used as evidence of superiority, but it is not validated against human judgments, so the claim that AvatarMakeup achieves better 'human preference' is unsupported. Please either validate AIME against a human study (e.g., correlation or agreement) or treat it as a qualitative illustration and rely on established metrics for the main claims.
  4. [Section IV-B] The Detail Refinement module is under-specified. The paper states that noise is injected at small timestamps t and that the base makeup serves as a structural prior, but it does not describe how the base makeup image is incorporated (e.g., as the initial latent, as an additional conditioning input, or through DDIM inversion). Without an explicit formulation for generating \hat{I}_\theta, the refinement stage is not reproducible and the ablation results for 'w/o Detail Refinement' cannot be independently assessed. Please provide the exact equations or pseudocode for the refinement guidance generation.
minor comments (7)
  1. [Abstract and Section IV-A] The word 'recoding' in the abstract should be 'recording'.
  2. [Section IV-A, Eq. (7)] The notation in Eq. (7) is ambiguous: S_i is defined as a set of pixel positions, but the equation writes 1/|S_i| sum over H,W in S_i; clarify that the sum is over all source pixels that map to the same UV pixel, and define I_UV(h,w) consistently.
  3. [Section V-A] The phrase '16 different-view fuidance images' contains a typo; it should be 'guidance'.
  4. [Table I] The text refers to 'Tab I' and 'Table I' inconsistently, and the table caption repeats the table number; unify the style.
  5. [Section V-C, Fig. 5] The qualitative comparison with GaussianEditor and TIP-Editor uses different conditioning modalities; this is acknowledged, but the caption in Fig. 5 should state the prompts used for each baseline to make the comparison reproducible.
  6. [References] References [2] and [64] are duplicates of the same GaussianAvatars paper; keep only one.
  7. [Section III-B, Eq. (6)] The norm subscript in Eq. (6) is typeset oddly; it should be the squared L2 norm of the difference between the noise and the predicted noise, and the rendering should be fixed.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: reported comparisons are against external reference makeup images, and the only self-citation is not load-bearing.

full rationale

The paper's derivation is a pipeline, not a closed loop. The coarse stage constructs a UV map by averaging Stable-Makeup guidance images (Eq. 7), then uses that UV map to synthesize coherent guidance and optimizes the avatar with the losses in Eqs. (8)-(10). The main quantitative claims are supported by DINO-I and CLIP-I scores that compare rendered avatars to the original LADN reference makeup images, which are external to the fitted UV map and guidance images; FID and KID are likewise computed against the reference image distribution. Thus the evaluation is not comparing the output to the same guidance used for optimization. The only self-citation is DreamLCM [58], mentioned alongside SDS as an example of image-guidance optimization; it is not used to justify any premise or to close any derivational loop. The AIME metric is author-proposed, but it is a supplementary GPT-4o preference score rather than a step in the derivation, so it is a validation concern rather than circularity. The load-bearing assumption in Sec. IV-A that Stable-Makeup aligns makeup well at canonical expressions is empirical and not quantitatively tested in the paper; however, an untested external dependency is a robustness risk, not a circular derivation. No prediction in the paper reduces to its input by construction.

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

The method rests on strong external dependencies: Stable-Makeup for makeup content, GaussianAvatars for geometry, and a face-parsing model for masks. The only new entity introduced to support the central claim is the AIME evaluation metric, which is not independently validated. The listed free parameters are manual hyperparameter choices rather than fitted constants, but they directly affect the reported consistency and quality numbers.

free parameters (5)
  • Loss weights lambda1 and lambda2 = 10.0 and 10.0
    Manual weights balancing makeup fidelity and identity preservation; not tuned on a validation set and no sensitivity analysis is reported.
  • UV map resolution = 256 x 256
    Resolution limits base makeup detail; higher resolution could change quality and consistency results.
  • Refinement diffusion timestamps = t in [20, 400]
    Noise injection range chosen to balance detail and consistency; no ablation of this interval is provided.
  • Guidance image counts = 16 views for UV fill; 5,000 base stage; 3,000 refinement stage
    Sampling counts chosen without ablation; they affect coverage of poses and expressions and therefore the consistency claim.
  • Training iterations, learning rate, spherical harmonics degree = 13,000 iterations; lr 1e-3; sh=0
    Optimization schedule and representation settings chosen by hand, with no sensitivity analysis reported.
assumptions (4)
  • domain assumption Stable-Makeup produces makeup images that align correctly with the target face under canonical expressions (Section IV-A).
    The paper states this as an experimental finding, but provides no quantitative alignment test. If alignment fails for a reference makeup image, the UV map and all downstream guidance are corrupted.
  • domain assumption The 3D Gaussians are optimally distributed on the FLAME mesh so that freezing position, rotation, and scale preserves geometry during animation (Section IV-B).
    The method only optimizes feature and opacity parameters. If the GaussianAvatars binding is inaccurate for extreme expressions, the frozen geometry will cause rendering artifacts.
  • domain assumption The face-parsing model correctly segments makeup regions across views and expressions (Section IV-B).
    Masks gate both the makeup loss and the identity-preservation loss. Mask errors would either block makeup application or cause identity leakage into non-makeup regions.
  • domain assumption Stable-Makeup generalizes to arbitrary reference makeup images and to rendered avatar images as input (Section III-B).
    The pipeline inherits all failure modes of a pretrained 2D makeup transfer model; no adaptation or fine-tuning of Stable-Makeup is described.
invented entities (1)
  • AI-Assisted Makeup Evaluation (AIME) metric
    purpose: A GPT-4o-based scoring protocol for makeup similarity, makeup quality, and identity preservation in the experiments.
    The metric is proposed and used only by the authors in this paper. It is not validated against human raters or established benchmarks, so its scores are not independently calibrated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AvatarMakeup: Realistic Makeup Transfer for 3D Animatable Head Avatars." pith.science (2026). https://pith.science/paper/Y45SLDTF

@misc{pith2026250702419,
  author       = {Pith},
  title        = {Pith review of: AvatarMakeup: Realistic Makeup Transfer for 3D Animatable Head Avatars},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y45SLDTF}},
  note         = {Machine review of arXiv:2507.02419}
}
read the original abstract

Similar to facial beautification in real life, 3D virtual avatars require personalized customization to enhance their visual appeal, yet this area remains insufficiently explored. Although current 3D Gaussian editing methods can be adapted for facial makeup purposes, these methods fail to meet the fundamental requirements for achieving realistic makeup effects: 1) ensuring a consistent appearance during drivable expressions, 2) preserving the identity throughout the makeup process, and 3) enabling precise control over fine details. To address these, we propose a specialized 3D makeup method named AvatarMakeup, leveraging a pretrained diffusion model to transfer makeup patterns from a single reference photo of any individual. We adopt a coarse-to-fine idea to first maintain the consistent appearance and identity, and then to refine the details. In particular, the diffusion model is employed to generate makeup images as supervision. Due to the uncertainties in diffusion process, the generated images are inconsistent across different viewpoints and expressions. Therefore, we propose a Coherent Duplication method to coarsely apply makeup to the target while ensuring consistency across dynamic and multiview effects. Coherent Duplication optimizes a global UV map by recoding the averaged facial attributes among the generated makeup images. By querying the global UV map, it easily synthesizes coherent makeup guidance from arbitrary views and expressions to optimize the target avatar. Given the coarse makeup avatar, we further enhance the makeup by incorporating a Refinement Module into the diffusion model to achieve high makeup quality. Experiments demonstrate that AvatarMakeup achieves state-of-the-art makeup transfer quality and consistency throughout animation.

Figures

Figures reproduced from arXiv: 2507.02419 by the authors.

Figure 1
Figure 1. 3D makeup transfer examples generated by AvatarMakeup. We improve the quality of makeup transfer by employing [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of AvatarMakeup. AvatarMakeup takes a reconstructed avatar and a reference makeup image as input and [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the inconsistency during optimization. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Qualitative comparision between our methods and [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Qualitative Comparison. GaussianEditor [20] alters the face color but generates low-quality eye shadow. TIP-Editor [22] struggles to preserve the identity of the original avatars while generating incorrect makeup colors, such as the mismatched lips color in the first r…
Figure 6
Figure 6. Figure 6: Additional makeup results generated using AvatarMakeup. Given a real-world reference makeup, our methods can transfer [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 6
Figure 6. Figure 6: Quantitative Results. We conduct quantitative experiments by calculating the four metrics comparing our methods and ClipFace [60]. The results are shown in Tab I. We can see that AvatarMakeup outperforms ClipFace in the DINO-I metric. Remarkably, AvatarMakeup achieves …

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

74 extracted references · 44 canonical work pages · cited by 1 Pith paper

  1. [1]

    3d gaussian splatting for real-time radiance field rendering,

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis, “3d gaussian splatting for real-time radiance field rendering,” ACM Transactions on Graphics (ToG), vol. 42, no. 4, pp. 1–14, 2023

    2023

  2. [2]

    Gaussianavatars: Photorealistic head avatars with rigged 3d gaussians,

    S. Qian, T. Kirschstein, L. Schoneveld, D. Davoli, S. Giebenhain, and M. Nießner, “Gaussianavatars: Photorealistic head avatars with rigged 3d gaussians,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , June 2024, pp. 20 299–20 309

    work page 2024

  3. [3]

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

    Z. Shao, Z. Wang, Z. Li, D. Wang, X. Lin, Y . Zhang, M. Fan, and Z. Wang, “SplattingAvatar: Realistic Real-Time Human Avatars with Mesh-Embedded Gaussian Splatting,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    work page 2024

  4. [4]

    Instaface: Identity-preserving facial editing with single image inference,

    M. Khan, M. Jia, X. Zhang, E. Yu, and K. Musial-Gabrys, “Instaface: Identity-preserving facial editing with single image inference,” arXiv preprint arXiv:2502.20577, 2025

    arXiv 2025

  5. [5]

    Beautygan: Instance-level facial makeup transfer with deep generative adversarial network,

    T. Li, R. Qian, C. Dong, S. Liu, Q. Yan, W. Zhu, and L. Lin, “Beautygan: Instance-level facial makeup transfer with deep generative adversarial network,” in Proceedings of the 26th ACM international conference on Multimedia, 2018, pp. 645–653

    work page 2018

  6. [6]

    Psgan: Pose and expression robust spatial-aware gan for customizable makeup transfer,

    W. Jiang, S. Liu, C. Gao, J. Cao, R. He, J. Feng, and S. Yan, “Psgan: Pose and expression robust spatial-aware gan for customizable makeup transfer,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2020, pp. 5194–5202

    work page 2020

  7. [7]

    Spatially-invariant style- codes controlled makeup transfer,

    H. Deng, C. Han, H. Cai, G. Han, and S. He, “Spatially-invariant style- codes controlled makeup transfer,” in Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition, 2021, pp. 6549– 6557

    work page 2021

  8. [8]

    Lipstick ain’t enough: beyond color matching for in-the-wild makeup transfer,

    T. Nguyen, A. T. Tran, and M. Hoai, “Lipstick ain’t enough: beyond color matching for in-the-wild makeup transfer,” in Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition , 2021, pp. 13 305–13 314

    work page 2021

Show all 74 references
  1. [9]

    Ramgan: Region attentive morphing gan for region-level makeup transfer,

    J. Xiang, J. Chen, W. Liu, X. Hou, and L. Shen, “Ramgan: Region attentive morphing gan for region-level makeup transfer,” in European Conference on Computer Vision . Springer, 2022, pp. 719–735

    2022

  2. [10]

    Ladn: Local adversarial disentangling network for facial makeup and de-makeup,

    Q. Gu, G. Wang, M. T. Chiu, Y .-W. Tai, and C.-K. Tang, “Ladn: Local adversarial disentangling network for facial makeup and de-makeup,” in Proceedings of the IEEE/CVF International conference on computer vision, 2019, pp. 10 481–10 490

    2019

  3. [11]

    Psgan++: robust detail-preserving makeup transfer and removal,

    S. Liu, W. Jiang, C. Gao, R. He, J. Feng, B. Li, and S. Yan, “Psgan++: robust detail-preserving makeup transfer and removal,” IEEE Transac- tions on Pattern Analysis and Machine Intelligence , vol. 44, no. 11, pp. 8538–8551, 2021

    2021

  4. [12]

    Facial attribute transformers for precise and robust makeup transfer,

    Z. Wan, H. Chen, J. An, W. Jiang, C. Yao, and J. Luo, “Facial attribute transformers for precise and robust makeup transfer,” in Proceedings of the IEEE/CVF winter conference on applications of computer vision , 2022, pp. 1717–1726

    2022

  5. [13]

    Beautyrec: Robust, efficient, and component-specific makeup transfer,

    Q. Yan, C. Guo, J. Zhao, Y . Dai, C. C. Loy, and C. Li, “Beautyrec: Robust, efficient, and component-specific makeup transfer,” in Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 1102–1110

    2023

  6. [14]

    Ssat ++: A semantic-aware and ver- satile makeup transfer network with local color consistency constraint,

    Z. Sun, Y . Chen, and S. Xiong, “Ssat ++: A semantic-aware and ver- satile makeup transfer network with local color consistency constraint,” IEEE Transactions on Neural Networks and Learning Systems , 2023

    2023

  7. [15]

    Ca-gan: Weakly supervised color aware gan for controllable makeup transfer,

    R. Kips, P. Gori, M. Perrot, and I. Bloch, “Ca-gan: Weakly supervised color aware gan for controllable makeup transfer,” in Computer Vision– ECCV 2020 Workshops: Glasgow, UK, August 23–28, 2020, Proceed- ings, Part III 16 . Springer, 2020, pp. 280–296

    2020

  8. [16]

    Elegant: Exquisite and locally editable gan for makeup transfer,

    C. Yang, W. He, Y . Xu, and Y . Gao, “Elegant: Exquisite and locally editable gan for makeup transfer,” in European Conference on Computer Vision. Springer, 2022, pp. 737–754

    2022

  9. [17]

    Stable-makeup: When real-world makeup transfer meets diffusion model,

    Y . Zhang, L. Wei, Q. Zhang, Y . Song, J. Liu, H. Li, X. Tang, Y . Hu, and H. Zhao, “Stable-makeup: When real-world makeup transfer meets diffusion model,” 2024. [Online]. Available: https: //arxiv.org/abs/2403.07764

    2024 arXiv

  10. [18]

    Geneavatar: Generic expression-aware volu- metric head avatar editing from a single image,

    C. Bao, Y . Zhang, Y . Li, X. Zhang, B. Yang, H. Bao, M. Pollefeys, G. Zhang, and Z. Cui, “Geneavatar: Generic expression-aware volu- metric head avatar editing from a single image,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), J...

    2024

  11. [19]

    Next3d: Generative neural texture rasterization for 3d-aware head avatars,

    J. Sun, X. Wang, L. Wang, X. Li, Y . Zhang, H. Zhang, and Y . Liu, “Next3d: Generative neural texture rasterization for 3d-aware head avatars,” in CVPR, 2023

    2023

  12. [20]

    Gaussianeditor: Swift and controllable 3d editing with gaussian splatting,

    Y . Chen, Z. Chen, C. Zhang, F. Wang, X. Yang, Y . Wang, Z. Cai, L. Yang, H. Liu, and G. Lin, “Gaussianeditor: Swift and controllable 3d editing with gaussian splatting,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2024, pp. 21 476–21 485

    2024

  13. [21]

    Dge: Direct gaussian 3d editing by consistent multi-view editing,

    M. Chen, I. Laina, and A. Vedaldi, “Dge: Direct gaussian 3d editing by consistent multi-view editing,” arXiv preprint arXiv:2404.18929 , 2024

    2024 arXiv

  14. [22]

    Tip-editor: An accurate 3d editor following both text-prompts and image-prompts,

    J. Zhuang, D. Kang, Y .-P. Cao, G. Li, L. Lin, and Y . Shan, “Tip-editor: An accurate 3d editor following both text-prompts and image-prompts,” ACM Transactions on Graphics (TOG) , vol. 43, no. 4, pp. 1–12, 2024

    2024

  15. [23]

    Learning a model of facial shape and expression from 4D scans,

    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, (Proc. SIGGRAPH Asia) , vol. 36, no. 6, pp. 194:1–194:17,

  16. [24]

    Smpl: A skinned multi-person linear model,

    M. Loper, N. Mahmood, J. Romero, G. Pons-Moll, and M. J. Black, “Smpl: A skinned multi-person linear model,” in Seminal Graphics Papers: Pushing the Boundaries, Volume 2 , 2023, pp. 851–866

    2023

  17. [25]

    Face2face: Real-time face capture and reenactment of rgb videos,

    J. Thies, M. Zollhofer, M. Stamminger, C. Theobalt, and M. Nießner, “Face2face: Real-time face capture and reenactment of rgb videos,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 2387–2395

    2016

  18. [26]

    Pifu: Pixel-aligned implicit function for high-resolution clothed human digitization,

    S. Saito, Z. Huang, R. Natsume, S. Morishima, A. Kanazawa, and H. Li, “Pifu: Pixel-aligned implicit function for high-resolution clothed human digitization,” in Proceedings of the IEEE/CVF international conference on computer vision , 2019, pp. 2304–2314

    2019

  19. [27]

    Pifuhd: Multi-level pixel- aligned implicit function for high-resolution 3d human digitization,

    S. Saito, T. Simon, J. Saragih, and H. Joo, “Pifuhd: Multi-level pixel- aligned implicit function for high-resolution 3d human digitization,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 84–93

    2020

  20. [28]

    Arch: Animatable reconstruction of clothed humans,

    Z. Huang, Y . Xu, C. Lassner, H. Li, and T. Tung, “Arch: Animatable reconstruction of clothed humans,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2020, pp. 3093–3102

    2020

  21. [29]

    Arch++: Animation- ready clothed human reconstruction revisited,

    T. He, Y . Xu, S. Saito, S. Soatto, and T. Tung, “Arch++: Animation- ready clothed human reconstruction revisited,” in Proceedings of the IEEE/CVF international conference on computer vision , 2021, pp. 11 046–11 056

    2021

  22. [30]

    Hiface: High-fidelity 3d face reconstruction by learning static and dynamic details,

    Z. Chai, T. Zhang, T. He, X. Tan, T. Baltrusaitis, H. Wu, R. Li, S. Zhao, C. Yuan, and J. Bian, “Hiface: High-fidelity 3d face reconstruction by learning static and dynamic details,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 9087–9098

    2023

  23. [31]

    Vid2avatar: 3d avatar reconstruction from videos in the wild via self-supervised scene decomposition,

    C. Guo, T. Jiang, X. Chen, J. Song, and O. Hilliges, “Vid2avatar: 3d avatar reconstruction from videos in the wild via self-supervised scene decomposition,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 12 858–12 868

    2023

  24. [32]

    Nerf: Representing scenes as neural radiance fields for view synthesis,

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,” Communications of the ACM , vol. 65, no. 1, pp. 99–106, 2021

    2021

  25. [33]

    Humanrf: High-fidelity neural radiance fields for humans in motion,

    M. Is ¸ık, M. R ¨unz, M. Georgopoulos, T. Khakhulin, J. Starck, L. Agapito, and M. Nießner, “Humanrf: High-fidelity neural radiance fields for humans in motion,” ACM Transactions on Graphics (TOG), vol. 42, no. 4, pp. 1–12, 2023. [Online]. Available: https://doi.org/10.1145/3592415

    2023 doi

  26. [34]

    Instantavatar: Learning avatars from monocular video in 60 seconds,

    T. Jiang, X. Chen, J. Song, and O. Hilliges, “Instantavatar: Learning avatars from monocular video in 60 seconds,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 16 922–16 932

    2023

  27. [35]

    Dynamic neural radiance fields for monocular 4d facial avatar reconstruction,

    G. Gafni, J. Thies, M. Zollhofer, and M. Nießner, “Dynamic neural radiance fields for monocular 4d facial avatar reconstruction,” in Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 8649–8658

    2021

  28. [36]

    Neural head avatars from monocular rgb videos,

    P.-W. Grassal, M. Prinzler, T. Leistner, C. Rother, M. Nießner, and J. Thies, “Neural head avatars from monocular rgb videos,” in Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2022, pp. 18 653–18 664

    2022

  29. [37]

    Im avatar: Implicit morphable head avatars from videos,

    Y . Zheng, V . F. Abrevaya, M. C. B ¨uhler, X. Chen, M. J. Black, and O. Hilliges, “Im avatar: Implicit morphable head avatars from videos,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 13 545–13 555

    2022

  30. [38]

    Headnerf: A real-time nerf-based parametric head model,

    Y . Hong, B. Peng, H. Xiao, L. Liu, and J. Zhang, “Headnerf: A real-time nerf-based parametric head model,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 20 374–20 384

    2022

  31. [39]

    Instant volumetric head avatars,

    W. Zielonka, T. Bolkart, and J. Thies, “Instant volumetric head avatars,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 4574–4584

    2023

  32. [40]

    Instant neural graphics primitives with a multiresolution hash encoding,

    T. M ¨uller, A. Evans, C. Schied, and A. Keller, “Instant neural graphics primitives with a multiresolution hash encoding,” ACM Transactions on Graphics (ToG), vol. 41, no. 4, pp. 1–15, 2022. 12

    2022

  33. [41]

    Headgas: Real-time animatable head avatars via 3d gaussian splatting,

    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,” in European Conference on Computer Vision. Springer, 2024, pp. 459–476

    2024

  34. [42]

    Npga: Neural parametric gaussian avatars,

    S. Giebenhain, T. Kirschstein, M. R ¨unz, L. Agapito, and M. Nießner, “Npga: Neural parametric gaussian avatars,” in SIGGRAPH Asia 2024 Conference Papers, 2024, pp. 1–11

    2024

  35. [43]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” 2021

    2021

  36. [44]

    Dreambooth: Fine tuning text-to-image diffusion models for subject- driven generation,

    N. Ruiz, Y . Li, V . Jampani, Y . Pritch, M. Rubinstein, and K. Aberman, “Dreambooth: Fine tuning text-to-image diffusion models for subject- driven generation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2023, pp. 22 500–22 510

    2023

  37. [45]

    Adding conditional control to text-to-image diffusion models,

    L. Zhang, A. Rao, and M. Agrawala, “Adding conditional control to text-to-image diffusion models,” 2023

    2023

  38. [46]

    Prompt-to-prompt image editing with cross attention control,

    A. Hertz, R. Mokady, J. Tenenbaum, K. Aberman, Y . Pritch, and D. Cohen-Or, “Prompt-to-prompt image editing with cross attention control,” arXiv preprint arXiv:2208.01626 , 2022

    2022 arXiv

  39. [47]

    Uni-controlnet: All-in-one control to text-to-image diffusion models,

    S. Zhao, D. Chen, Y .-C. Chen, J. Bao, S. Hao, L. Yuan, and K.-Y . K. Wong, “Uni-controlnet: All-in-one control to text-to-image diffusion models,” Advances in Neural Information Processing Systems , 2023

    2023

  40. [48]

    Omniedit: Building image editing generalist models through specialist supervision,

    C. Wei, Z. Xiong, W. Ren, X. Du, G. Zhang, and W. Chen, “Omniedit: Building image editing generalist models through specialist supervision,” in The Thirteenth International Conference on Learning Representa- tions, 2024

    2024

  41. [49]

    Freeedit: Mask-free reference-based image editing with multi- modal instruction,

    R. He, K. Ma, L. Huang, S. Huang, J. Gao, X. Wei, J. Dai, J. Han, and S. Liu, “Freeedit: Mask-free reference-based image editing with multi- modal instruction,” arXiv preprint arXiv:2409.18071 , 2024

    2024 arXiv

  42. [50]

    Mige: A unified framework for multimodal instruction-based image generation and editing,

    X. Tian, W. Li, B. Xu, Y . Yuan, Y . Wang, and H. Shen, “Mige: A unified framework for multimodal instruction-based image generation and editing,” arXiv preprint arXiv:2502.21291 , 2025

    2025 arXiv

  43. [51]

    Unpaired image-to-image translation using cycle-consistent adversarial networks,

    J.-Y . Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired image-to-image translation using cycle-consistent adversarial networks,” in Proceedings of the IEEE international conference on computer vision , 2017, pp. 2223–2232

    2017

  44. [52]

    Pairedcyclegan: Asymmetric style transfer for applying and removing makeup,

    H. Chang, J. Lu, F. Yu, and A. Finkelstein, “Pairedcyclegan: Asymmetric style transfer for applying and removing makeup,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2018, pp. 40–48

    2018

  45. [53]

    Pro- tecting facial privacy: Generating adversarial identity masks via style- robust makeup transfer,

    S. Hu, X. Liu, Y . Zhang, M. Li, L. Y . Zhang, H. Jin, and L. Wu, “Pro- tecting facial privacy: Generating adversarial identity masks via style- robust makeup transfer,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 15 014–15 023

    2022

  46. [54]

    Generative adversarial nets,

    I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial nets,” in Proceedings of the 27th International Conference on Neural Information Processing Systems - Volume 2 , ser. NIPS’14. Cambridge, MA, U...

    2014

  47. [55]

    Denoising diffusion probabilistic mod- els,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic mod- els,” in Proceedings of the 34th International Conference on Neural Information Processing Systems, ser. NIPS ’20. Red Hook, NY , USA: Curran Associates Inc., 2020

    2020

  48. [56]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervision,” 2021. [Online]. Available: https://arxiv.org/abs/2103.00020

    2021 arXiv

  49. [57]

    Dreamfusion: Text- to-3d using 2d diffusion,

    B. Poole, A. Jain, J. T. Barron, and B. Mildenhall, “Dreamfusion: Text- to-3d using 2d diffusion,” arXiv, 2022

    2022

  50. [58]

    Dreamlcm: Towards high quality text-to-3d generation via latent consistency model,

    Y . Zhong, X. Zhang, Y . Zhao, and Y . Wei, “Dreamlcm: Towards high quality text-to-3d generation via latent consistency model,” in Proceedings of the 32nd ACM International Conference on Multimedia, ser. MM ’24. New York, NY , USA: Association for Computing Machinery, 2024, p...

    2024

  51. [59]

    Bisenet v2: Bilateral network with guided aggregation for real- time semantic segmentation,

    C. Yu, C. Gao, J. Wang, G. Yu, C. Shen, and N. Sang, “Bisenet v2: Bilateral network with guided aggregation for real- time semantic segmentation,” Int. J. Comput. Vision , vol. 129, no. 11, p. 3051–3068, Nov. 2021. [Online]. Available: https: //doi.org/10.1007/s11263-021-01515-2

    2021 doi

  52. [60]

    Clipface: Text-guided editing of textured 3d morphable models,

    S. Aneja, J. Thies, A. Dai, and M. Nießner, “Clipface: Text-guided editing of textured 3d morphable models,” in ACM SIGGRAPH 2023 Conference Proceedings, 2023, pp. 1–11

    2023

  53. [61]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980 , 2014

    2014 arXiv

  54. [62]

    Nersemble: Multi-view radiance field reconstruction of human heads,

    T. Kirschstein, S. Qian, S. Giebenhain, T. Walter, and M. Nießner, “Nersemble: Multi-view radiance field reconstruction of human heads,” ACM Trans. Graph. , vol. 42, no. 4, jul 2023. [Online]. Available: https://doi.org/10.1145/3592455

    2023 doi

  55. [63]

    Ladn: Local adversarial disentangling network for facial makeup and de-makeup,

    Q. Gu, G. Wang, M. T. Chiu, Y .-W. Tai, and C.-K. Tang, “Ladn: Local adversarial disentangling network for facial makeup and de-makeup,”

  56. [64]

    Gaussianavatars: Photorealistic head avatars with rigged 3d gaussians,

    S. Qian, T. Kirschstein, L. Schoneveld, D. Davoli, S. Giebenhain, and M. Nießner, “Gaussianavatars: Photorealistic head avatars with rigged 3d gaussians,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 20 299–20 309

    2024

  57. [65]

    Dinov2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V . V o, M. Szafraniec, V . Khali- dov, P. Fernandez, D. Haziza, F. Massa, A. El-Nouby, R. Howes, P.-Y . Huang, H. Xu, V . Sharma, S.-W. Li, W. Galuba, M. Rabbat, M. Assran, N. Ballas, G. Synnaeve, I. Misra, H. Jegou, J. Mairal, P. Labatu...

    2023

  58. [66]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium,

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,” Advances in neural information processing systems , vol. 30, 2017

    2017

  59. [67]

    Rethinking the inception architecture for computer vision,

    C. Szegedy, V . Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, “Rethinking the inception architecture for computer vision,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 2818–2826

    2016

  60. [68]

    Demysti- fying mmd gans,

    M. Bi ´nkowski, D. J. Sutherland, M. Arbel, and A. Gretton, “Demysti- fying mmd gans,” arXiv preprint arXiv:1801.01401 , 2018

    2018 arXiv

  61. [69]

    Gpt-4o system card,

    A. Hurst, A. Lerer, A. P. Goucher, A. Perelman, A. Ramesh, A. Clark, A. Ostrow, A. Welihinda, A. Hayes, A. Radford et al., “Gpt-4o system card,” arXiv preprint arXiv:2410.21276 , 2024

    2024 arXiv

  62. [70]

    Instructpix2pix: Learning to follow image editing instructions,

    T. Brooks, A. Holynski, and A. A. Efros, “Instructpix2pix: Learning to follow image editing instructions,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2023, pp. 18 392–18 402

    2023

  63. [71]

    Gpt-4 technical report,

    OpenAI, J. Achiam, and e. a. Steven Adler, “Gpt-4 technical report,”

  64. [2017]

    Available: https://doi.org/10.1145/3130800.3130813

    [Online]. Available: https://doi.org/10.1145/3130800.3130813

  65. [2019]

    Available: https://arxiv.org/abs/1904.11272

    [Online]. Available: https://arxiv.org/abs/1904.11272

    1904 arXiv

  66. [2024]

    Available: https://arxiv.org/abs/2303.08774

    [Online]. Available: https://arxiv.org/abs/2303.08774

    arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.