Pith. sign in

REVIEW 4 major objections 5 minor 94 references

DevilSight: Augmenting Monocular Human Avatar Reconstruction through a Virtual Perspective

T0 review · 4 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 A monocular 3D Gaussian avatar can reach multi-view fidelity if a video generative model supplies a synthetic back-view video as pseudo-supervision, with fine-tuning and super-resolution making that video faithful.

desk verdict Plausible idea, weak evidence: back-view pseudo-supervision from an undisclosed pretrained model whose training data may overlap the test set undercuts the main claim. read the letter →

arxiv 2509.00403 v1 pith:5BBIVHCH submitted 2025-08-30 cs.CV

Yushuo Chen , Ruizhi Shao , Youxin Pang , Hongwen Zhang , Xinyi Wu , Rihui Wu , Yebin Liu This is my paper
classification cs.CV
keywords monocularhumanavatarreconstruction3DGaussiansplattingvideogenerativemodelpseudo-supervisionnovelviewsynthesisphysicalidentityinversiondiffusiontransformersuper-resolutiongeneration
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 tries to show that a monocular human avatar built with 3D Gaussian splatting can reach the fidelity of multi-view methods if, before training, a video generative model is asked to synthesize the one viewpoint the camera never saw: the subject's back. That synthetic back-view video is treated as real footage and used as extra training data, which simultaneously lets the avatar keep high-frequency dynamic details visible in the input and suppress artifacts in regions that were never observed. Two supporting mechanisms make the generated video usable: fine-tuning the generator on the input clip to lock in the exact subject's appearance and motion (called physical identity inversion), and a patch-based denoising pass that doubles the generated resolution to match the captured footage. The authors report that this recipe outperforms strong monocular and multi-view baselines on reconstruction and novel-view synthesis, and that its ablations confirm both mechanisms contribute.

What carries the argument

The central object is the back-view synthetic video produced by Human4DiT, a diffusion-transformer video model conditioned on identity and pose, which the paper treats as additional pseudo-ground-truth frames for a 3D Gaussian avatar. The avatar is represented as two pose-conditioned Gaussian maps anchored to the SMPL template, following Animatable Gaussians. The two enabling mechanisms are: (1) Physical Identity Inversion, a subject-driven fine-tuning technique that injects a learnable identity token into the diffusion model so the generated rear view matches the input subject's appearance and motion; and (2) patch-based super-resolution generation, which splits the latent image into nine o

What would settle it

Run the method on a newly captured subject that provably does not appear in any public human video dataset, retrain or restrict the generative model so its training set excludes the test subject, and compare novel-view PSNR/LPIPS against the same avatar trained without the generated back view; if the improvement vanishes, the reported gains are memorization artifacts. A second check: replace the generated back view with the true multi-view back view on THuman4.0; if the avatar trained on the true back view does not clearly beat the one trained on the generated back view, the generator is repro

Watch

Extended reading notes

Core claim

The central claim is that generative priors from a 360-degree human video model can serve as pseudo-supervision for dynamic avatar reconstruction. The authors argue that the failure of prior work is a data problem, not just a representation problem: monocular input provides no observation of rear-view appearance, so pose-conditioned networks either smooth over unseen regions or hallucinate artifacts. DevilSight's contribution is to fill that missing observation by generating a rear-view video of the same person performing the same motion, then treating that video as ground truth. To make generation faithful, they fine-tune the generator with a learnable identity embedding initialized from th

Load-bearing premise

The pipeline rests on the assumption that the video generative model, once fine-tuned on the input clip, produces a rear-view video of the same person performing the same motion that is physically consistent — and that the model has not already memorized the test subject from its own training data.

Editorial extensions

If this is right

  • Monocular avatar pipelines that currently restrict network complexity to avoid novel-view artifacts can instead keep full capacity and rely on generative pseudo-views for regularization.
  • Reconstructed avatars support novel-view synthesis of the back of the body with quality comparable to input views, directly from a single video.
  • The generated back view also improves dynamic details such as fluttering clothing that template-based methods miss, because the pseudo-supervision preserves physical identity rather than just appearance.
  • The same back-view generation strategy extends to novel pose synthesis: generating front and back views for a new pose sequence and adding them as training data animates the avatar under arbitrary poses with consistent detail.
  • Using generated videos as explicit pseudo-ground-truth, rather than as score-distillation targets, avoids the oversaturation and smoothing typical of SDS-based distillation while achieving better novel-view FID.

Reading between the lines

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

  • The framework generates only one back view for efficiency, so a natural extension is to generate left, right, and top views as well; whether the view-consistency priors of the generator hold across more vantage points is a testable scaling question the paper leaves open.
  • A risk not addressed in the paper is training-set overlap: if the video generative model was trained on the same multi-view datasets used for evaluation, its 'novel' back views could be memorized ground truth rather than genuine hallucination, which would make part of the reported gain circular.
  • Physical identity inversion costs about 20 hours of per-subject fine-tuning, so an important practical follow-up is a lightweight adapter that transfers identity without per-subject gradient updates; the paper's ablations suggest the full fine-tuning is what captures dynamic clothing detail.
  • The patch-based super-resolution result depends on overlap weighting, but the paper does not study sensitivity to patch overlap or stride; small controlled experiments varying those parameters would clarify how robust the resolution claim is.
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 / 5 minor

Summary. The paper proposes DevilSight, a framework for reconstructing animatable 3D Gaussian avatars from monocular video. The key idea is to use Human4DiT, a video diffusion transformer, to generate a back-view video of the same subject and motion, and to use that generated video as pseudo-supervision for the avatar representation. Two supporting components are introduced: Physical Identity Inversion, which fine-tunes Human4DiT on the input video with a learnable identity embedding to improve motion and appearance consistency, and a patch-based super-resolution strategy that doubles the resolution of generated frames. The method is evaluated on THuman4.0 and a new Mono2K dataset against HumanNeRF, GaussianAvatar, AnimatableGaussians, and two video-to-4D baselines. The paper claims that the generated back view enriches unseen regions and regularizes the avatar, leading to superior reconstruction quality and novel-view synthesis.

Significance. If the claims were established, the idea of using a video generative model as a virtual viewpoint for dynamic human avatar reconstruction would be a timely and valuable contribution. The pipeline is nontrivial and the two proposed strategies—identity fine-tuning and patch-based super-resolution—are sensible engineering choices. The paper also provides a useful comparison against strong baselines. However, the central empirical claim of outperforming state-of-the-art methods is not supported by the reported numbers, and a training-data leakage risk in Human4DiT remains unresolved. As presented, the evidence does not demonstrate that the generative pseudo-supervision provides independent information about unseen views, which is the load-bearing contribution of the work.

major comments (4)
  1. [§4.2, Table 1] The sentence "our method significantly outperforms other state-of-the-art approaches" is not supported by the reported numbers. On input-view reconstruction, the method is worse than AnimatableGaussians (AG) on both datasets in PSNR and SSIM (THuman4.0: 32.97 vs 34.06 dB, 0.9769 vs 0.9798; Mono2K: 32.19 vs 33.15 dB, 0.9768 vs 0.9807). Novel-view gains are small (≤0.5 dB) and inconsistent: on Mono2K Novel View 1 the PSNR is lower than AG (26.26 vs 26.29), and per-subject Table B shows ties within noise. With only five evaluation sequences and no variance or significance testing, the central claim of SOTA superiority is not established. The claims in §4.2 and the abstract should be tempered or supported with a more extensive evaluation.
  2. [§3.2–3.3; Supplementary Table A] The core assumption is that Human4DiT provides independent pseudo-supervision for unseen back views. This requires that Human4DiT has not memorized the evaluation subjects. The paper states only that Human4DiT was trained on "large-scale multi-dimensional datasets" and does not disclose whether THuman4.0 or Mono2K subjects, or their multi-view captures, were used in its training. Since Human4DiT is from the same group and THuman4.0 is a multi-view dataset used here for evaluation, the risk is concrete: if any of the five test subjects appeared in training, the generated back view is a recalled ground-truth image and the pseudo-supervision is circular. The low FID of the generated video alone (59.25 vs AG's 100.34 on novel views, Supplementary Table A) is consistent with strong generalization but also with memorization. The authors must disclose the training split of Human4DiT and evaluat
  3. [§3.4, Supplement A] The generated video is used as hard pseudo-ground truth without accounting for generator error. Supplementary Table A shows that the generated video itself has substantially lower fidelity than the 3DGS baselines on input views (PSNR 32.38 vs AG 34.06; FID 38.55 vs 17.91). Table 1 shows a corresponding input-view degradation for the full method relative to AG on PSNR/SSIM, suggesting that hard supervision with generated content hurts the observed view. The paper should either use a confidence-weighted or uncertainty-aware supervision, or provide evidence that the input-view regression is acceptable; otherwise the trade-off undermines the claimed advantage.
  4. [§4.3, Supplement A] The two named contributions (Physical Identity Inversion and super-resolution generation) are only ablated qualitatively (Figs. 7 and 8). Without quantitative ablation, it is unclear whether the reported (modest) novel-view gains are due to the pseudo-supervision or to other training choices. Also, several hyperparameters are listed in Supplement A without numerical values: λ_LPIPS, λ_offset, λ_scale, the number/overlap of patches, fine-tuning steps, and the back-view camera rotation. These should be specified to make the method reproducible and to allow readers to assess the sensitivity of the claimed regularization.
minor comments (5)
  1. [§4] The evaluation uses ground-truth SMPL-X poses fitted from multiple views, which is a favorable condition for a 'monocular' reconstruction task. This should be stated prominently and discussed as a limitation.
  2. [§3.3] The term 'Physical Identity Inversion' is not formally defined. Please clarify what 'physical identity' means beyond appearance and specify exactly which parameters are fine-tuned (e.g., which attention layers).
  3. [Table 1] Tables 1 and A use unusual spacing in decimal numbers; harmonize formatting for readability.
  4. [§4.2 / Fig. 6] The comparison with video-to-4D methods L4GM and GVFDiffusion is only qualitative. If quantitative evaluation is possible, adding numbers would strengthen the comparison.
  5. [Eq. (3)] The conditions cref and cΘ are not defined with respect to actual inputs. Specify how the identity reference and pose conditions are computed, especially during the fine-tuning stage.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found; Human4DiT is used as an external generative prior with empirical validation, and the claimed improvements are not by-construction identities.

full rationale

The paper's pipeline is: (1) estimate SMPL poses from the monocular video; (2) fine-tune Human4DiT on that video to inject identity ('Physical Identity Inversion'); (3) generate a back-view video with the same poses; (4) use that generated video as pseudo-supervision to train a 3DGS avatar. At no point is a target quantity defined in terms of itself. The generated back-view is not produced by the avatar being trained, nor is it a re-rendering of the input; it is an output of a fixed generative model conditioned on identity and pose. The avatar's loss on those frames is a standard reconstruction loss against this pseudo ground truth. The only self-referential element is the citation to Human4DiT [59], whose authors overlap with the present paper. This is a load-bearing tool, but it is not invoked as a 'uniqueness theorem' and the paper provides its own ablations (Fig. 7, Table A) testing the generated video's fidelity. The concern that Human4DiT may have memorized THuman4.0/Mono2K evaluation subjects is a data-leakage validity threat, not a circularity: the paper does not state that those subjects are in Human4DiT's training set, and no equation or construction in the manuscript reduces the novel-view predictions to the input video. Without evidence that the pseudo-supervision equals ground truth by construction, this remains a correctness risk rather than a demonstrated circular step. Therefore no circularity is established.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the availability and reliability of Human4DiT, a self-cited prior model, plus several standard assumptions about SMPL-X and 3DGS. The empirical evaluation adds no formally derived free parameters beyond training hyperparameters, but those are not reported, and the oracle-pose protocol diverges from the monocular problem statement.

free parameters (5)
  • λ_LPIPS, λ_offset, λ_scale = not specified
    Loss weights in Eq. 4 of the supplementary, chosen by hand; no values or tuning procedure are reported.
  • Number of patches and overlap = 9 patches, overlap unspecified
    Patch count is fixed at 9 in Fig. 4; the overlap amount is not reported and affects output consistency.
  • Fine-tuning steps = ~20k steps
    The number of steps for physical identity inversion is given as approximately 20k, but no learning rate or schedule is provided.
  • c_id learnable embedding = learned
    The identity embedding is initialized from CLIP and fine-tuned on the input video; its behavior is central to the method.
  • Back-view camera rotation = 180 degrees
    The camera is rotated 180 degrees around the root; this is a design choice that could affect which unseen regions are covered.
assumptions (5)
  • domain assumption Human4DiT can generate high-quality, view-consistent human videos from a single image with pose control.
    The entire pseudo-supervision strategy rests on the reliability of this self-cited prior model (Section 3.2).
  • standard math SMPL-X and linear blend skinning accurately model human body shape and dynamics.
    The avatar is anchored to the SMPL template and deformed using LBS (Section 3.2), inheriting the limitations of the template.
  • domain assumption Fine-tuning only identity-related attention layers preserves multi-view consistency while adding physical identity.
    The authors alternate tuning on input video and Human4DiT's multi-view dataset to avoid overfitting (Section 3.3), but no evidence is shown that this preserves consistency.
  • domain assumption Patch-based denoising with overlapping windows maintains spatial consistency across patches.
    The super-resolution algorithm (Section 3.4) assumes weighted merging of independently denoised patches yields coherent images; no quantitative validation of this assumption is given.
  • domain assumption Ground-truth SMPL-X poses fitted from multi-view are representative of monocular pose estimates.
    The main evaluation uses oracle poses (Section 4), which is not the same as the monocular setting claimed in the title and abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DevilSight: Augmenting Monocular Human Avatar Reconstruction through a Virtual Perspective." pith.science (2026). https://pith.science/paper/5BBIVHCH

@misc{pith2026250900403,
  author       = {Pith},
  title        = {Pith review of: DevilSight: Augmenting Monocular Human Avatar Reconstruction through a Virtual Perspective},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5BBIVHCH}},
  note         = {Machine review of arXiv:2509.00403}
}
read the original abstract

We present a novel framework to reconstruct human avatars from monocular videos. Recent approaches have struggled either to capture the fine-grained dynamic details from the input or to generate plausible details at novel viewpoints, which mainly stem from the limited representational capacity of the avatar model and insufficient observational data. To overcome these challenges, we propose to leverage the advanced video generative model, Human4DiT, to generate the human motions from alternative perspective as an additional supervision signal. This approach not only enriches the details in previously unseen regions but also effectively regularizes the avatar representation to mitigate artifacts. Furthermore, we introduce two complementary strategies to enhance video generation: To ensure consistent reproduction of human motion, we inject the physical identity into the model through video fine-tuning. For higher-resolution outputs with finer details, a patch-based denoising algorithm is employed. Experimental results demonstrate that our method outperforms recent state-of-the-art approaches and validate the effectiveness of our proposed strategies.

Figures

Figures reproduced from arXiv: 2509.00403 by the authors.

Figure 1
Figure 1. We present DevilSight, which reconstruct 3DGS human avatar with fine-grained dynamic details from monocular video. Our [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our method. Our method leverages priors from Human4DiT to enable robust monocular avatar reconstruction. It comprises three key components: (a) fine-tuning the model for personalized content generation, (b) generating consistent rear-view motion with superresolution, and (c) reconstructing 3DGS avatars using pseudo multi-view data. fidelity and meet quality expectations. We propose physi￾cal identity inv… view at source ↗
Figure 3
Figure 3. Freeview Rendering of our reconstructed subjects from THuman4.0 and Mono2K. By leveraging the priors from Hu￾man4DiT, we not only accurately reconstruct dynamic details from the monocular input but also supports novel view synthesis with equivalent quality. Zoom in to see more details. 3.3. Physical Identity Inversion by Finetuning In Human4DiT, the human identity is injected to the net￾work through a CLIP [54] embe… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Illustration of Super-resolution Generation. At each diffusion timestep, we partition the image into 9 overlapping patches. The noises are then predicted independently by patch and accumulated by weighted sum. Human4DiT generates images at a standard resolution of 768 …
Figure 5
Figure 5. Figure 5: Qualitative Comparisons on input view reconstruction and novel view synthesis. Our approach achieves high-fidelity recon￾struction by leveraging priors from Human4DiT. Zoom in to see more details. ods, when evaluating HumanNeRF on the Mono2K dataset, we resized the ima…
Figure 7
Figure 7. Figure 7 [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 6
Figure 6. Figure 6: Qualitative Comparisons with SOTA video-to-4D methods. By leveraging both human priors and generative priors, our method effectively recovers fine-grained dynamic details. ages video generation that rigorously preserves the subject’s physical identity while ensuring vi…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

94 extracted references · 55 canonical work pages

  1. [1]

    Single-image 3d human digitization with shape-guided diffusion

    Badour AlBahar, Shunsuke Saito, Hung-Yu Tseng, Changil Kim, Johannes Kopf, and Jia-Bin Huang. Single-image 3d human digitization with shape-guided diffusion. In SIG- GRAPH Asia 2023 Conference Papers , pages 1–11, 2023. 3

    2023

  2. [2]

    Guide3D: Create 3D Avatars from Text and Image Guidance

    Yukang Cao, Yan-Pei Cao, Kai Han, Ying Shan, and Kwan- Yee K Wong. Guide3d: Create 3d avatars from text and im- age guidance. arXiv preprint arXiv:2308.09705 , 2023. 2, 3

    work page Pith review arXiv 2023

  3. [3]

    Dreamavatar: Text-and-shape guided 3d hu- man avatar generation via diffusion models

    Yukang Cao, Yan-Pei Cao, Kai Han, Ying Shan, and Kwan- Yee K Wong. Dreamavatar: Text-and-shape guided 3d hu- man avatar generation via diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 958–968, 2024. 3

    2024

  4. [4]

    Videocrafter2: Overcoming data limitations for high-quality video diffusion models, 2024

    Haoxin Chen, Yong Zhang, Xiaodong Cun, Menghan Xia, Xintao Wang, Chao Weng, and Ying Shan. Videocrafter2: Overcoming data limitations for high-quality video diffusion models, 2024. 3

    2024

  5. [5]

    Gener- alizable human gaussians from single-view image

    Jinnan Chen, Chen Li, Jianfeng Zhang, Lingting Zhu, Buzhen Huang, Hanlin Chen, and Gim Hee Lee. Gener- alizable human gaussians from single-view image. arXiv preprint arXiv:2406.06050, 2024. 3

    arXiv 2024

  6. [6]

    Meshavatar: Learning high-quality triangular human avatars from multi-view videos

    Yushuo Chen, Zerong Zheng, Zhe Li, Chao Xu, and Yebin Liu. Meshavatar: Learning high-quality triangular human avatars from multi-view videos. In European Conference on Computer Vision, pages 250–269. Springer, 2024. 1, 2, 4

    2024

  7. [7]

    Scaling recti- fied flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M ¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling recti- fied flow transformers for high-resolution image synthesis. In Forty-first international conference on machine learning,

  8. [8]

    Learning neural volumetric representations of dy- namic humans in minutes

    Chen Geng, Sida Peng, Zhen Xu, Hujun Bao, and Xiaowei Zhou. Learning neural volumetric representations of dy- namic humans in minutes. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 8759–8770, 2023. 2

    2023

Show all 94 references
  1. [9]

    Humans in 4d: Re- constructing and tracking humans with transformers

    Shubham Goel, Georgios Pavlakos, Jathushan Rajasegaran, Angjoo Kanazawa, and Jitendra Malik. Humans in 4d: Re- constructing and tracking humans with transformers. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 14783–14794, 2023. 3

    2023

  2. [10]

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

    Chen Guo, Tianjian Jiang, Xu Chen, Jie Song, and Otmar Hilliges. Vid2avatar: 3d avatar reconstruction from videos in the wild via self-supervised scene decomposition. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12858–12868, 2023. 1, 3

    2023

  3. [11]

    Sparsectrl: Adding sparse controls to text-to-video diffusion models

    Yuwei Guo, Ceyuan Yang, Anyi Rao, Maneesh Agrawala, Dahua Lin, and Bo Dai. Sparsectrl: Adding sparse controls to text-to-video diffusion models. In European Conference on Computer Vision, pages 330–348. Springer, 2024. 3

    2024

  4. [12]

    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. Advances in neural information processing systems , 30, 2017. 13

    2017

  5. [13]

    Sith: Single-view tex- tured human reconstruction with image-conditioned diffu- sion

    I Ho, Jie Song, Otmar Hilliges, et al. Sith: Single-view tex- tured human reconstruction with image-conditioned diffu- sion. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 538–549, 2024. 2, 3

    2024

  6. [14]

    Expres- sive gaussian human avatars from monocular rgb video

    Hezhen Hu, Zhiwen Fan, Tianhao Wu, Yihan Xi, Seoyoung Lee, Georgios Pavlakos, Zhangyang Wang, et al. Expres- sive gaussian human avatars from monocular rgb video. Ad- vances in Neural Information Processing Systems, 37:5646– 5660, 2025. 1, 3

    2025

  7. [15]

    Animate anyone: Consistent and controllable image- to-video synthesis for character animation

    Li Hu. Animate anyone: Consistent and controllable image- to-video synthesis for character animation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8153–8163, 2024. 3

    2024

  8. [16]

    Gaussianavatar: Towards realistic human avatar modeling from a single video via animatable 3d gaussians

    Liangxiao Hu, Hongwen Zhang, Yuxiang Zhang, Boyao Zhou, Boning Liu, Shengping Zhang, and Liqiang Nie. Gaussianavatar: Towards realistic human avatar modeling from a single video via animatable 3d gaussians. In Pro- ceedings of the IEEE/CVF conference on computer vision and pat...

    2024

  9. [17]

    Gauhuman: Articu- lated gaussian splatting from monocular human videos

    Shoukang Hu, Tao Hu, and Ziwei Liu. Gauhuman: Articu- lated gaussian splatting from monocular human videos. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 20418–20431, 2024. 1, 3

    2024

  10. [18]

    Humannorm: Learning normal diffusion model for high-quality and realistic 3d hu- man generation

    Xin Huang, Ruizhi Shao, Qi Zhang, Hongwen Zhang, Ying Feng, Yebin Liu, and Qing Wang. Humannorm: Learning normal diffusion model for high-quality and realistic 3d hu- man generation. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition , pages...

    2024

  11. [19]

    Dreamwaltz: Make a scene with complex 3d animatable avatars.Advances in Neural Information Processing Systems , 36:4566–4584,

    Yukun Huang, Jianan Wang, Ailing Zeng, He Cao, Xianbiao Qi, Yukai Shi, Zheng-Jun Zha, and Lei Zhang. Dreamwaltz: Make a scene with complex 3d animatable avatars.Advances in Neural Information Processing Systems , 36:4566–4584,

  12. [20]

    Tech: Text-guided reconstruction of lifelike clothed humans

    Yangyi Huang, Hongwei Yi, Yuliang Xiu, Tingting Liao, Ji- axiang Tang, Deng Cai, and Justus Thies. Tech: Text-guided reconstruction of lifelike clothed humans. In 2024 Interna- tional Conference on 3D Vision (3DV) , pages 1531–1542. IEEE, 2024. 3

    2024

  13. [21]

    Sel- frecon: Self reconstruction your digital avatar from monoc- ular video

    Boyi Jiang, Yang Hong, Hujun Bao, and Juyong Zhang. Sel- frecon: Self reconstruction your digital avatar from monoc- ular video. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 5605– 5615, 2022. 2

    2022

  14. [22]

    Avatar- craft: Transforming text into neural human avatars with pa- rameterized shape and pose control

    Ruixiang Jiang, Can Wang, Jingbo Zhang, Menglei Chai, Mingming He, Dongdong Chen, and Jing Liao. Avatar- craft: Transforming text into neural human avatars with pa- rameterized shape and pose control. In Proceedings of the IEEE/CVF International Conference on Computer Vision ,...

    2023

  15. [23]

    Mvhuman: tailoring 2d diffusion with multi-view sampling for realistic 3d human generation

    Suyi Jiang, Haimin Luo, Haoran Jiang, Ziyu Wang, Jingyi Yu, and Lan Xu. Mvhuman: tailoring 2d diffusion with multi-view sampling for realistic 3d human generation. arXiv preprint arXiv:2312.10120, 2023. 2, 3 9

    2023 arXiv

  16. [24]

    In- stantavatar: Learning avatars from monocular video in 60 seconds

    Tianjian Jiang, Xu Chen, Jie Song, and Otmar Hilliges. In- stantavatar: Learning avatars from monocular video in 60 seconds. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 16922– 16932, 2023. 3

    2023

  17. [25]

    Neuman: Neural human radiance field from a single video

    Wei Jiang, Kwang Moo Yi, Golnoosh Samei, Oncel Tuzel, and Anurag Ranjan. Neuman: Neural human radiance field from a single video. In European Conference on Computer Vision, pages 402–418. Springer, 2022. 1, 3

    2022

  18. [26]

    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. 4

    2023

  19. [27]

    Hugs: Human gaussian splats

    Muhammed Kocabas, Jen-Hao Rick Chang, James Gabriel, Oncel Tuzel, and Anurag Ranjan. Hugs: Human gaussian splats. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , pages 505–515, 2024. 3

    2024

  20. [28]

    Dreamhuman: Animatable 3d avatars from text

    Nikos Kolotouros, Thiemo Alldieck, Andrei Zanfir, Ed- uard Bazavan, Mihai Fieraru, and Cristian Sminchisescu. Dreamhuman: Animatable 3d avatars from text. Advances in Neural Information Processing Systems, 36:10516–10529,

  21. [29]

    Guess the unseen: Dynamic 3d scene reconstruction from partial 2d glimpses

    Inhee Lee, Byungjun Kim, and Hanbyul Joo. Guess the unseen: Dynamic 3d scene reconstruction from partial 2d glimpses. 2024. 3

    2024

  22. [30]

    Gart: Gaussian articulated template mod- els

    Jiahui Lei, Yufu Wang, Georgios Pavlakos, Lingjie Liu, and Kostas Daniilidis. Gart: Gaussian articulated template mod- els. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , pages 19876–19887,

  23. [31]

    Tava: Template-free animatable volumetric actors

    Ruilong Li, Julian Tanke, Minh V o, Michael Zollh ¨ofer, J¨urgen Gall, Angjoo Kanazawa, and Christoph Lassner. Tava: Template-free animatable volumetric actors. In Eu- ropean Conference on Computer Vision , pages 419–436. Springer, 2022. 2

    2022

  24. [32]

    Posevocab: Learning joint-structured pose embeddings for human avatar modeling

    Zhe Li, Zerong Zheng, Yuxiao Liu, Boyao Zhou, and Yebin Liu. Posevocab: Learning joint-structured pose embeddings for human avatar modeling. In ACM SIGGRAPH 2023 con- ference proceedings, pages 1–11, 2023. 1, 2

    2023

  25. [33]

    Ani- matable gaussians: Learning pose-dependent gaussian maps for high-fidelity human avatar modeling

    Zhe Li, Zerong Zheng, Lizhen Wang, and Yebin Liu. Ani- matable gaussians: Learning pose-dependent gaussian maps for high-fidelity human avatar modeling. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 19711–19722, 2024. 1, 2, 4, 6, 13, 14

    2024

  26. [34]

    Diffusion4d: Fast spatial-temporal consis- tent 4d generation via video diffusion models.arXiv preprint arXiv:2405.16645, 2024

    Hanwen Liang, Yuyang Yin, Dejia Xu, Hanxue Liang, Zhangyang Wang, Konstantinos N Plataniotis, Yao Zhao, and Yunchao Wei. Diffusion4d: Fast spatial-temporal consis- tent 4d generation via video diffusion models.arXiv preprint arXiv:2405.16645, 2024. 3

    2024 arXiv

  27. [35]

    Tada! text to animatable digital avatars

    Tingting Liao, Hongwei Yi, Yuliang Xiu, Jiaxiang Tang, Yangyi Huang, Justus Thies, and Michael J Black. Tada! text to animatable digital avatars. In2024 International Con- ference on 3D Vision (3DV), pages 1508–1519. IEEE, 2024. 3

    2024

  28. [36]

    Neural actor: Neural free-view synthesis of human actors with pose con- trol

    Lingjie Liu, Marc Habermann, Viktor Rudnev, Kripasindhu Sarkar, Jiatao Gu, and Christian Theobalt. Neural actor: Neural free-view synthesis of human actors with pose con- trol. ACM transactions on graphics (TOG) , 40(6):1–16,

  29. [37]

    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. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9298–9309, 2023. 2

    2023

  30. [38]

    Humangaus- sian: Text-driven 3d human generation with gaussian splat- ting

    Xian Liu, Xiaohang Zhan, Jiaxiang Tang, Ying Shan, Gang Zeng, Dahua Lin, Xihui Liu, and Ziwei Liu. Humangaus- sian: Text-driven 3d human generation with gaussian splat- ting. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 6646–6657,

  31. [39]

    Wonder3d: Sin- gle image to 3d using cross-domain diffusion

    Xiaoxiao Long, Yuan-Chen Guo, Cheng Lin, Yuan Liu, Zhiyang Dou, Lingjie Liu, Yuexin Ma, Song-Hai Zhang, Marc Habermann, Christian Theobalt, et al. Wonder3d: Sin- gle image to 3d using cross-domain diffusion. In Proceed- ings of the IEEE/CVF conference on computer vision and pa...

    2024

  32. [40]

    Matthew Loper, Naureen Mahmood, Javier Romero, Ger- ard Pons-Moll, and Michael J. Black. SMPL: A skinned multi-person linear model. ACM Trans. Graphics (Proc. SIGGRAPH Asia), 34(6):248:1–248:16, 2015. 1, 2

    2015

  33. [41]

    Avatarstudio: Text-driven editing of 3d dynamic human head avatars

    Mohit Mendiratta, Xingang Pan, Mohamed Elgharib, Kar- tik Teotia, Ayush Tewari, Vladislav Golyanik, Adam Ko- rtylewski, and Christian Theobalt. Avatarstudio: Text-driven editing of 3d dynamic human head avatars. ACM Transac- tions on Graphics (ToG), 42(6):1–18, 2023. 2, 3

    2023

  34. [42]

    Expressive whole-body 3d gaussian avatar

    Gyeongsik Moon, Takaaki Shiratori, and Shunsuke Saito. Expressive whole-body 3d gaussian avatar. In European Conference on Computer Vision , pages 19–35. Springer,

  35. [43]

    Humansplat: Generalizable single-image human gaus- sian splatting with structure priors

    Panwang Pan, Zhuo Su, Chenguo Lin, Zhen Fan, Yongjie Zhang, Zeming Li, Tingting Shen, Yadong Mu, and Yebin Liu. Humansplat: Generalizable single-image human gaus- sian splatting with structure priors. Advances in Neural In- formation Processing Systems, 37:74383–74410, 2025. 3

    2025

  36. [44]

    Effi- cient4d: Fast dynamic 3d object generation from a single- view video

    Zijie Pan, Zeyu Yang, Xiatian Zhu, and Li Zhang. Effi- cient4d: Fast dynamic 3d object generation from a single- view video. arXiv preprint arXiv:2401.08742, 2024. 3

    2024

  37. [45]

    Ash: Animatable gaussian splats for efficient and photoreal human rendering

    Haokai Pang, Heming Zhu, Adam Kortylewski, Christian Theobalt, and Marc Habermann. Ash: Animatable gaussian splats for efficient and photoreal human rendering. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1165–1175, 2024. 3

    2024

  38. [46]

    Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed A. A. Osman, Dimitrios Tzionas, and Michael J. Black. Expressive body capture: 3D hands, face, and body from a single image. In Proceedings IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pa...

    2019

  39. [47]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF inter- national conference on computer vision , pages 4195–4205,

  40. [48]

    Selfnerf: 10 Fast training nerf for human from monocular self-rotating video

    Bo Peng, Jun Hu, Jingtao Zhou, and Juyong Zhang. Selfnerf: 10 Fast training nerf for human from monocular self-rotating video. arXiv preprint arXiv:2210.01651, 2022. 2

    2022 arXiv

  41. [49]

    Ani- matable neural radiance fields for modeling dynamic human bodies

    Sida Peng, Junting Dong, Qianqian Wang, Shangzhan Zhang, Qing Shuai, Xiaowei Zhou, and Hujun Bao. Ani- matable neural radiance fields for modeling dynamic human bodies. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 14314–14323, 2021. 1, 2

    2021

  42. [50]

    Neural body: Implicit neural representations with structured latent codes for novel view synthesis of dynamic humans

    Sida Peng, Yuanqing Zhang, Yinghao Xu, Qianqian Wang, Qing Shuai, Hujun Bao, and Xiaowei Zhou. Neural body: Implicit neural representations with structured latent codes for novel view synthesis of dynamic humans. InProceedings of the IEEE/CVF conference on computer vision and ...

    2021

  43. [51]

    Rmavatar: Photo- realistic human avatar reconstruction from monocular video based on rectified mesh-embedded gaussians

    Sen Peng, Weixing Xie, Zilong Wang, Xiaohu Guo, Zhong- gui Chen, Baorong Yang, and Xiao Dong. Rmavatar: Photo- realistic human avatar reconstruction from monocular video based on rectified mesh-embedded gaussians. arXiv preprint arXiv:2501.07104, 2025. 3

    2025 arXiv

  44. [52]

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

    Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Milden- hall. Dreamfusion: Text-to-3d using 2d diffusion. arXiv preprint arXiv:2209.14988, 2022. 2, 3

    2022 arXiv

  45. [53]

    3dgs-avatar: Animatable avatars via deformable 3d gaussian splatting

    Zhiyin Qian, Shaofei Wang, Marko Mihajlovic, Andreas Geiger, and Siyu Tang. 3dgs-avatar: Animatable avatars via deformable 3d gaussian splatting. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5020–5030, 2024. 3

    2024

  46. [54]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  47. [55]

    Sam 2: Segment anything in images and videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R”adle, Chloe Rolland, Laura Gustafson, Eric Mintun, Junt- ing Pan, Kalyan Vasudev Alwala, Nicolas Carion, Chao- Yuan Wu, Ross Girshick, Piotr Doll´ar, and Christoph Feic...

    2024 arXiv

  48. [56]

    L4gm: Large 4d gaussian reconstruction model

    Jiawei Ren, Cheng Xie, Ashkan Mirzaei, Karsten Kreis, Zi- wei Liu, Antonio Torralba, Sanja Fidler, Seung Wook Kim, Huan Ling, et al. L4gm: Large 4d gaussian reconstruction model. Advances in Neural Information Processing Systems, 37:56828–56858, 2024. 3, 7

    2024

  49. [57]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 2, 3

    2022

  50. [58]

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

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir 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 , pages 2250...

    2023

  51. [59]

    Human4dit: 360-degree human video generation with 4d diffusion transformer

    Ruizhi Shao, Youxin Pang, Zerong Zheng, Jingxiang Sun, and Yebin Liu. Human4dit: 360-degree human video generation with 4d diffusion transformer. arXiv preprint arXiv:2405.17405, 2024. 2, 4

    2024 arXiv

  52. [60]

    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

  53. [61]

    A-nerf: Articulated neural radiance fields for learn- ing human shape, appearance, and pose

    Shih-Yang Su, Frank Yu, Michael Zollh ¨ofer, and Helge Rhodin. A-nerf: Articulated neural radiance fields for learn- ing human shape, appearance, and pose. Advances in neural information processing systems, 34:12278–12291, 2021. 2

    2021

  54. [62]

    Danbo: Disentangled articulated neural body representations via graph neural networks

    Shih-Yang Su, Timur Bagautdinov, and Helge Rhodin. Danbo: Disentangled articulated neural body representations via graph neural networks. InEuropean Conference on Com- puter Vision, pages 107–124. Springer, 2022. 2

    2022

  55. [63]

    Sv3d: Novel multi-view syn- thesis and 3d generation from a single image using latent video diffusion

    Vikram V oleti, Chun-Han Yao, Mark Boss, Adam Letts, David Pankratz, Dmitry Tochilkin, Christian Laforte, Robin Rombach, and Varun Jampani. Sv3d: Novel multi-view syn- thesis and 3d generation from a single image using latent video diffusion. In European Conference on Computer...

    2024

  56. [64]

    Disentangled clothed avatar generation from text descriptions

    Jionghao Wang, Yuan Liu, Zhiyang Dou, Zhengming Yu, Yongqing Liang, Cheng Lin, Rong Xie, Li Song, Xin Li, and Wenping Wang. Disentangled clothed avatar generation from text descriptions. In European Conference on Com- puter Vision, pages 381–401. Springer, 2024. 3

    2024

  57. [65]

    Arah: Animatable volume rendering of articulated hu- man sdfs

    Shaofei Wang, Katja Schwarz, Andreas Geiger, and Siyu Tang. Arah: Animatable volume rendering of articulated hu- man sdfs. In European conference on computer vision, pages 1–19. Springer, 2022. 1, 2

    2022

  58. [66]

    Disco: Disentangled control for realistic human dance generation

    Tan Wang, Linjie Li, Kevin Lin, Yuanhao Zhai, Chung- Ching Lin, Zhengyuan Yang, Hanwang Zhang, Zicheng Liu, and Lijuan Wang. Disco: Disentangled control for realistic human dance generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition ,...

    2024

  59. [67]

    Videocomposer: Compositional video synthesis with motion controllability

    Xiang Wang, Hangjie Yuan, Shiwei Zhang, Dayou Chen, Ji- uniu Wang, Yingya Zhang, Yujun Shen, Deli Zhao, and Jin- gren Zhou. Videocomposer: Compositional video synthesis with motion controllability. Advances in Neural Information Processing Systems, 36:7594–7611, 2023. 3

    2023

  60. [68]

    Emu3: Next-token prediction is all you need

    Xinlong Wang, Xiaosong Zhang, Zhengxiong Luo, Quan Sun, Yufeng Cui, Jinsheng Wang, Fan Zhang, Yueze Wang, Zhen Li, Qiying Yu, et al. Emu3: Next-token prediction is all you need. arXiv preprint arXiv:2409.18869, 2024. 3

    2024 arXiv

  61. [69]

    Humancoser: Layered 3d human generation via semantic-aware diffusion model

    Yi Wang, Jian Ma, Ruizhi Shao, Qiao Feng, Yu-Kun Lai, and Kun Li. Humancoser: Layered 3d human generation via semantic-aware diffusion model. In 2024 IEEE Inter- national Symposium on Mixed and Augmented Reality (IS- MAR), pages 436–445. IEEE, 2024. 3

    2024

  62. [70]

    Tram: Global trajectory and motion of 3d humans from in- the-wild videos

    Yufu Wang, Ziyun Wang, Lingjie Liu, and Kostas Daniilidis. Tram: Global trajectory and motion of 3d humans from in- the-wild videos. In European Conference on Computer Vi- sion, pages 467–487. Springer, 2024. 3

    2024

  63. [71]

    Image quality assessment: from error visibility to 11 structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to 11 structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004. 7

    2004

  64. [72]

    Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distilla- tion

    Zhengyi Wang, Cheng Lu, Yikai Wang, Fan Bao, Chongxuan Li, Hang Su, and Jun Zhu. Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distilla- tion. Advances in Neural Information Processing Systems , 36:8406–8441, 2023. 3

    2023

  65. [73]

    Wonderhuman: Hallucinating unseen parts in dynamic 3d human reconstruction

    Zilong Wang, Zhiyang Dou, Yuan Liu, Cheng Lin, Xiao Dong, Yunhui Guo, Chenxu Zhang, Xin Li, Wenping Wang, and Xiaohu Guo. Wonderhuman: Hallucinating unseen parts in dynamic 3d human reconstruction. arXiv preprint arXiv:2502.01045, 2025. 3

    2025

  66. [74]

    Gomavatar: Efficient an- imatable human modeling from monocular video using gaussians-on-mesh

    Jing Wen, Xiaoming Zhao, Zhongzheng Ren, Alexander G Schwing, and Shenlong Wang. Gomavatar: Efficient an- imatable human modeling from monocular video using gaussians-on-mesh. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 2059–20...

  67. [75]

    Hu- mannerf: Free-viewpoint rendering of moving people from monocular video

    Chung-Yi Weng, Brian Curless, Pratul P Srinivasan, Jonathan T Barron, and Ira Kemelmacher-Shlizerman. Hu- mannerf: Free-viewpoint rendering of moving people from monocular video. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern Recognition , pages 162...

    2022

  68. [76]

    Show-o: One single transformer to unify multimodal understanding and generation

    Jinheng Xie, Weijia Mao, Zechen Bai, David Junhao Zhang, Weihao Wang, Kevin Qinghong Lin, Yuchao Gu, Zhijie Chen, Zhenheng Yang, and Mike Zheng Shou. Show-o: One single transformer to unify multimodal understanding and generation. arXiv preprint arXiv:2408.12528, 2024. 2, 3

    2024 arXiv

  69. [77]

    Sv4d: Dynamic 3d content generation with multi-frame and multi-view consistency

    Yiming Xie, Chun-Han Yao, Vikram V oleti, Huaizu Jiang, and Varun Jampani. Sv4d: Dynamic 3d content generation with multi-frame and multi-view consistency. arXiv preprint arXiv:2407.17470, 2024. 3

    2024 arXiv

  70. [78]

    Seeavatar: Pho- torealistic text-to-3d avatar generation with constrained ge- ometry and appearance

    Yuanyou Xu, Zongxin Yang, and Yi Yang. Seeavatar: Pho- torealistic text-to-3d avatar generation with constrained ge- ometry and appearance. arXiv preprint arXiv:2312.08889 ,

    arXiv

  71. [79]

    Magicanimate: Temporally consistent human im- age animation using diffusion model

    Zhongcong Xu, Jianfeng Zhang, Jun Hao Liew, Hanshu Yan, Jia-Wei Liu, Chenxu Zhang, Jiashi Feng, and Mike Zheng Shou. Magicanimate: Temporally consistent human im- age animation using diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Reco...

    2024

  72. [80]

    Diffusion 2: Dynamic 3d content generation via score composition of video and multi-view diffusion models

    Zeyu Yang, Zijie Pan, Chun Gu, and Li Zhang. Diffusion 2: Dynamic 3d content generation via score composition of video and multi-view diffusion models. arXiv preprint arXiv:2404.02148, 2024. 3

    2024 arXiv

  73. [81]

    Ip- adapter: Text compatible image prompt adapter for text-to- image diffusion models

    Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip- adapter: Text compatible image prompt adapter for text-to- image diffusion models. arXiv preprint arXiv:2308.06721,

    arXiv

  74. [82]

    Monohuman: Animatable human neu- ral field from monocular video

    Zhengming Yu, Wei Cheng, Xian Liu, Wayne Wu, and Kwan-Yee Lin. Monohuman: Animatable human neu- ral field from monocular video. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16943–16953, 2023. 2

    2023

  75. [83]

    Mip-splatting: Alias-free 3d gaussian splat- ting

    Zehao Yu, Anpei Chen, Binbin Huang, Torsten Sattler, and Andreas Geiger. Mip-splatting: Alias-free 3d gaussian splat- ting. Conference on Computer Vision and Pattern Recogni- tion (CVPR), 2024. 2

    2024

  76. [84]

    Avatarbooth: High-quality and customizable 3d human avatar generation

    Yifei Zeng, Yuanxun Lu, Xinya Ji, Yao Yao, Hao Zhu, and Xun Cao. Avatarbooth: High-quality and customizable 3d human avatar generation. arXiv preprint arXiv:2306.09864,

    arXiv

  77. [85]

    Gaussian vari- ation field diffusion for high-fidelity video-to-4d synthesis

    Bowen Zhang, Sicheng Xu, Chuxin Wang, Jiaolong Yang, Feng Zhao, Dong Chen, and Baining Guo. Gaussian vari- ation field diffusion for high-fidelity video-to-4d synthesis. arXiv preprint arXiv:2507.23785, 2025. 3, 7

    2025 arXiv

  78. [86]

    Avatarverse: High-quality & stable 3d avatar creation from text and pose

    Huichao Zhang, Bowen Chen, Hao Yang, Liao Qu, Xu Wang, Li Chen, Chao Long, Feida Zhu, Daniel Du, and Min Zheng. Avatarverse: High-quality & stable 3d avatar creation from text and pose. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 7124–7132, 2024. 3

    2024

  79. [87]

    4diffusion: Multi-view video dif- fusion model for 4d generation

    Haiyu Zhang, Xinyuan Chen, Yaohui Wang, Xihui Liu, Yun- hong Wang, and Yu Qiao. 4diffusion: Multi-view video dif- fusion model for 4d generation. Advances in Neural Infor- mation Processing Systems, 37:15272–15295, 2024. 3

    2024

  80. [88]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF international conference on computer vision, pages 3836–3847, 2023. 3

    2023

  81. [89]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 586–595, 2018. 7, 13

    2018

  82. [90]

    Structured local radiance fields for human avatar modeling

    Zerong Zheng, Han Huang, Tao Yu, Hongwen Zhang, Yan- dong Guo, and Yebin Liu. Structured local radiance fields for human avatar modeling. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 15893–15903, 2022. 1, 2, 6

    2022

  83. [91]

    Avatarrex: Real-time expressive full- body avatars

    Zerong Zheng, Xiaochen Zhao, Hongwen Zhang, Boning Liu, and Yebin Liu. Avatarrex: Real-time expressive full- body avatars. ACM Transactions on Graphics (TOG), 42(4): 1–19, 2023. 1, 2

    2023

  84. [92]

    Champ: Controllable and consistent human image animation with 3d parametric guidance

    Shenhao Zhu, Junming Leo Chen, Zuozhuo Dai, Yinghui Xu, Xun Cao, Yao Yao, Hao Zhu, and Siyu Zhu. Champ: Controllable and consistent human image animation with 3d parametric guidance. In European Conference on Computer Vision (ECCV), 2024. 3

    2024

  85. [93]

    Driv- able 3d gaussian avatars

    Wojciech Zielonka, Timur Bagautdinov, Shunsuke Saito, Michael Zollh ¨ofer, Justus Thies, and Javier Romero. Driv- able 3d gaussian avatars. arXiv preprint arXiv:2311.08581,

    arXiv

  86. [2023]

    back-view

    3 12 DevilSight: Augmenting Monocular Human Avatar Reconstruction through a Virtual Perspective Supplementary Material A. Implementation Details The avatar is finally optimized by: L = Lrgb + λLPIPSLLPIPS + λoffsetLoffset + λscaleLscale, (4) where Lrgb is the L1 photometric lo...

Pith tools

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