Pith. sign in

REVIEW 4 major objections 5 minor 81 references

ReMu: Reconstructing Multi-layer 3D Clothed Human from Image Layers

T0 review · 4 major / 5 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 ReMu reconstructs multi-layer 3D clothing from single-view image layers.

desk verdict A practical new capture protocol and a coherent pipeline for multi-layer garment reconstruction, but the 'penetration-free' claim leans on a 2D occlusion proxy rather than a real 3D intersection measure. read the letter →

arxiv 2508.01381 v1 pith:YI7UBYSD submitted 2025-08-02 cs.GR cs.CV

Onat Vuran , Hsuan-I Ho This is my paper · ORCID
classification cs.GRcs.CV
keywords multi-layergarmentreconstructionsingle-view3Dimagelayerscapturecollision-awareoptimizationneuralunsigneddistancefieldsinverselinearblendskinningtemplate-freelayeredclothedhumanavatars
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

ReMu claims that fully layered 3D clothing on a human body can be reconstructed from a set of ordinary single-view photographs, one per garment layer, without any garment templates and without category-specific training. The setup, called Image Layers, replaces expensive multi-view capture with a static RGB camera and a subject who puts on each layer in front of it. The paper further claims that the resulting garment meshes are nearly penetration-free, with Intersection Ratios below 0.5% in all tested categories, which is what makes them usable in downstream clothing simulation. If these claims hold, capturing a layered avatar for virtual try-on, animation, or simulation becomes a phone-camera task rather than a studio task.

What carries the argument

The load-bearing mechanism is a layered representation that combines triangle meshes with implicit neural unsigned distance fields (UDFs). Each garment layer is first a watertight mesh produced by an off-the-shelf single-view reconstruction model; inverse linear blend skinning moves all layers into one canonical body pose; a collision-aware displacement step moves each garment vertex along the normal line of its nearest body point, placing non-garment vertices exactly on the inner layer's surface and penetrated garment vertices a small thickness ($\epsilon$) outside it; and finally a per-layer UDF network, trained to output the unsigned distance to the garment surface, is meshed by Marching Cubes to yield smooth, complete boundaries.

What would settle it

Take one 4D-DRESS subject or a real static-camera capture and deliberately vary body pose between image layers, for example by shifting weight or bending an arm between the inner-layer and outer-layer photographs, then run the full pipeline and measure the Intersection Ratio on the canonical body; if the ratio rises substantially above the reported sub-0.5% level, the inverse-LBS registration premise is falsified.

Watch

Extended reading notes

Core claim

The central discovery is that the unobserved inner layers that make multi-layer clothing reconstruction ill-posed need not be hallucinated: if the same subject is photographed once per clothing layer, each layer is directly visible in its own image. ReMu registers these per-layer observations by fitting a body model per image, deforming every reconstructed watertight mesh into a shared canonical T/A pose with inverse linear blend skinning, and then running an inner-to-outer penetration removal that pushes garment vertices outside the previous layer only where they belong to the garment. A final stage fits a neural unsigned distance field to each layer to smooth segmentation noise and fragmented boundaries. The result is a template-free, category-agnostic pipeline whose reconstructed garments are claimed to be nearly collision-free and competitive with category-specific methods trained on synthetic data.

Load-bearing premise

The load-bearing premise is that the subject is captured in a near-static T/A pose for every image layer, so that inverse linear blend skinning can register all garment meshes to a single canonical body; if the pose or body shape drifts between captures, alignment breaks down and the reported penetration-freeness no longer holds.

Editorial extensions

If this is right

  • A layered, physically usable clothed avatar can be captured with one static RGB camera and one photograph per garment layer, eliminating calibrated multi-view rigs.
  • Because the pipeline is template-free and category-agnostic, garment types absent from any training set, such as long coats and dresses, can be reconstructed without adaptation.
  • With Intersection Ratios below 0.5% in the reported benchmark, the reconstructed layers can be handed to physics-based simulation tools, as demonstrated with Blender.
  • In the same-input comparison, the method reaches 3D geometry metrics comparable to or better than category-specific baselines on lower and inner garments, and comparable on outer garments.

Reading between the lines

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

  • A natural extension the authors leave implicit is to replace the fixed T/A-pose requirement with a video sequence in which the subject moves between layers, letting non-rigid registration handle pose change instead of inverse LBS; the paper's own limitation statement points exactly there.
  • Because the UDF refinement fits each layer independently, nothing in that stage enforces collision-freeness, and the ablation indeed shows a small Intersection Ratio increase after refinement; a collision-aware UDF objective is a directly testable follow-up.
  • Image Layers data could be mined to train feed-forward models that predict multi-layer garments directly from a single image, using the aligned penetration-free layers as ground truth; the paper notes large-scale data collection as future work.
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 introduces ReMu, a training-free and template-free pipeline that reconstructs multi-layer 3D clothing from “Image Layers”: a set of single-view RGB images of a subject wearing successive layers of clothing in a near-static T/A pose. The pipeline first fits an SMPL-X body and uses SITH to recover a watertight clothed mesh for each image, segments the target garment via multi-view renders and Grounded SAM voting, registers the garment meshes to a shared canonical body with inverse linear blend skinning, removes inter-layer penetrations by displacing vertices along body normals, and finally refines the surfaces by fitting neural unsigned distance fields. The authors evaluate on 12 subjects from the 4D-DRESS dataset using Chamfer distance, normal consistency, and a newly proposed Intersection Ratio, reporting improvements over SMPLicit, ClothWild, ISP, and extended “+” variants, together with an ablation study and an in-the-wild iPhone capture demo.

Significance. If the claims hold, the Image Layers capture protocol is a genuinely attractive idea: multi-layer garment geometry for downstream simulation is obtained from a single static RGB camera without category-specific training data. The paper gives clean ablations showing that the penetration-removal step itself is effective on the proposed proxy metric, and the in-the-wild demo indicates practical feasibility. The main contributions are empirical rather than theoretical; the manuscript does not include code or machine-checked proofs, so reproducibility rests on the textual description and off-the-shelf components. The qualitative generality to unseen garment categories such as long coats and dresses is a strength, as is the simple capture setup. However, the headline “nearly penetration-free” claim is not yet established because the Intersection Ratio metric is a two-view occlusion proxy rather than a geometric intersection measure, and the quantitative evaluation reports no variance or significance testing over its 12 subjects. These issues gate the central claims of the paper.

major comments (4)
  1. [Sec. 4.1, IR definition; Figs. 5 and 6] The Intersection Ratio defined in Sec. 4.1 as (A - A_hat)/A from frontal and back rasterized views is an occlusion proxy, not a geometric intersection statistic. It counts any outer-garment area hidden by the body or by inner garments in those two views, including regions that are legitimately open to the body (e.g., an unzipped coat or a flared skirt), and it cannot detect interpenetrations that are invisible from the two chosen viewpoints. Since the abstract's central claim is that ReMu reconstructs “nearly penetration-free 3D clothed humans,” and since the motivation is downstream physical simulation, the relevant quantity is actual 3D mesh intersection volume or area between each garment and the body/inner stack and between adjacent garments. Please report such a geometric intersection measure, or validate the two-view IR against it and quantify the discrepancy. The comparison is further complicated by the Fig. 5 caption, which states that the proposed method and several baselines use double-sided meshes while ISP and the ground-truth reference are single-sided, so the rasterized areas entering IR are not computed on comparable surface representations.
  2. [Sec. 4.1, Tables 2 and 3] All quantitative comparisons are aggregate means over only 12 subjects, with no standard deviations, per-subject breakdowns, or significance tests. The claimed improvements in Chamfer distance and normal consistency over the baselines and the trade-offs in the ablation study cannot be assessed for statistical reliability from the reported numbers. Please report per-subject results and run paired significance tests for the primary method-vs-baseline comparisons and for the ablations. The supplementary statement in Sec. 7.1 that CD was omitted because it “was not statistically significant” is also unsupported, since no statistical test is described anywhere in the paper.
  3. [Sec. 3.3 and Limitations] The UDF refinement step is supervised only by distances to the post-penetration-removal meshes and contains no collision or penetration term, as the Limitations paragraph itself acknowledges (“our UDF fitting processing does not incorporate other constraints”). Table 3 shows that the full pipeline increases IR in every category relative to the penetration-removal stage alone (e.g., outer IR goes from 0.060% to 0.485%). With the metric issue above, this means the claim that the final output is “nearly penetration-free” is not demonstrated. If a true 3D intersection metric is adopted, the refinement step must be evaluated with it; if refinement reintroduces geometric intersections, the loss should be augmented with a collision penalty or the residual intersections should be bounded explicitly.
  4. [Sec. 4.1 and Discussion/Limitations] The quantitative protocol provides ground-truth poses and labels, and the Limitations section states that the method is tailored to subjects in the T/A pose. The evaluation therefore does not test sensitivity to pose or body-shape differences between image layers, which is the most likely failure mode of the inverse-LBS registration and of the penetration-removal step in real capture. Please add a controlled robustness study that perturbs the pose or body shape between layers and reports CD, NC, and the penetration metric as a function of misalignment, and qualify the abstract and introduction claims to the T/A-pose setting if the degradation is substantial.
minor comments (5)
  1. [Tables 2 and 4 (method names)] The method name is spelled inconsistently as “SMPLlicit” and “Smplicit+” in Tables 2 and in the text, while the reference [12] is “SMPLicit”; please unify the spelling throughout.
  2. [Eq. (5) and Supplementary Eq. (8)] The UDF fitting loss in the main text is written without the clamping operation and with a different normalization than the supplementary formulation; the sampling distribution for x_k is also unspecified. Please align the two equations and give the exact sampling procedure in the main paper.
  3. [Fig. 5] The figure compares meshes with differing sidedness and uses different canonical body models for different methods (SMPL-X for ReMu, SMPL for baselines); this makes the visual comparison and the IR numbers partly dependent on representation choices. Please render all methods with a consistent back-face convention, or state clearly why the comparison is unaffected.
  4. [Supp. Sec. 7.1] The claim that CD was omitted from the positional-encoding ablation because it was “not statistically significant” is not supported by any reported significance test; either provide the test or report the CD values.
  5. [Sec. 3.1] The 3D segmentation step uses 60 rendered views, voting, and Grounded SAM with a language prompt; the paper gives no ablation on the number of views, the mask threshold, or the effect of segmentation errors on the final garment boundaries, which would help readers gauge the robustness of the pipeline to this component.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: ReMu is evaluated against external 4D-DRESS ground truth; self-cited SiTH and 4D-DRESS are tooling and benchmark, and the IR caveats are validity concerns, not definitional reductions.

full rationale

ReMu is an empirical reconstruction pipeline, not a formal derivation, and no step could be exhibited where an output is defined in terms of the quantity it is claimed to predict, or where a fitted parameter is renamed as a prediction. The garment meshes are produced by SiTH reconstruction, SAM-based segmentation, inverse LBS alignment, geometric penetration removal, and UDF re-fitting; the UDF loss (Eq. 5 and Eq. 8) supervises the network with the pipeline's own intermediate meshes, but this is an internal smoothing/reparametrization step, and the benchmark scores the final Marching Cubes meshes against 4D-DRESS ground-truth scans rather than against the fit target. The hand-set scalars (epsilon = 2 mm, Marching Cubes threshold = 3 mm, positional-encoding count) are optimization and meshing constants, not parameters fitted to the benchmark targets. The self-citations to SiTH and 4D-DRESS are a published reconstruction tool and a public benchmark by a co-author; neither is invoked as an unverified uniqueness theorem or as the reason the reported IR values are low. The IR metric in Sec. 4.1 is a two-view rasterization occlusion proxy rather than a true 3D intersection measure, and the double-sided versus single-sided mesh caveat in Fig. 5 complicates direct comparison; these are correctness and validity limitations, not circularity. The stated T/A-pose limitation further scopes the claim rather than making it tautological. Verdict: no significant circularity.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the reliability of off-the-shelf components (SITH, SMPL-X, GroundedSAM) and on the assumption of a static T/A pose. No new physical entities are introduced. The method's hyperparameters are hand-set, not fitted to the evaluation targets.

free parameters (6)
  • thickness parameter epsilon = 2 mm
    Controls how far a penetrating garment vertex is moved outside the previous layer in Eq. (3). Chosen by hand; directly affects the final Intersection Ratio.
  • Marching Cubes threshold = 3 mm
    Used to extract a double-sided garment mesh from the UDF. Chosen to give garments a small thickness; affects geometry thickness and the final mesh.
  • number of positional encodings = N = 4
    Balances wrinkle detail and smoothness. Ablation in the supplementary shows N=0 loses detail and N=8 captures noisy high-frequency features.
  • MLP architecture and learning rate = 4 layers, hidden dims 128/256/256/128, LR 1e-4
    Standard hyperparameters chosen by experimentation for the neural UDF fitting.
  • number of rendered views = 60
    60 circularly sampled views (10-degree horizontal separation, 30-degree upper/lower) used for 3D segmentation via GroundedSAM.
  • off-surface point noise sigma = 0.01
    Gaussian displacement used in sampling points near the garment surface for UDF training.
assumptions (5)
  • domain assumption SITH reconstructs an accurate watertight clothed-human mesh from a single RGB image.
    The entire pipeline builds on SITH output meshes (Sec. 3.1). If SITH fails on loose garments or complex poses, the downstream steps inherit the error.
  • domain assumption SMPL-X body fitting and inverse linear blend skinning correctly align garment meshes to a canonical T-pose body.
    Inverse LBS in Eq. (2) assumes the skinning weights from the nearest SMPL-X vertex are valid for the reconstructed mesh. This is a standard assumption but untested for off-skin vertices.
  • domain assumption GroundedSAM 2D segmentation, when reprojected to 3D via voting, yields correct garment labels.
    The 3D garment area Sk is defined by this voting procedure (Sec. 3.1, supplementary). Segmentation errors propagate into the garment boundary.
  • domain assumption The subject is in a near-static T/A pose across all Image Layers.
    The method requires the pose to be consistent enough for inverse LBS to snap all layers to the same canonical body. This is acknowledged in the Limitations section.
  • domain assumption A neural unsigned distance field can represent garment surfaces, and Marching Cubes with a small threshold yields a plausible double-sided mesh.
    The refinement stage in Sec. 3.3 relies on UDF fitting and MC extraction. The supplementary details the choice of threshold, but the faithfulness of the representation is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ReMu: Reconstructing Multi-layer 3D Clothed Human from Image Layers." pith.science (2026). https://pith.science/paper/YI7UBYSD

@misc{pith2026250801381,
  author       = {Pith},
  title        = {Pith review of: ReMu: Reconstructing Multi-layer 3D Clothed Human from Image Layers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YI7UBYSD}},
  note         = {Machine review of arXiv:2508.01381}
}
read the original abstract

The reconstruction of multi-layer 3D garments typically requires expensive multi-view capture setups and specialized 3D editing efforts. To support the creation of life-like clothed human avatars, we introduce ReMu for reconstructing multi-layer clothed humans in a new setup, Image Layers, which captures a subject wearing different layers of clothing with a single RGB camera. To reconstruct physically plausible multi-layer 3D garments, a unified 3D representation is necessary to model these garments in a layered manner. Thus, we first reconstruct and align each garment layer in a shared coordinate system defined by the canonical body pose. Afterwards, we introduce a collision-aware optimization process to address interpenetration and further refine the garment boundaries leveraging implicit neural fields. It is worth noting that our method is template-free and category-agnostic, which enables the reconstruction of 3D garments in diverse clothing styles. Through our experiments, we show that our method reconstructs nearly penetration-free 3D clothed humans and achieves competitive performance compared to category-specific methods. Project page: https://eth-ait.github.io/ReMu/

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

81 extracted references · 76 canonical work pages

  1. [1]

    Photorealistic monocular 3d reconstruction of humans wear- ing clothing

    Thiemo Alldieck, Mihai Zanfir, and Cristian Sminchisescu. Photorealistic monocular 3d reconstruction of humans wear- ing clothing. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 2, 3

    2022

  2. [2]

    CloSe: A 3D clothing segmentation dataset and model

    Dimitrije Anti ´c, Garvita Tiwari, Batuhan Ozcomlekci, Ric- cardo Marin, and Gerard Pons-Moll. CloSe: A 3D clothing segmentation dataset and model. In International Confer- ence on 3D Vision (3DV), 2024. 3

    work page 2024

  3. [3]

    4dhumanoutfit: a multi-subject 4d dataset of human motion sequences in varying outfits ex- hibiting large displacements

    Matthieu Armando, Laurence Boissieux, Edmond Boyer, Jean-Sebastien Franco, Martin Humenberger, Christophe Legras, Vincent Leroy, Mathieu Marsot, Julien Pansiot, Sergi Pujades, Rim Rekik, Gregory Rogez, Anilkumar Swamy, and Stefanie Wuhrer. 4dhumanoutfit: a multi-subject 4d dataset of human motion sequences in varying outfits ex- hibiting large displacemen...

    work page 2023

  4. [4]

    Multi-garment net: Learning to dress 3d people from images

    Bharat Lal Bhatnagar, Garvita Tiwari, Christian Theobalt, and Gerard Pons-Moll. Multi-garment net: Learning to dress 3d people from images. In Proceedings of the IEEE Interna- tional Conference on Computer Vision (ICCV), 2019. 3

    work page 2019

  5. [5]

    Chat- garment: Garment estimation, generation and editing via large language models

    Siyuan Bian, Chenghao Xu, Yuliang Xiu, Artur Grigorev, Zhen Liu, Cewu Lu, Michael J Black, and Yao Feng. Chat- garment: Garment estimation, generation and editing via large language models. In Proceedings of the IEEE Confer- ence on Computer Vision and Pattern Recognition (CVPR) , pages 2924–2934, 2025. 3

    work page 2025

  6. [6]

    Federica Bogo, Angjoo Kanazawa, Christoph Lassner, Peter Gehler, Javier Romero, and Michael J. Black. Keep it SMPL: Automatic estimation of 3D human pose and shape from a single image. In ECCV. Springer, 2016. 3

    work page 2016

  7. [7]

    Smpler-x: Scaling up expressive human pose and shape estimation

    Zhongang Cai, Wanqi Yin, Ailing Zeng, Chen Wei, Qing- ping Sun, Yanjun Wang, Hui En Pang, Haiyi Mei, Mingyuan Zhang, Lei Zhang, Chen Change Loy, Lei Yang, and Ziwei Liu. Smpler-x: Scaling up expressive human pose and shape estimation. In Advances in Neural Information Processing Systems (NeurIPS), 2023. 3, 15

    work page 2023

  8. [8]

    Z. Cao, G. Hidalgo Martinez, T. Simon, S. Wei, and Y . A. Sheikh. Openpose: Realtime multi-person 2d pose estima- tion using part affinity fields. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019. 3

    work page 2019

Show all 81 references
  1. [9]

    Animatable neural radiance fields from monocular rgb videos, 2021

    Jianchuan Chen, Ying Zhang, Di Kang, Xuefei Zhe, Linchao Bao, Xu Jia, and Huchuan Lu. Animatable neural radiance fields from monocular rgb videos, 2021. 2

    2021

  2. [10]

    Neural unsigned distance fields for implicit function learning

    Julian Chibane, Aymen Mir, and Gerard Pons-Moll. Neural unsigned distance fields for implicit function learning. InAd- vances in Neural Information Processing Systems (NeurIPS),

  3. [11]

    High-quality streamable free-viewpoint video

    Alvaro Collet, Ming Chuang, Pat Sweeney, Don Gillett, Den- nis Evseev, David Calabrese, Hugues Hoppe, Adam Kirk, and Steve Sullivan. High-quality streamable free-viewpoint video. ACM Transactions on Graphics (TOG), 2015. 2

    2015

  4. [12]

    Smplicit: Topology-aware generative model for clothed people

    Enric Corona, Albert Pumarola, Guillem Aleny `a, Ger- ard Pons-Moll, and Francesc Moreno-Noguer. Smplicit: Topology-aware generative model for clothed people. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021. 2, 3, 6

    2021

  5. [13]

    Structured 3d features for reconstructing relightable and an- imatable avatars

    Enric Corona, Mihai Zanfir, Thiemo Alldieck, Eduard Gabriel Bazavan, Andrei Zanfir, and Cristian Sminchisescu. Structured 3d features for reconstructing relightable and an- imatable avatars. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR...

    2023

  6. [14]

    Deepgarment: 3d garment shape estimation from a single image

    R Dan ˇeˇrek, Endri Dibra, Cengiz ¨Oztireli, Remo Ziegler, and Markus Gross. Deepgarment: 3d garment shape estimation from a single image. In Computer Graphics Forum. Wiley Online Library, 2017. 3

    2017

  7. [15]

    Perfor- mance capture from sparse multi-view video

    Edilson de Aguiar, Carsten Stoll, Christian Theobalt, Naveed Ahmed, Hans-Peter Seidel, and Sebastian Thrun. Perfor- mance capture from sparse multi-view video. ACM Trans- actions on Graphics (TOG), 2008. 2

    2008

  8. [16]

    DrapeNet: Garment Generation and Self- Supervised Draping

    Luca De Luigi, Ren Li, Benoit Guillard, Mathieu Salzmann, and Pascal Fua. DrapeNet: Garment Generation and Self- Supervised Draping. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 3

    2023

  9. [17]

    Tela: Text to layer-wise 3d clothed human generation

    Junting Dong, Qi Fang, Zehuan Huang, Xudong Xu, Jingbo Wang, Sida Peng, and Bo Dai. Tela: Text to layer-wise 3d clothed human generation. In Proceedings of the Euro- pean Conference on Computer Vision (ECCV), pages 19–36,

  10. [18]

    Pina: Learning a personalized implicit neural avatar from a single rgb-d video sequence

    Zijian Dong, Chen Guo, Jie Song, Xu Chen, Andreas Geiger, and Otmar Hilliges. Pina: Learning a personalized implicit neural avatar from a single rgb-d video sequence. In CVPR,

  11. [19]

    Laga: Layered 3d avatar genera- tion and customization via gaussian splatting

    Jia Gong, Shenyu Ji, Lin Geng Foo, Kang Chen, Hossein Rahmani, and Jun Liu. Laga: Layered 3d avatar genera- tion and customization via gaussian splatting. arXiv preprint arXiv:2405.12663, 2024. 3

    2024 arXiv

  12. [20]

    Black, and Otmar Hilliges

    Artur Grigorev, Bernhard Thomaszewski, Michael J. Black, and Otmar Hilliges. Hood: Hierarchical graphs for general- ized modelling of clothing dynamics. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion (CVPR), 2023. 2

    2023

  13. [21]

    ContourCraft: Learning to resolve intersections in neural multi-garment simulations

    Artur Grigorev, Giorgio Becherini, Michael Black, Ot- mar Hilliges, and Bernhard Thomaszewski. ContourCraft: Learning to resolve intersections in neural multi-garment simulations. In ACM SIGGRAPH 2024 Conference Papers, pages 1–10, 2024. 2

    2024

  14. [22]

    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 Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 2

    2023

  15. [23]

    Reloo: Reconstructing humans dressed in loose garments from monocular video in the wild

    Chen Guo, Tianjian Jiang, Manuel Kaufmann, Chengwei Zheng, Julien Valentin, Jie Song, and Otmar Hilliges. Reloo: Reconstructing humans dressed in loose garments from monocular video in the wild. In Proceedings of the Euro- pean Conference on Computer Vision (ECCV), 2024. 3

    2024

  16. [24]

    Vid2avatar-pro: Authentic avatar from videos in the wild via universal prior

    Chen Guo, Junxuan Li, Yash Kant, Yaser Sheikh, Shunsuke Saito, and Chen Cao. Vid2avatar-pro: Authentic avatar from videos in the wild via universal prior. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion (CVPR), 2025. 2

    2025

  17. [25]

    High-fidelity 3d human digitization from single 2k resolution images

    Sang-Hun Han, Min-Gyu Park, Ju Hong Yoon, Ju-Mi Kang, Young-Jae Park, and Hae-Gon Jeon. High-fidelity 3d human digitization from single 2k resolution images. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 2, 3

    2023

  18. [26]

    Arch++: Animation-ready clothed human recon- struction revisited

    Tong He, Yuanlu Xu, Shunsuke Saito, Stefano Soatto, and Tony Tung. Arch++: Animation-ready clothed human recon- struction revisited. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2021. 2, 3

    2021

  19. [27]

    Deep fashion3d: A dataset and benchmark for 3d garment recon- struction from single images

    Zhu Heming, Cao Yu, Jin Hang, Chen Weikai, Du Dong, Wang Zhangye, Cui Shuguang, and Han Xiaoguang. Deep fashion3d: A dataset and benchmark for 3d garment recon- struction from single images. In Proceedings of the Eu- ropean Conference on Computer Vision (ECCV) . Springer,

  20. [28]

    Learn- ing locally editable virtual humans

    Hsuan-I Ho, Lixin Xue, Jie Song, and Otmar Hilliges. Learn- ing locally editable virtual humans. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion (CVPR), 2023. 2

    2023

  21. [29]

    Sith: Single- view textured human reconstruction with image-conditioned diffusion

    Hsuan-I Ho, Jie Song, and Otmar Hilliges. Sith: Single- view textured human reconstruction with image-conditioned diffusion. In Proceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR), 2024. 2, 3, 6, 13

    2024

  22. [30]

    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 Conference on Computer Vision and Pattern...

    2024

  23. [31]

    Arch: Animatable reconstruction of clothed hu- mans

    Zeng Huang, Yuanlu Xu, Christoph Lassner, Hao Li, and Tony Tung. Arch: Animatable reconstruction of clothed hu- mans. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020. 2, 3

    2020

  24. [32]

    Bcnet: Learning body and cloth shape from a single image

    Boyi Jiang, Juyong Zhang, Yang Hong, Jinhao Luo, Ligang Liu, and Hujun Bao. Bcnet: Learning body and cloth shape from a single image. In Proceedings of the European Con- ference on Computer Vision (ECCV). Springer, 2020. 2, 3

    2020

  25. [33]

    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 Conference on Com- puter Vision and Pattern Recognition (CVPR), 2022. 2

    2022

  26. [34]

    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 Conference on Com- puter Vision and Pattern Recognition (CVPR), 2023

    2023

  27. [35]

    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 Proceedings of the European confer- ence on computer vision (ECCV), 2022. 2

    2022

  28. [36]

    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 (TOG), 2023. 2

    2023

  29. [37]

    Gala: Generating animatable layered assets from a sin- gle scan

    Taeksoo Kim, Byungjun Kim, Shunsuke Saito, and Hanbyul Joo. Gala: Generating animatable layered assets from a sin- gle scan. In Proceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR), 2024. 3, 13

    2024

  30. [38]

    Berg, Wan-Yen Lo, Piotr Dollar, and Ross Girshick

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C. Berg, Wan-Yen Lo, Piotr Dollar, and Ross Girshick. Segment anything. In Proceedings of the IEEE International Conference on Computer Vision (I...

    2023

  31. [39]

    Pshuman: Photorealistic single-view human reconstruction using cross-scale diffusion

    Peng Li, Wangguandong Zheng, Yuan Liu, Tao Yu, Yang- guang Li, Xingqun Qi, Mengfei Li, Xiaowei Chi, Siyu Xia, Wei Xue, et al. Pshuman: Photorealistic single-view human reconstruction using cross-scale diffusion. arXiv preprint arXiv:2409.10141, 2024. 2, 3

    2024 arXiv

  32. [40]

    ISP: Multi-Layered Garment Draping with Implicit Sewing Patterns

    Ren Li, Benoit Guillard, and Pascal Fua. ISP: Multi-Layered Garment Draping with Implicit Sewing Patterns. In Ad- vances in Neural Information Processing Systems (NeurIPS),

  33. [41]

    Garment recovery with shape and deformation priors

    Ren Li, Corentin Dumery, Beno ˆıt Guillard, and Pascal Fua. Garment recovery with shape and deformation priors. InPro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1586–1595, 2024. 3

    2024

  34. [42]

    High-Fidelity Clothed Avatar Reconstruction from a Single Image

    Tingting Liao, Xiaomei Zhang, Yuliang Xiu, Hongwei Yi, Xudong Liu, Guo-Jun Qi, Yong Zhang, Xuan Wang, Xi- angyu Zhu, and Zhen Lei. High-Fidelity Clothed Avatar Reconstruction from a Single Image. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tio...

    2023

  35. [43]

    Layga: Layered gaussian avatars for 10 animatable clothing transfer

    Siyou Lin, Zhe Li, Zhaoqi Su, Zerong Zheng, Hongwen Zhang, and Yebin Liu. Layga: Layered gaussian avatars for 10 animatable clothing transfer. In SIGGRAPH Conference Pa- pers, 2024. 2, 3

    2024

  36. [44]

    A point-cloud- based multiview stereo algorithm for free-viewpoint video

    Yebin Liu, Qionghai Dai, and Wenli Xu. A point-cloud- based multiview stereo algorithm for free-viewpoint video. IEEE Visualization and Computer Graphics (TVCG) , 2009. 2

    2009

  37. [45]

    Garverselod: High-fidelity 3d garment recon- struction from a single in-the-wild image using a dataset with levels of details

    Zhongjin Luo, Haolin Liu, Chenghong Li, Wanghao Du, Zirong Jin, Wanhu Sun, Yinyu Nie, Weikai Chen, and Xi- aoguang Han. Garverselod: High-fidelity 3d garment recon- struction from a single in-the-wild image using a dataset with levels of details. ACM Transactions on Graphics (TOG) ,

  38. [46]

    Qianli Ma, Jinlong Yang, Siyu Tang, and Michael J. Black. The power of points for modeling humans in clothing. In Proceedings of the IEEE International Conference on Com- puter Vision (ICCV), 2021. 2

    2021

  39. [47]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis. In Proceedings of the European Conference on Com- puter Vision (ECCV), 2020. 2, 14

    2020

  40. [48]

    3d clothed human reconstruction in the wild

    Gyeongsik Moon, Hyeongjin Nam, Takaaki Shiratori, and Kyoung Mu Lee. 3d clothed human reconstruction in the wild. In Proceedings of the European Conference on Com- puter Vision (ECCV), 2022. 2, 3, 6

    2022

  41. [49]

    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 of the IEEE Conference on Computer Vision and Pattern Recognitio...

    2019

  42. [50]

    Im- plicit neural representations with structured latent codes for human body modeling

    Sida Peng, Chen Geng, Yuanqing Zhang, Yinghao Xu, Qian- qian Wang, Qing Shuai, Xiaowei Zhou, and Hujun Bao. Im- plicit neural representations with structured latent codes for human body modeling. IEEE Transactions on Pattern Anal- ysis and Machine Intelligence (TPAMI), 2023. 2

    2023

  43. [51]

    Rec-mv: Reconstruct- ing 3d dynamic cloth from monocular videos

    Lingteng Qiu, Guanying Chen, Jiapeng Zhou, Mutian Xu, Junle Wang, and Xiaoguang Han. Rec-mv: Reconstruct- ing 3d dynamic cloth from monocular videos. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 3

    2023

  44. [52]

    Grounded sam: Assembling open-world models for diverse visual tasks,

    Tianhe Ren, Shilong Liu, Ailing Zeng, Jing Lin, Kun- chang Li, He Cao, Jiayu Chen, Xinyu Huang, Yukang Chen, Feng Yan, Zhaoyang Zeng, Hao Zhang, Feng Li, Jie Yang, Hongyang Li, Qing Jiang, and Lei Zhang. Grounded sam: Assembling open-world models for diverse visual tasks,

  45. [53]

    Pifu: Pixel-aligned implicit function for high-resolution clothed human digitiza- tion

    Shunsuke Saito, Zeng Huang, Ryota Natsume, Shigeo Mor- ishima, Angjoo Kanazawa, and Hao Li. Pifu: Pixel-aligned implicit function for high-resolution clothed human digitiza- tion. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2019. 2, 3

    2019

  46. [54]

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

    Shunsuke Saito, Tomas Simon, Jason Saragih, and Hanbyul Joo. Pifuhd: Multi-level pixel-aligned implicit function for high-resolution 3d human digitization. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion (CVPR), 2020. 2, 3

    2020

  47. [55]

    SNUG: Self-Supervised Neural Dynamic Garments

    Igor Santesteban, Miguel A Otaduy, and Dan Casas. SNUG: Self-Supervised Neural Dynamic Garments. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 2

    2022

  48. [56]

    Otaduy, Nils Thuerey, and Dan Casas

    Igor Santesteban, Miguel A. Otaduy, Nils Thuerey, and Dan Casas. ULNeF: Untangled layered neural fields for mix-and- match virtual try-on. InAdvances in Neural Information Pro- cessing Systems (NeurIPS), 2022. 3

    2022

  49. [57]

    X-avatar: Ex- pressive human avatars

    Kaiyue Shen, Chen Guo, Manuel Kaufmann, Juan Zarate, Julien Valentin, Jie Song, and Otmar Hilliges. X-avatar: Ex- pressive human avatars. In Proceedings of the IEEE Confer- ence on Computer Vision and Pattern Recognition (CVPR) ,

  50. [58]

    Canonical- fusion: Generating drivable 3d human avatars from multi- ple images

    Jisu Shin, Junmyeong Lee, Seongmin Lee, Min-Gyu Park, Ju-Mi Kang, Ju Hong Yoon, and Hae-Gon Jeon. Canonical- fusion: Generating drivable 3d human avatars from multi- ple images. In Proceedings of the European Conference on Computer Vision (ECCV), pages 38–56. Springer, 2025. 2

    2025

  51. [59]

    Model-based multiple view reconstruction of people

    Jonathan Starck and Adrian Hilton. Model-based multiple view reconstruction of people. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2003. 2

    2003

  52. [60]

    Surface capture for performance-based animation

    Jonathan Starck and Adrian Hilton. Surface capture for performance-based animation. IEEE computer graphics and applications, 2007. 2

    2007

  53. [61]

    State of the art on neural rendering

    Ayush Tewari, Ohad Fried, Justus Thies, Vincent Sitzmann, Stephen Lombardi, Kalyan Sunkavalli, Ricardo Martin- Brualla, Tomas Simon, Jason Saragih, Matthias Nießner, et al. State of the art on neural rendering. In Computer Graphics Forum, 2020. 2

    2020

  54. [62]

    Advances in neural rendering

    Ayush Tewari, Justus Thies, Ben Mildenhall, Pratul Srini- vasan, Edgar Tretschk, Wang Yifan, Christoph Lassner, Vin- cent Sitzmann, Ricardo Martin-Brualla, Stephen Lombardi, et al. Advances in neural rendering. In Computer Graphics Forum, 2022. 2

    2022

  55. [63]

    Sizer: A dataset and model for parsing 3d clothing and learning size sensitive 3d clothing

    Garvita Tiwari, Bharat Lal Bhatnagar, Tony Tung, and Ger- ard Pons-Moll. Sizer: A dataset and model for parsing 3d clothing and learning size sensitive 3d clothing. In Pro- ceedings of the European Conference on Computer Vision (ECCV). Springer, 2020. 3

    2020

  56. [64]

    High resolution 3d shape texture from multiple videos

    Vagia Tsiminaki, Jean-S ´ebastien Franco, and Edmond Boyer. High resolution 3d shape texture from multiple videos. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2014. 2

    2014

  57. [65]

    4d-dress: A 4d dataset of real-world human clothing with se- mantic annotations

    Wenbo Wang, Hsuan-I Ho, Chen Guo, Boxiang Rong, Ar- tur Grigorev, Jie Song, Juan Jose Zarate, and Otmar Hilliges. 4d-dress: A 4d dataset of real-world human clothing with se- mantic annotations. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (...

    2024

  58. [66]

    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

  59. [67]

    Srinivasan, Jonathan T

    Chung-Yi Weng, Brian Curless, Pratul P. Srinivasan, Jonathan T. Barron, and Ira Kemelmacher-Shlizerman. Hu- manNeRF: Free-viewpoint rendering of moving people from 11 monocular video. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 2

    2022

  60. [68]

    Neural fields in visual computing and beyond

    Yiheng Xie, Towaki Takikawa, Shunsuke Saito, Or Litany, Shiqin Yan, Numair Khan, Federico Tombari, James Tomp- kin, Vincent Sitzmann, and Srinath Sridhar. Neural fields in visual computing and beyond. InComputer Graphics Forum,

  61. [69]

    Yuliang Xiu, Jinlong Yang, Dimitrios Tzionas, and Michael J. Black. ICON: Implicit Clothed humans Obtained from Normals. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 2, 3, 6

    2022

  62. [70]

    Yuliang Xiu, Jinlong Yang, Xu Cao, Dimitrios Tzionas, and Michael J. Black. ECON: Explicit Clothed humans Opti- mized via Normal integration. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 2, 3

    2023

  63. [71]

    Deep learning technique for human parsing: A survey and outlook

    Lu Yang, Wenhe Jia, Shan Li, and Qing Song. Deep learning technique for human parsing: A survey and outlook. arXiv preprint arXiv:2301.00394, 2023. 3

    2023 arXiv

  64. [72]

    Shan Yang, Zherong Pan, Tanya Amert, Ke Wang, Licheng Yu, Tamara Berg, and Ming C. Lin. Physics-inspired gar- ment recovery from a single-view image. ACM Transactions on Graphics (TOG), 2018. 3

    2018

  65. [73]

    Function4d: Real-time human vol- umetric capture from very sparse consumer rgbd sensors

    Tao Yu, Zerong Zheng, Kaiwen Guo, Pengpeng Liu, Qiong- hai Dai, and Yebin Liu. Function4d: Real-time human vol- umetric capture from very sparse consumer rgbd sensors. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021. 3

    2021

  66. [74]

    Sifu: Side- view conditioned implicit function for real-world usable clothed human reconstruction

    Zechuan Zhang, Zongxin Yang, and Yi Yang. Sifu: Side- view conditioned implicit function for real-world usable clothed human reconstruction. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2024. 2, 3

    2024

  67. [75]

    Humannerf: Efficiently gen- erated human radiance field from sparse inputs

    Fuqiang Zhao, Wei Yang, Jiakai Zhang, Pei Lin, Yingliang Zhang, Jingyi Yu, and Lan Xu. Humannerf: Efficiently gen- erated human radiance field from sparse inputs. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 2

    2022

  68. [76]

    Black, and Otmar Hilliges

    Yufeng Zheng, Wang Yifan, Gordon Wetzstein, Michael J. Black, and Otmar Hilliges. Pointavatar: Deformable point- based head avatars from videos. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 2

    2023

  69. [77]

    Physavatar: Learning the physics of dressed 3d avatars from visual observations, 2024

    Yang Zheng, Qingqing Zhao, Guandao Yang, Wang Yi- fan, Donglai Xiang, Florian Dubost, Dmitry Lagun, Thabo Beeler, Federico Tombari, Leonidas Guibas, and Gordon Wetzstein. Physavatar: Learning the physics of dressed 3d avatars from visual observations, 2024. 2

    2024

  70. [78]

    Pamir: Parametric model-conditioned implicit representa- tion for image-based human reconstruction

    Zerong Zheng, Tao Yu, Yebin Liu, and Qionghai Dai. Pamir: Parametric model-conditioned implicit representa- tion for image-based human reconstruction. IEEE Transac- tions on Pattern Analysis and Machine Intelligence (TPAMI),

  71. [79]

    Registering explicit to implicit: Towards high-fidelity gar- ment mesh reconstruction from single images

    Heming Zhu, Lingteng Qiu, Yuda Qiu, and Xiaoguang Han. Registering explicit to implicit: Towards high-fidelity gar- ment mesh reconstruction from single images. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 3 12 ReMu: Reconstr...

    2022

  72. [80]

    Body Fitting and Garment Separation In the garment reconstruction stage, we first fit a single- layer watertight mesh to each image using SITH [29]

    Implementation Details 6.1. Body Fitting and Garment Separation In the garment reconstruction stage, we first fit a single- layer watertight mesh to each image using SITH [29]. SITH starts by fitting an SMPLX [49] body to the image and then hallucinates the back view of the pe...

  73. [81]

    More Experimental Results 7.1. Ablation of Positional Encodings for Neural UDF In this ablation study, we showcase the results for various complex garments with high-frequency features, where the neural implicit UDFs are trained with different numbers of positional encodings. ...

Pith tools

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