pith. machine review for the scientific record. sign in

arxiv: 2605.04728 · v1 · submitted 2026-05-06 · 💻 cs.CV

Recognition: unknown

Anny-Fit: All-Age Human Mesh Recovery

Laura Bravo-S\'anchez , Matthieu Armando , Romain Br\'egier , Gr\'egory Rogez , Serena Yeung-Levy , Fabien Baradel
Authors on Pith no claims yet

Pith reviewed 2026-05-08 17:26 UTC · model grok-4.3

classification 💻 cs.CV
keywords all-age human mesh recovery3D pose and shape estimationmulti-person scenesjoint camera-space optimizationVLM semantic attributesdepth-scale ambiguitypseudo-ground-truth distillationzero-shot age adaptation
0
0 comments X

The pith

A joint camera-space optimization recovers accurate 3D human meshes for people of all ages by combining depth maps, outlines, keypoints, and age estimates.

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

Most existing methods recover 3D human meshes by treating each person independently and assuming adult body proportions. This breaks down in everyday scenes that mix adults and children, because depth and scale become ambiguous when body sizes differ. Anny-Fit instead optimizes every person together inside the camera coordinate system. It feeds the optimizer four complementary signals from separate networks: metric depth maps, instance segmentation, 2D keypoints, and age-gender labels produced by a vision-language model. The joint process yields meshes that fit the image better, respect depth order, and match true 3D shape more closely than independent adult-only fitting.

Core claim

Anny-Fit is a multi-person camera-space optimization framework that fuses metric depth maps, instance segmentation, 2D keypoints, and VLM-derived age and gender attributes to jointly recover 3D human meshes across the full age spectrum. These signals together resolve the depth-scale ambiguity that arises when body proportions vary with age. The resulting meshes show higher 2D reprojection accuracy, better relative depth ordering, lower 3D error, and improved shape fidelity, while also supplying pseudo-ground-truth labels that let adult-trained HMR models learn semantically meaningful shape parameters without retraining.

What carries the argument

Joint camera-space optimization that integrates metric depth, instance segmentation, 2D keypoints, and age-gender attributes from off-the-shelf networks to constrain all-age multi-person scenes.

Load-bearing premise

The separate networks that supply depth maps, segmentations, keypoints, and age-gender labels must deliver signals accurate and complementary enough to resolve scale without introducing dominant new errors or biases.

What would settle it

Applying the method to a dataset of mixed-age group images with known 3D ground truth and observing that 3D joint errors or depth-ordering mistakes exceed those from standard per-person adult HMR methods would falsify the benefit of the joint optimization.

Figures

Figures reproduced from arXiv: 2605.04728 by Fabien Baradel, Gr\'egory Rogez, Laura Bravo-S\'anchez, Matthieu Armando, Romain Br\'egier, Serena Yeung-Levy.

Figure 1
Figure 1. Figure 1: Anny-Fit recovers multi-person 3D human meshes of all ages directly in camera space. By integrating expert semantic, depth, keypoint, and segmentation cues, it improves all-age HMR and enables zero-shot adaptation of adult-only models. Abstract Recovering 3D human pose and shape from a single im￾age remains a cornerstone of human-centric vision, yet most methods assume adult subjects and optimize each pers… view at source ↗
Figure 2
Figure 2. Figure 2: Depth-scale ambiguity. Unlike adult-only settings where body size reliably indicates depth, the all-age setting gen￾eralizes the problem such that size alone cannot distinguish depth: identical 2D reprojections can correspond to either distant adults or nearby children. We leverage visual cues to infer shape and re-constrain the problem. recent Anny [7] model, which offers two key advantages for our purpos… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of Anny-Fit. We refine initial human meshes estimated using an HMR network through iterative optimization. Anny-Fit leverages pre-computed cues from expert vision models to guide fitting. To mitigate degenerate depth solutions that satisfy 2D reprojection losses, we incorporate shape attribute estimation and an explicit multi-person depth loss. 3. Method 3.1. Problem formulation Our goal is to rec… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results, in front and top view. Our method, Anny-Fit, exploits the advantages of SOTA models and generalizes them to the all-age setting. Compared to BEV, this translates to improvements in detection, depth ordering, shape and pose estimation. these challenges, our key insight is to reformulate shape cue estimation as a categorization task over each dimension of β (which has been shown to be mo… view at source ↗
Figure 6
Figure 6. Figure 6: Effect of shape. Top: Incorrect shape initialization pre￾vents the optimization from converging. Bottom: Accurate shape initialization resolves depth ordering. highlights the flexibility of Anny-Fit, which directly bene￾fits from advances in HMR methods even when they are not designed for all-age reconstruction. 4.1. Ablation experiments We study each component of our method on the Relative Human validatio… view at source ↗
Figure 5
Figure 5. Figure 5: Effect of depth loss. Adding a depth-based loss from an expert model preserves the relative depth relationships between people. All results use the same initialization and shape prediction. Circles denote incorrect (red) and correct (green) placement. non-specialized, adult-only model (CameraHMR) can be adapted to the all-age setting in a training-free manner, achieving competitive performance with special… view at source ↗
Figure 8
Figure 8. Figure 8 view at source ↗
Figure 9
Figure 9. Figure 9: Shape parameter anchors. Examples of the text descriptors and Anny shape parameter values for each anchor. Left: Age mapping. Right: Gender mapping. Other shape parameters are set to 0.5 here for visualization. (Age set to 0.66 for gender). or significant occlusions. We hypothesize that combining both the full body crop and the head crop would likely over￾come the limitations of each approach, but at an in… view at source ↗
Figure 10
Figure 10. Figure 10: Age distribution across different subsets of the Relative view at source ↗
Figure 11
Figure 11. Figure 11: Results on CMU toddler. For each method we show the camera and top views view at source ↗
Figure 12
Figure 12. Figure 12: Visualization of 2D Keypoint reprojection error percentiles. We filter out fits that are not within the 3rd and 95th percentiles. We overlay the ground-truth bounding boxes and estimated age and gender view at source ↗
Figure 13
Figure 13. Figure 13: Gender confusion matrix on Relative Human test. Note how after retraining with our Anny-Fit fits, the model can accurately predict gender in an unseen dataset view at source ↗
Figure 15
Figure 15. Figure 15: Additional examples of reconstructions with Anny-Fit (Ours) compared to SOTA on the Relative Human dataset. view at source ↗
read the original abstract

Recovering 3D human pose and shape from a single image remains a cornerstone of human-centric vision, yet most methods assume adult subjects and optimize each person independently. These assumptions fail in real-world, all-age scenes, where body proportions and depth must be resolved jointly. We introduce Anny-Fit, a multi-person, camera-space optimization framework for all-age 3D human mesh recovery (HMR). Unlike existing per-person fitting methods, Anny-Fit jointly optimizes all individuals directly in the camera coordinate system, enforcing global spatial consistency. At the core of our approach is the use of multiple forms of expert knowledge -- including metric depth maps, instance segmentation, 2D keypoints, and, VLM-derived semantic attributes such as age and gender -- each obtained from dedicated off-the-shelf networks. These complementary signals jointly guide the optimization, constraining the depth-scale ambiguity characteristic of all-age scenes. Across diverse datasets, Anny-Fit consistently improves 2D reprojection accuracy (+13 to 16), relative depth ordering (+6 to 7), 3D estimation error (-9 to -29) and shape estimation (+25 to +82), producing more coherent scenes. Finally, we show that VLM-based semantic knowledge can be distilled into an HMR model via the pseudo-ground-truth annotations produced by Anny-Fit on training data, enabling it to learn semantically meaningful shape parameters while improving HMR performance. Our approach bridges adult-only and all-age modeling by enabling zero-shot adaptation of adult-trained HMR pipelines to the full age spectrum without retraining. Code is publicly available at https://github.com/naver/anny-fit.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces Anny-Fit, a multi-person camera-space optimization framework for all-age 3D human mesh recovery from single images. It jointly optimizes all persons using complementary signals from off-the-shelf networks (metric depth maps, instance segmentation, 2D keypoints, and VLM-derived age/gender attributes) to enforce global spatial consistency and resolve depth-scale ambiguity. The paper reports consistent quantitative improvements in 2D reprojection accuracy (+13 to 16), relative depth ordering (+6 to 7), 3D estimation error (-9 to -29), and shape estimation (+25 to +82), along with a distillation step that uses the optimized pseudo-ground-truth to inject semantic knowledge into standard HMR models for zero-shot all-age adaptation without retraining. Code is released publicly.

Significance. If the results hold under rigorous validation, the work is significant for bridging adult-centric HMR methods to realistic all-age multi-person scenes. The public code release supports reproducibility, and the distillation mechanism offers a practical route to incorporate VLM semantic priors into parametric body models.

major comments (2)
  1. [Methods (joint optimization)] The central optimization (described in the methods) assumes that off-the-shelf metric depth maps and VLM age/gender signals are sufficiently accurate and complementary to jointly constrain all persons without introducing dominant per-instance scale drift or age-specific biases. No quantitative error analysis or validation of these input signals on all-age data (especially ages under 10) is provided, which directly underpins the claimed gains in relative depth ordering and 3D error.
  2. [Experiments and results] The reported quantitative gains (e.g., +13 to 16 in 2D reprojection, -9 to -29 in 3D error) lack accompanying details on experimental protocol, baseline implementations, error bars, dataset splits, and ablations isolating each signal's contribution. This makes it impossible to verify whether the improvements are robust or attributable to the proposed joint formulation.
minor comments (2)
  1. [Abstract] The abstract refers to 'VLM-derived semantic attributes' without naming the specific VLM or extraction procedure; this detail should be added for clarity.
  2. [Methods] Notation for optimization variables and loss terms could be introduced with a summary table or equation list in the methods for easier reference.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive feedback. We address each major comment point by point below, providing clarifications from the manuscript and outlining revisions where needed to strengthen the presentation.

read point-by-point responses

  1. Referee: [Methods (joint optimization)] The central optimization (described in the methods) assumes that off-the-shelf metric depth maps and VLM age/gender signals are sufficiently accurate and complementary to jointly constrain all persons without introducing dominant per-instance scale drift or age-specific biases. No quantitative error analysis or validation of these input signals on all-age data (especially ages under 10) is provided, which directly underpins the claimed gains in relative depth ordering and 3D error.

    Authors: We agree that an explicit quantitative validation of the off-the-shelf signals on all-age data would strengthen the methodological justification. The manuscript relies on the complementarity of the signals (metric depth, segmentation, keypoints, and VLM attributes) to mitigate individual inaccuracies, as evidenced by the consistent improvements in relative depth ordering and 3D error across datasets. However, we did not include a dedicated error analysis of the input networks on young children. In the revised version, we will add a new subsection with quantitative evaluation of each signal's accuracy on all-age benchmarks (including ages under 10), along with an analysis of how the joint optimization reduces per-instance drift. revision: yes

  2. Referee: [Experiments and results] The reported quantitative gains (e.g., +13 to 16 in 2D reprojection, -9 to -29 in 3D error) lack accompanying details on experimental protocol, baseline implementations, error bars, dataset splits, and ablations isolating each signal's contribution. This makes it impossible to verify whether the improvements are robust or attributable to the proposed joint formulation.

    Authors: The experimental protocol, baseline implementations (including per-person fitting methods), dataset splits, and evaluation metrics are detailed in Section 4 and the supplementary material. That said, we acknowledge the referee's point that error bars, explicit ablations for each signal, and more granular protocol descriptions would improve verifiability. We will revise the experiments section to include standard error bars over multiple runs, a full ablation table isolating the contribution of depth, segmentation, keypoints, and VLM signals, and expanded descriptions of the baselines and splits. revision: yes

Circularity Check

0 steps flagged

No significant circularity: optimization relies on independent off-the-shelf signals and distillation uses separate evaluation

full rationale

The paper's core derivation uses metric depth, segmentation, 2D keypoints and VLM age/gender attributes from dedicated off-the-shelf networks as external inputs to a joint camera-space optimization. These signals are not derived from the fitted meshes themselves. The subsequent distillation step generates pseudo-GT meshes from Anny-Fit outputs to supervise an HMR model, but the reported gains (+13-16 reprojection, -9 to -29 3D error, etc.) are measured on held-out test data against external ground truth rather than reducing to quantities defined inside the same fitted equations. No load-bearing self-citations, self-definitional loops, or ansatz smuggling appear in the derivation chain; the framework remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 0 axioms · 0 invented entities

The central claim rests on the reliability of four external off-the-shelf networks whose outputs are treated as fixed inputs to the joint optimization; no explicit free parameters or invented entities are named in the abstract, but the optimization itself necessarily involves weighting terms among the signals.

free parameters (1)
  • signal weighting coefficients
    The joint optimization must combine depth, segmentation, keypoint, and VLM signals; the abstract does not specify whether these weights are learned, hand-tuned, or fixed.

pith-pipeline@v0.9.0 · 5620 in / 1429 out tokens · 25866 ms · 2026-05-08T17:26:34.281752+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

71 extracted references · 11 canonical work pages · 4 internal anchors

  1. [1]

    Qwen2.5-VL Technical Report

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, et al. Qwen2.5-vl technical report.arXiv preprint arXiv:2502.13923, 2025. 3, 4, 7

    work page internal anchor Pith review arXiv 2025

  2. [2]

    Multi-hmr: Multi-person whole-body human mesh recovery in a single shot

    Fabien Baradel, Matthieu Armando, Salma Galaaoui, Ro- main Br ´egier, Philippe Weinzaepfel, Gr ´egory Rogez, and Thomas Lucas. Multi-hmr: Multi-person whole-body human mesh recovery in a single shot. InEuropean Conference on Computer Vision, pages 202–218. Springer, 2024. 2, 3, 6

    2024

  3. [3]

    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. InProceedings of the Computer Vi- sion and Pattern Recognition Conference, pages 2924–2934,

  4. [4]

    Keep it smpl: Automatic estimation of 3d human pose and shape from a single image

    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. InEuropean conference on computer vision, pages 561–578. Springer, 2016. 2

    2016

  5. [5]

    Ask, pose, unite: Scaling data acquisi- tion for close interaction meshes with vision language mod- els

    Laura Bravo-S ´anchez, Jaewoo Heo, Zhenzhen Weng, and Kuan-Chieh Wang. Ask, pose, unite: Scaling data acquisi- tion for close interaction meshes with vision language mod- els. InSynthetic Data for Computer Vision Workshop@ CVPR 2025, 2025. 2

    2025

  6. [6]

    Condimen: Conditional multi-person mesh recovery

    Romain Br ´egier, Fabien Baradel, Thomas Lucas, Salma Galaaoui, Matthieu Armando, Philippe Weinzaepfel, and Gr´egory Rogez. Condimen: Conditional multi-person mesh recovery. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 3880–3890, 2025. 2, 3, 8

    2025

  7. [7]

    Human mesh modeling for anny body.arXiv preprint arXiv:2511.03589, 2025

    Romain Br ´egier, Gu ´enol´e Fiche, Laura Bravo-S ´anchez, Thomas Lucas, Matthieu Armando, Philippe Weinzaepfel, Gr´egory Rogez, and Fabien Baradel. Human mesh modeling for anny body.arXiv preprint arXiv:2511.03589, 2025. 2, 3, 4, 6, 7, 1

    work page arXiv 2025

  8. [8]

    Meyer, Yuning Chai, Dennis Park, and Yong Jae Lee

    Mu Cai, Haotian Liu, Siva Karthik Mustikovela, Gregory P. Meyer, Yuning Chai, Dennis Park, and Yong Jae Lee. Mak- ing large multimodal models understand arbitrary visual prompts. InIEEE Conference on Computer Vision and Pat- tern Recognition, 2024. 7

    2024

  9. [9]

    Accurate 3d body shape regression using metric and semantic attributes

    Vasileios Choutas, Lea M ¨uller, Chun-Hao P Huang, Siyu Tang, Dimitrios Tzionas, and Michael J Black. Accurate 3d body shape regression using metric and semantic attributes. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2718–2728, 2022. 2, 3

    2022

  10. [10]

    Adversarial parametric pose prior

    Andrey Davydov, Anastasia Remizova, Victor Constantin, Sina Honari, Mathieu Salzmann, and Pascal Fua. Adversarial parametric pose prior. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 10997–11005, 2022. 3

    2022

  11. [11]

    PoseScript: Linking 3D Human Poses and Natural Lan- guage.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    Delmas, Ginger and Weinzaepfel, Philippe and Lucas, Thomas and Moreno-Noguer, Francesc and Rogez, Gr´egory. PoseScript: Linking 3D Human Poses and Natural Lan- guage.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024. 3

    2024

  12. [12]

    PoseFix: Correcting 3D Human Poses with Natural Language

    Delmas, Ginger and Weinzaepfel, Philippe and Moreno- Noguer, Francesc and Rogez, Gr´egory. PoseFix: Correcting 3D Human Poses with Natural Language. InProceedings of the IEEE/CVF international conference on computer vision (ICCV), 2023

    2023

  13. [13]

    PoseEmbroider: Towards a 3D, Visual, Semantic-aware Human Pose Rep- resentation

    Delmas, Ginger and Weinzaepfel, Philippe and Moreno- Noguer, Francesc and Rogez, Gr ´egory. PoseEmbroider: Towards a 3D, Visual, Semantic-aware Human Pose Rep- resentation. InEuropean Conference on Computer Vision (ECCV), 2024. 3

    2024

  14. [14]

    Teach clip to develop a number sense for ordinal regression

    Yao Du, Qiang Zhai, Weihang Dai, and Xiaomeng Li. Teach clip to develop a number sense for ordinal regression. InEuropean Conference on Computer Vision, pages 1–17. Springer, 2024. 5

    2024

  15. [15]

    Chatpose: Chatting about 3d human pose

    Yao Feng, Jing Lin, Sai Kumar Dwivedi, Yu Sun, Priyanka Patel, and Michael J Black. Chatpose: Chatting about 3d human pose. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2093–2103,

    2093

  16. [16]

    Three- dimensional reconstruction of human interactions

    Mihai Fieraru, Mihai Zanfir, Elisabeta Oneata, Alin-Ionut Popa, Vlad Olaru, and Cristian Sminchisescu. Three- dimensional reconstruction of human interactions. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7214–7223, 2020. 3

    2020

  17. [17]

    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. 2

    2023

  18. [18]

    Computer vision for medical infant motion analysis: State of the art and rgb-d data set

    Nikolas Hesse, Christoph Bodensteiner, Michael Arens, Ulrich G Hofmann, Raphael Weinberger, and A Sebas- tian Schroeder. Computer vision for medical infant motion analysis: State of the art and rgb-d data set. InProceed- ings of the European conference on computer vision (ECCV) workshops, pages 0–0, 2018. 3

    2018

  19. [19]

    Learning an infant body model from rgb-d data for accurate full body motion analysis

    Nikolas Hesse, Sergi Pujades, Javier Romero, Michael J Black, Christoph Bodensteiner, Michael Arens, Ulrich G Hofmann, Uta Tacke, Mijna Hadders-Algra, Raphael Wein- berger, et al. Learning an infant body model from rgb-d data for accurate full body motion analysis. InInternational Conference on Medical Image Computing and Computer- Assisted Intervention, ...

    2018

  20. [20]

    Closely interactive human reconstruction with proxemics and physics-guided adaption

    Buzhen Huang, Chen Li, Chongyang Xu, Liang Pan, Yan- gang Wang, and Gim Hee Lee. Closely interactive human reconstruction with proxemics and physics-guided adaption. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1011–1021, 2024. 3

    2024

  21. [21]

    Panoptic studio: A massively multiview sys- tem for social interaction capture.IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 2017

    Hanbyul Joo, Tomas Simon, Xulong Li, Hao Liu, Lei Tan, Lin Gui, Sean Banerjee, Timothy Scott Godisart, Bart Nabbe, Iain Matthews, Takeo Kanade, Shohei Nobuhara, and Yaser Sheikh. Panoptic studio: A massively multiview sys- tem for social interaction capture.IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 2017. 6, 3

    2017

  22. [22]

    End-to-end recovery of human shape and pose

    Angjoo Kanazawa, Michael J Black, David W Jacobs, and Jitendra Malik. End-to-end recovery of human shape and pose. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 7122–7131, 2018. 3 9

    2018

  23. [23]

    Harmony4d: A video dataset for in- the-wild close human interactions.Advances in Neural In- formation Processing Systems, 37:107270–107285, 2024

    Rawal Khirodkar, Jyun-Ting Song, Jinkun Cao, Zhengyi Luo, and Kris Kitani. Harmony4d: A video dataset for in- the-wild close human interactions.Advances in Neural In- formation Processing Systems, 37:107270–107285, 2024. 3

    2024

  24. [24]

    Learning to reconstruct 3d human pose and shape via model-fitting in the loop

    Nikos Kolotouros, Georgios Pavlakos, Michael J Black, and Kostas Daniilidis. Learning to reconstruct 3d human pose and shape via model-fitting in the loop. InProceedings of the IEEE/CVF international conference on computer vision, pages 2252–2261, 2019. 2, 3

    2019

  25. [25]

    Cliff: Carrying location information in full frames into human pose and shape estimation

    Zhihao Li, Jianzhuang Liu, Zhensong Zhang, Songcen Xu, and Youliang Yan. Cliff: Carrying location information in full frames into human pose and shape estimation. InEuro- pean Conference in Computer Vision, 2022. 3

    2022

  26. [26]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European conference on computer vision, pages 740–755. Springer, 2014. 8

    2014

  27. [27]

    Improved baselines with visual instruction tuning, 2023

    Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. Improved baselines with visual instruction tuning, 2023. 3

    2023

  28. [28]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. InNeurIPS, 2023. 3

    2023

  29. [29]

    Smpl: A skinned multi- person linear model

    Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J Black. Smpl: A skinned multi- person linear model. InSeminal Graphics Papers: Pushing the Boundaries, Volume 2, pages 851–866. 2023. 3, 6

    2023

  30. [30]

    Dposer-x: Diffusion model as robust 3d whole-body human pose prior.arXiv preprint arXiv:2508.00599, 2025

    Junzhe Lu, Jing Lin, Hongkun Dou, Ailing Zeng, Yue Deng, Xian Liu, Zhongang Cai, Lei Yang, Yulun Zhang, Hao- qian Wang, and Ziwei Liu. Dposer-x: Diffusion model as robust 3d whole-body human pose prior.arXiv preprint arXiv:2508.00599, 2025. 3

    work page arXiv 2025

  31. [31]

    SmolVLM: Redefining small and efficient multimodal models

    Andr ´es Marafioti, Orr Zohar, Miquel Farr ´e, Merve Noyan, Elie Bakouch, Pedro Cuenca, Cyril Zakka, Loubna Ben Allal, Anton Lozhkov, Nouamane Tazi, et al. Smolvlm: Redefining small and efficient multimodal models.arXiv preprint arXiv:2504.05299, 2025. 7

    work page internal anchor Pith review arXiv 2025

  32. [32]

    V olumetricsmpl: A neural volumet- ric body model for efficient interactions, contacts, and colli- sions

    Marko Mihajlovic, Siwei Zhang, Gen Li, Kaifeng Zhao, Lea Muller, and Siyu Tang. V olumetricsmpl: A neural volumet- ric body model for efficient interactions, contacts, and colli- sions. InProceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 5060–5070, 2025. 8

    2025

  33. [33]

    On self-contact and human pose

    Lea Muller, Ahmed AA Osman, Siyu Tang, Chun-Hao P Huang, and Michael J Black. On self-contact and human pose. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 9990–9999,

  34. [34]

    Generative proxemics: A prior for 3d social interaction from images

    Lea Muller, Vickie Ye, Georgios Pavlakos, Michael Black, and Angjoo Kanazawa. Generative proxemics: A prior for 3d social interaction from images. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9687–9697, 2024. 2

    2024

  35. [35]

    Black, and Angjoo Kanazawa

    Lea Muller, Vickie Ye, Georgios Pavlakos, Michael J. Black, and Angjoo Kanazawa. Generative proxemics: A prior for 3D social interaction from images. 2024. 3, 6, 8

    2024

  36. [36]

    Camerahmr: Aligning people with perspective

    Priyanka Patel and Michael J Black. Camerahmr: Aligning people with perspective. In2025 International Conference on 3D Vision (3DV), pages 1562–1571. IEEE, 2025. 2, 3, 5, 6, 7, 8

    2025

  37. [37]

    Agora: Avatars in geography optimized for regression analysis

    Priyanka Patel, Chun-Hao P Huang, Joachim Tesch, David T Hoffmann, Shashank Tripathi, and Michael J Black. Agora: Avatars in geography optimized for regression analysis. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 13468–13478, 2021. 2, 3

    2021

  38. [38]

    Expressive body capture: 3d hands, face, and body from a single image

    Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed AA Osman, Dimitrios Tzionas, and Michael J Black. Expressive body capture: 3d hands, face, and body from a single image. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10975–10985, 2019. 3

    2019

  39. [39]

    Pexels stock photos.https://www.pexels

    Pexels. Pexels stock photos.https://www.pexels. com/, 2025. 3

    2025

  40. [40]

    Unidepthv2: Universal monocular metric depth estimation made simpler

    Luigi Piccinelli, Christos Sakaridis, Yung-Hsu Yang, Mat- tia Segu, Siyuan Li, Wim Abbeloos, and Luc Van Gool. UniDepthV2: Universal monocular metric depth estimation made simpler.arXiv preprint arXiv:2502.20110, 2025. 6

    work page arXiv 2025

  41. [41]

    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. arXiv preprint arXiv:2401.14159, 2024. 6

    work page internal anchor Pith review arXiv 2024

  42. [42]

    Dex: Deep expectation of apparent age from a single image

    Rasmus Rothe, Radu Timofte, and Luc Van Gool. Dex: Deep expectation of apparent age from a single image. InProceed- ings of the IEEE international conference on computer vision workshops, pages 10–15, 2015. 3

    2015

  43. [43]

    Neural localizer fields for continuous 3d human pose and shape estimation

    Istv ´an S ´ar´andi and Gerard Pons-Moll. Neural localizer fields for continuous 3d human pose and shape estimation. Advances in Neural Information Processing Systems, 37: 140032–140065, 2024. 1

    2024

  44. [44]

    Syn- thetic training for accurate 3d human pose and shape esti- mation in the wild

    Akash Sengupta, Ignas Budvytis, and Roberto Cipolla. Syn- thetic training for accurate 3d human pose and shape esti- mation in the wild. InBritish Machine Vision Conference (BMVC), 2020. 3

    2020

  45. [45]

    Deep regression forests for age estimation

    Wei Shen, Yilu Guo, Yan Wang, Kai Zhao, Bo Wang, and Alan L Yuille. Deep regression forests for age estimation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2304–2313, 2018. 3

    2018

  46. [46]

    Body talk: Crowdshaping realistic 3d avatars with words.ACM Transactions on Graphics (TOG), 35(4):1–14, 2016

    Stephan Streuber, M Alejandra Quiros-Ramirez, Matthew Q Hill, Carina A Hahn, Silvia Zuffi, Alice O’Toole, and Michael J Black. Body talk: Crowdshaping realistic 3d avatars with words.ACM Transactions on Graphics (TOG), 35(4):1–14, 2016. 3

    2016

  47. [47]

    Sat- hmr: Real-time multi-person 3d mesh estimation via scale- adaptive tokens

    Chi Su, Xiaoxuan Ma, Jiajun Su, and Yizhou Wang. Sat- hmr: Real-time multi-person 3d mesh estimation via scale- adaptive tokens. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, 2025. 3, 6

    2025

  48. [48]

    Aios: All-in-one-stage expressive human pose and shape estimation

    Qingping Sun, Yanjun Wang, Ailing Zeng, Wanqi Yin, Chen Wei, Wenjia Wang, Haiyi Mei, Chi-Sing Leung, Ziwei Liu, Lei Yang, et al. Aios: All-in-one-stage expressive human pose and shape estimation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,

  49. [49]

    Monocular, One-stage, Regression of Multiple 3D People

    Yu Sun, Qian Bao, Wu Liu, Yili Fu, Black Michael J., and 10 Tao Mei. Monocular, One-stage, Regression of Multiple 3D People. InICCV, 2021

    2021

  50. [50]

    Putting people in their place: Monocular regression of 3d people in depth

    Yu Sun, Wu Liu, Qian Bao, Yili Fu, Tao Mei, and Michael J Black. Putting people in their place: Monocular regression of 3d people in depth. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 13243–13252, 2022. 2, 3, 6, 7, 1

    2022

  51. [51]

    Deco: Dense estimation of 3d human-scene contact in the wild

    Shashank Tripathi, Agniv Chatterjee, Jean-Claude Passy, Hongwei Yi, Dimitrios Tzionas, and Michael J Black. Deco: Dense estimation of 3d human-scene contact in the wild. In Proceedings of the IEEE/CVF international conference on computer vision, pages 8001–8013, 2023. 2

    2023

  52. [52]

    Body size and depth disambiguation in multi-person reconstruction from single images

    Nicolas Ugrinovic, Adria Ruiz, Antonio Agudo, Alberto Sanfeliu, and Francesc Moreno-Noguer. Body size and depth disambiguation in multi-person reconstruction from single images. In2021 International Conference on 3D Vision (3DV), pages 53–63. IEEE, 2021. 2, 3, 7

    2021

  53. [53]

    arXiv preprint arXiv:2511.13282 (2025) 6, 14, 26

    Kaiwen Wang, Kaili Zheng, Yiming Shi, Chenyi Guo, and Ji Wu. Towards metric-aware multi-person mesh recovery by jointly optimizing human crowd in camera space.arXiv preprint arXiv:2511.13282, 2025. 3

    work page arXiv 2025

  54. [54]

    Refit: Recurrent fit- ting network for 3d human recovery

    Yufu Wang and Kostas Daniilidis. Refit: Recurrent fit- ting network for 3d human recovery. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 14644–14654, 2023. 2

    2023

  55. [55]

    Prompthmr: Promptable human mesh recovery

    Yufu Wang, Yu Sun, Priyanka Patel, Kostas Daniilidis, Michael J Black, and Muhammed Kocabas. Prompthmr: Promptable human mesh recovery. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 1148–1159, 2025. 2, 3

    2025

  56. [56]

    Humphreys, Lee M

    Zhenzhen Weng, Laura Bravo-S ´anchez, Zeyu Wang, Christopher Howard, Maria Xenochristou, Nicole Meis- ter, Angjoo Kanazawa, Arnold Milstein, Elika Bergelson, Kathryn L. Humphreys, Lee M. Sanders, and Serena Yeung- Levy. Artificial intelligence–powered 3d analysis of video- based caregiver-child interactions.Science Advances, 11(8): eadp4422, 2025. 3

    2025

  57. [57]

    Reconstructing humans with a biome- chanically accurate skeleton

    Yan Xia, Xiaowei Zhou, Etienne V ouga, Qixing Huang, and Georgios Pavlakos. Reconstructing humans with a biome- chanically accurate skeleton. InProceedings of the Com- puter Vision and Pattern Recognition Conference, pages 5355–5365, 2025. 2

    2025

  58. [58]

    Ghum & ghuml: Generative 3d human shape and articulated pose models

    Hongyi Xu, Eduard Gabriel Bazavan, Andrei Zanfir, William T Freeman, Rahul Sukthankar, and Cristian Smin- chisescu. Ghum & ghuml: Generative 3d human shape and articulated pose models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6184–6193, 2020. 3

    2020

  59. [59]

    Vitpose+: Vision transformer foundation model for generic body pose estimation.arXiv preprint arXiv:2212.04246,

    Yufei Xu, Jing Zhang, Qiming Zhang, and Dacheng Tao. Vitpose+: Vision transformer foundation model for generic body pose estimation.arXiv preprint arXiv:2212.04246,

    work page arXiv

  60. [60]

    Hi4d: 4d instance segmentation of close human interaction

    Yifei Yin, Chen Guo, Manuel Kaufmann, Juan Zarate, Jie Song, and Otmar Hilliges. Hi4d: 4d instance segmentation of close human interaction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,

  61. [61]

    Monocular 3d pose and shape estimation of mul- tiple people in natural scenes-the importance of multiple scene constraints

    Andrei Zanfir, Elisabeta Marinoiu, and Cristian Sminchis- escu. Monocular 3d pose and shape estimation of mul- tiple people in natural scenes-the importance of multiple scene constraints. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 2148–2157,

  62. [62]

    Pymaf-x: To- wards well-aligned full-body model regression from monoc- ular images.IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 45(10):12287–12303, 2023

    Hongwen Zhang, Yating Tian, Yuxiang Zhang, Mengcheng Li, Liang An, Zhenan Sun, and Yebin Liu. Pymaf-x: To- wards well-aligned full-body model regression from monoc- ular images.IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 45(10):12287–12303, 2023. 2

    2023

  63. [63]

    Metrichmr: Metric human mesh recovery from monocular images.arXiv preprint arXiv:2506.09919, 2025

    He Zhang, Chentao Song, Hongwen Zhang, and Tao Yu. Metrichmr: Metric human mesh recovery from monocular images.arXiv preprint arXiv:2506.09919, 2025. 2, 3

    work page arXiv 2025

  64. [64]

    A Survey of Large Language Models

    Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, et al. A survey of large language mod- els.arXiv preprint arXiv:2303.18223, 2023. 5

    work page internal anchor Pith review arXiv 2023

  65. [65]

    Single view metrology in the wild

    Rui Zhu, Xingyi Yang, Yannick Hold-Geoffroy, Federico Perazzi, Jonathan Eisenmann, Kalyan Sunkavalli, and Man- mohan Chandraker. Single view metrology in the wild. In European Conference on Computer Vision, pages 316–333. Springer, 2020. 2

    2020

  66. [66]

    Kbody: Towards general, robust, and aligned monocular whole-body estima- tion

    Nikolaos Zioulis and James F O’Brien. Kbody: Towards general, robust, and aligned monocular whole-body estima- tion. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 6215–6225,

  67. [67]

    2 11 Anny-Fit: All-Age Human Mesh Recovery Supplementary Material

  68. [68]

    Age estimation prompt

    Implementation details 7.1. Anny model mapping We utilize the semantic shape space of the Anny model [7] to propose a direct mapping from shape attribute descrip- tors to normalized shape values. This mapping inherently accounts for the body model interpolation described in the same work. Figure 9 illustrates our complete mapping scheme for all experiment...

  69. [69]

    punishment value

    Experiment details 8.1. Metrics Percentage of Correct Keypoints (PCK). To robustly ac- count for missing keypoint detections, prior work often as- signs a fixed “punishment value” to unmatched predictions when computing the MPJPE. However, such heuristics dis- tort the numerical scale of the evaluation and can introduce undesirable incentives—e.g., a miss...

    work page arXiv

  70. [70]

    As a case study, we explore diverse body shapes

    Beyond all-age shape estimation While not our primary objective, our shape estimation for- mulation generalizes to account for attributes beyond age and gender. As a case study, we explore diverse body shapes. To manage a diverse range of shapes with a com- pact categorization for VLM querying, we define a dis- crete mapping to weight and muscle attribute...

  71. [71]

    While this multi-expert strat- egy improves robustness, it also makes performance depen- dent on the accuracy of each expert

    Limitations Our method integrates multiple expert predictions to guide the optimization for all-age human reconstruction, which enhances overall robustness. While this multi-expert strat- egy improves robustness, it also makes performance depen- dent on the accuracy of each expert. Errors in keypoints or depth under occlusion or extreme viewpoints can pro...