Pith. sign in

REVIEW 3 major objections 5 minor 30 references

From Skin to Skeleton: Towards Biomechanically Accurate 3D Digital Humans

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read SKEL gives the standard SMPL human body model a biomechanical skeleton driven by the same shape and pose parameters as the skin surface.

desk verdict A genuinely useful skin-skeleton parametric model with an accuracy claim that is self-referential but explicitly acknowledged; deserves serious peer review and a conditional accept. read the letter →

arxiv 2509.06607 v1 pith:6P6SPRFP submitted 2025-09-08 cs.GR cs.CV

Marilyn Keller , Keenon Werling , Soyong Shin , Scott Delp , Sergi Pujades , C. Karen Liu , Michael J. Black This is my paper · ORCID
classification cs.GRcs.CV
keywords parametrichumanbodymodelbiomechanicalskeletonSMPLmotioncapturejointregressiondatasetgenerationdigitalanimationmusculoskeletal
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to close the gap between body models used in computer vision and the anatomically accurate skeletons used in biomechanics. It builds BioAMASS, a dataset pairing over nine hours of SMPL body motion with fitted anatomical skeletons, by painting virtual markers on the skin and solving for the skeleton underneath. From this pairing it learns SKEL, a parametric model in which one set of shape and pose parameters $(\boldsymbol{\beta}, q)$ drives both the body surface and the bones, with anatomically constrained spine, shoulder-blade, and forearm motion. If the paper is right, the many existing systems that already estimate SMPL bodies can be upgraded to also output biomechanical joint locations and skeleton poses, enabling biomechanics from ordinary video and archived motion data.

What carries the argument

The load-bearing mechanism is a three-stage pipeline. First, a custom 24-bone biomechanical skeleton with 46 degrees of freedom—featuring a constant-curvature spine, an ellipsoid-sliding scapula, and a radius–ulna forearm—is fit inside posed SMPL meshes using virtual markers and a recent bilevel biomechanical optimization, producing the paired BioAMASS dataset. Second, a non-negative least-squares regressor maps SMPL mesh vertices to the 24 anatomical joint locations. Third, each bone's orientation is decomposed into a fixed base rotation learned from the dataset and a shape-dependent correction $R_{\boldsymbol{\beta}}^i(\boldsymbol{\beta})$ that aligns the bone with its parent–child joint segment; these pieces re-rig the SMPL mesh so the biomechanical pose parameters move skin and skeleton in synchrony. The decomposition is what lets SKEL keep SMPL's shape space while giving the bones anatomically coherent orientations inside every body shape.

What would settle it

Take a subject with both a CT or MRI scan and synchronized marker motion capture, fit SKEL to the motion, and compare its femur-head and humerus-head joint centers with the imaging-derived centers; a systematic offset, especially at the shoulder, would falsify the claimed biomechanical accuracy.

Watch

Extended reading notes

Core claim

SKEL is presented as the first model in which the body surface and the anatomical skeleton are directly controlled by the same shape and pose parameters. Trained on the BioAMASS dataset, SKEL places a properly scaled skeleton inside any SMPL body; its regressed joint locations are closer to functional anatomical joints than SMPL's artist-defined joints, and its bones fit inside the body surface better than earlier skeleton-from-skin methods. The model uses fewer pose parameters than SMPL (46 versus 72) but models the spine as a constant-curvature chain, the scapula as sliding on an ellipsoid around the thorax, and forearm pronation and supination as a radius–ulna motion, so its articulation is biomechanically constrained rather than a collection of ball joints.

Load-bearing premise

The whole result rests on the assumption that fitting the skeleton to markers painted on the skin of synthetic bodies yields essentially correct joint locations, even though those fits are only pseudo-ground-truth and are never checked against medical images or bone-pin measurements.

Editorial extensions

If this is right

  • Existing SMPL-based vision pipelines can be post-processed by fitting SKEL, turning datasets such as 3DPW and BEDLAM into sources of biomechanical skeleton parameters.
  • Regressing anatomical joints from skin (or fitting SKEL) yields joint locations closer to the functional joints than SMPL's own joints, with errors below one centimeter for most joints in the reported evaluation.
  • Because SKEL has 46 biomechanical degrees of freedom instead of SMPL's 72 ball-joint degrees, posed bodies exhibit more realistic shoulder-blade sliding, spine bending, and forearm twist.
  • Given a biomechanical skeleton, SKEL can generate a plausible skin surface while preserving the skeleton's bone lengths, allowing motion-capture data to be visualized with body shapes of different weights.

Reading between the lines

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

  • The decisive test the paper does not run is against independent anatomy: registering SKEL's femur-head and humerus-head joints to CT or MRI-derived joint centers on the same subjects would show whether the pseudo-ground-truth fits are unbiased.
  • If SKEL is used to train video-based biomechanics estimators, the natural downstream check is whether joint angles and moments computed from SKEL agree with marker-based motion capture to within clinical tolerances; that is the metric that would justify “biomechanics in the wild.”
  • Since the learning and rigging pipeline is independent of the particular skeleton model, the same procedure could be rerun with a clinically validated skeleton the moment one exists, upgrading SKEL without architectural changes.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper introduces SKEL, a parametric body model that re-rigs the SMPL surface model with a biomechanical skeleton, BSM, so that skin and skeleton are controlled by the same pose vector q and shape vector β. To train SKEL, the authors construct BioAMASS, a dataset of about 9 hours of motion from 113 AMASS subjects, by placing virtual markers on SMPL meshes and fitting BSM with the AddBiomechanics optimization pipeline. They learn a non-negative joint regressor from SMPL vertices to BSM anatomical joint locations, estimate per-bone base rotations and shape-dependent corrective rotations, and define a new kinematic tree with 46 biomechanical degrees of freedom, including a constant-curvature spine, sliding scapulae, and radioulnar pronation/supination. The paper evaluates joint-location errors against the BSM fits used as pseudo-ground-truth, reports SKEL-to-SMPL surface fitting errors on DFAUST, and shows qualitative comparisons with OSSO on MOYO.

Significance. If the accuracy claims are taken at face value, SKEL is a valuable bridge between statistical body models and biomechanics: it is, to my knowledge, the first model in which skin and a biomechanical skeleton are driven by the same pose and shape parameters, and the BioAMASS dataset and the SMPL-to-BSM fitting pipeline are potentially useful resources. The re-rigging formulation is well motivated, and the paper includes several thoughtful design choices, such as using OSSO to personalize virtual marker offsets and modeling the shoulder blade with an ellipsoid constraint. The main limitation is that the central claim of biomechanical accuracy is evaluated against the same AddBiomechanics fitting pipeline that generated the training labels, and the paper's own Sec. 7 acknowledges this is pseudo-ground-truth. Consequently, the sub-centimeter reported errors demonstrate reproducibility of a fitting pipeline rather than independently established anatomical accuracy; the qualitative OSSO comparison also does not quantify the 'bones fit inside the body surface better' claim.

major comments (3)
  1. [Sec. 6.2, Sec. 4.3.2, Sec. 7] The evaluation of the joint regressor and the SKEL fits is circular. The joint locations J^B used as 'ground truth' in Fig. 10 and Fig. 11 are produced by the same AddBiomechanics optimization (Eq. (2)) that generated the BioAMASS training labels. The learned regressor and SKEL are therefore scored by their ability to reproduce the fitting pipeline, not by their anatomical correctness. The paper itself states in Sec. 7 that these fits 'should not be considered as actual ground truth, but rather a pseudo-ground truth.' To support the abstract's claim of 'more biomechanically accurate joint locations than SMPL,' the authors should add an external validation, for example comparing a subset of subjects against OSSO skeletons trained on medical images, functional joint-center estimates, or published anthropometric or imaging measurements; alternatively, the claims throughout the abstract, Sec. 1, and Sec. 6.2 should be softened to 'consistency with a marker-based biomechanical fitting pipeline.'
  2. [Sec. 6.4] The claim that the bones 'fit inside the body surface better than previous methods' is supported only by qualitative side-by-side images of OSSO and SKEL on MOYO. No quantitative metric is reported for bone containment, bone-skin penetration, or agreement of bone orientations with an independent reference. I recommend adding a scalar evaluation, such as the fraction of bone vertices inside the SMPL skin mesh, the signed distance of bone vertices to the skin surface, or a comparison of joint angles against the AddBiomechanics reference, on a held-out set of poses and subjects.
  3. [Sec. 4.3.2, Eq. (2)] The per-marker offsets δ and the marker weights λ_k are free parameters in the bi-level optimization that creates BioAMASS, and the paper does not state how δ is initialized or regularized. Unrestrained per-marker offsets can absorb systematic skin-to-bone offsets and shift the estimated joint centers, so the skeleton locations might be determined more by δ than by the marker trajectories. Please report the magnitude and variability of the learned offsets, and provide a sensitivity analysis showing that joint locations in BioAMASS are stable when δ is removed or differently regularized.
minor comments (5)
  1. [Sec. 4.2, Eq. (1)] The formula for t_spine is difficult to parse; the term '(r−cos(α)) ∗ −sin(q_z)' appears to have a missing operand or an unusual double sign, and the definitions of q_x, q_y, q_z are incomplete. Adding explicit parentheses and a short derivation would improve readability.
  2. [Sec. 5.2, Eq. (6)] The text refers to 'the green term' in the equation, but the printed equation is monochrome; please replace the color reference with an explicit description of the factors being discussed.
  3. [Sec. 5.2 and Sec. 6.3] Sec. 5.2 states that SKEL can match SMPL meshes 'with an average vertex-to-vertex error below 3 cm,' while Sec. 6.3 reports average mean differences of 1.1 cm for males and 0.9 cm for females on DFAUST. Please reconcile these numbers, and clarify whether the 3 cm figure refers to extreme poses outside DFAUST.
  4. [Sec. 4.3.2] The prior P(s,β) is said to use biological sex as in Werling et al., but the paper does not explain how biological sex is obtained from the SMPL shape parameters β. Please specify the sex estimation procedure or state that sex is assumed from the source dataset.
  5. [Fig. 12] The color bar is labeled 'Blue: 0 cm, Red: 2 cm,' yet the text reports a maximum female difference of 1.9 cm; please adjust the visualization range or the text so the figure and caption are consistent.

Circularity Check

1 steps flagged · score 5.0 of 10

The quantitative 'biomechanical accuracy' claim is benchmarked against the same AddBiomechanics pseudo-ground-truth that generated the training labels; no independent anatomical validation breaks the loop.

  1. fitted input called prediction [Sec. 4.3.2 (Eq. 2), Sec. 5.1, Sec. 6.2 and Figs. 10–11]
    "We train these regressors from the posed vertices and joints of the BioAMASS dataset. ... For each frame of the DFAUST dataset, BioAMASS provides the anatomical joint locations J^B that we consider ground truth. Then, from the frame's SMPL mesh, we use our learned joint regressor to regress the anatomical joint location J^reg."

    The training targets for the joint regressor are the BSM joint locations produced by solving Eq. (2) with AddBiomechanics. The evaluation in Sec. 6.2 uses the same kind of output ('J^B that we consider ground truth') as the reference for the regressed joints. The reported sub-centimeter errors therefore quantify how faithfully the learned regressor reproduces the AddBiomechanics/BSM fitting pipeline, not an independently established anatomical accuracy. The paper itself concedes in Sec.

full rationale

The construction of SKEL itself—learning anatomical joint regressors from BioAMASS, deriving bone orientations from Eq. (4), and re-rigging SMPL with BSM degrees of freedom—is a coherent, non-circular engineering pipeline. The regressor is trained on one subset of AMASS and evaluated on DFAUST, so the experiment does demonstrate generalization across subjects and poses relative to the chosen labels. The circularity is confined to the interpretation of those labels as 'ground truth' for biomechanical accuracy. The labels are outputs of AddBiomechanics fits to virtual markers placed on SMPL meshes; the same AddBiomechanics/BSM pipeline generated the training supervision. Thus the quantitative evaluation of 'more biomechanically accurate joint locations' (Figs. 10–11) measures consistency with the pseudo-ground-truth generator, not with an independent skeletal measurement such as medical imaging, bone-pin data, or functional joint-center tests. The paper explicitly acknowledges this limitation in Sec. 7, which is commendable but does not remove the self-referential nature of the benchmark. The qualitative OSSO comparison (Sec. 6.4) is not quantified and does not independently establish the 'bones fit inside the body surface better' claim. No uniqueness theorem, ansatz-smuggling via citation, or definitional equation identity was found in the rigging mathematics; the core issue is the closed evaluation loop around the pseudo-ground-truth.

Assumptions & free parameters 7 free parameters · 5 assumptions · 2 invented entities

The SKEL model rests on several fitted parameters (bone scales, marker offsets, regressor weights, base rotations) and on the assumption that the marker-based biomechanical optimization yields reliable pseudo-ground-truth. The custom spine joint is an invented modeling construct without independent evidence. The paper acknowledges the pseudo-ground-truth limitation but does not provide external validation, which is the main epistemic gap.

free parameters (7)
  • BSM bone scales s = Per subject, R^{24x3}, not reported in the paper
    Optimized in Eq. 2 via AddBiomechanics for each AMASS subject; these scales define limb lengths and are central to the skeleton fit.
  • Per-marker offsets δ = Per marker per subject, R^{N_m x 3}
    Optimized in Eq. 2 to allow soft tissue motion; these offsets are fit to the data and influence the resulting skeleton pose.
  • Marker weights λ_k = Low for soft markers, high for bony markers
    Hand-set weights in Eq. 2 that control the trade-off between bony and soft marker errors; chosen by the authors after experimentation.
  • Joint regressor coefficients = 24 x 6890 x 3 matrix, learned
    Learned via non-negative least squares on BioAMASS (Sec. 5.1); maps SMPL vertices to anatomical joint locations.
  • Base bone rotations R_base^i = Learned per bone from Eq. 4
    Minimized over all BioAMASS frames to orient each bone around its own axis; this is a fitted quantity that the final model depends on.
  • Spine arc length l = Chosen by design, not reported
    Parameter of the custom constant-curvature spine joint (Eq. 1); fixed value but no justification or sensitivity analysis is given.
  • Mapping from SKEL DOFs to SMPL pose correctives = Hand-picked correspondence
    The paper states each SKEL DOF is mapped to a corresponding SMPL DOF to transfer pose-dependent deformations; this mapping is chosen manually and affects mesh accuracy.
assumptions (5)
  • domain assumption AddBiomechanics optimization recovers sufficiently accurate skeletal joint locations from surface markers
    The entire BioAMASS dataset and all evaluations treat these fits as ground truth; the paper explicitly labels them pseudo-ground-truth in Sec. 7, so the accuracy of this assumption is load-bearing.
  • domain assumption The BSM kinematic structure (Rajagopal lower body, Seth scapula, custom constant-curvature spine) accurately represents human biomechanics
    Adopted from prior biomechanics models, but the custom spine joint is new and not validated against measured spine kinematics.
  • domain assumption Skin markers, including soft markers, plus a scale prior, constrain the underlying skeleton pose and bone lengths
    This is the standard assumption of marker-based motion capture; soft tissue motion is only downweighted via marker weights, not modeled explicitly.
  • ad hoc to paper Uniform body density can be assumed to estimate weight from SMPL shape
    Used in Sec. 4.3.2 to define the scale prior P(s,β); this is a coarse approximation that could bias bone scaling for different body compositions.
  • ad hoc to paper SMPL pose-dependent deformations and skinning weights transfer to the SKEL rig with acceptable error
    The paper states this transfer creates artifacts in extreme poses and leads to about 1 cm mesh error, so the assumption is approximate by the authors' own admission.
invented entities (2)
  • BSM (Biomechanical Skeleton Model)
    purpose: A 24-bone OpenSim model with 46 degrees of freedom, including custom spine and scapula joints, used as the skeleton to fit inside SMPL meshes
    New assembly of joints including a novel constant-curvature spine; no external validation of this specific model against independent skeletal measurements.
  • Constant-curvature spine joint
    purpose: Models lumbar, thoracic, and cervical bending with constant arc length using a single termination angle (Eq. 1)
    Introduced in Sec. 4.2 as a custom extension to OpenSim; no comparison to measured spine kinematics or clinical data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of From Skin to Skeleton: Towards Biomechanically Accurate 3D Digital Humans." pith.science (2026). https://pith.science/paper/6P6SPRFP

@misc{pith2026250906607,
  author       = {Pith},
  title        = {Pith review of: From Skin to Skeleton: Towards Biomechanically Accurate 3D Digital Humans},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6P6SPRFP}},
  note         = {Machine review of arXiv:2509.06607}
}
read the original abstract

Great progress has been made in estimating 3D human pose and shape from images and video by training neural networks to directly regress the parameters of parametric human models like SMPL. However, existing body models have simplified kinematic structures that do not correspond to the true joint locations and articulations in the human skeletal system, limiting their potential use in biomechanics. On the other hand, methods for estimating biomechanically accurate skeletal motion typically rely on complex motion capture systems and expensive optimization methods. What is needed is a parametric 3D human model with a biomechanically accurate skeletal structure that can be easily posed. To that end, we develop SKEL, which re-rigs the SMPL body model with a biomechanics skeleton. To enable this, we need training data of skeletons inside SMPL meshes in diverse poses. We build such a dataset by optimizing biomechanically accurate skeletons inside SMPL meshes from AMASS sequences. We then learn a regressor from SMPL mesh vertices to the optimized joint locations and bone rotations. Finally, we re-parametrize the SMPL mesh with the new kinematic parameters. The resulting SKEL model is animatable like SMPL but with fewer, and biomechanically-realistic, degrees of freedom. We show that SKEL has more biomechanically accurate joint locations than SMPL, and the bones fit inside the body surface better than previous methods. By fitting SKEL to SMPL meshes we are able to "upgrade" existing human pose and shape datasets to include biomechanical parameters. SKEL provides a new tool to enable biomechanics in the wild, while also providing vision and graphics researchers with a better constrained and more realistic model of human articulation. The model, code, and data are available for research at https://skel.is.tue.mpg.de..

Figures

Figures reproduced from arXiv: 2509.06607 by the authors.

Figure 1
Figure 1. (a) We fit our new Biomechanical Skeleton Model, BSM, to SMPL [Loper et al [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Creation of the paired skeleton and body dataset. Given a SMPL [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The markers defined on SMPL: bony in orange, soft in blue. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (9 more)
Figure 6
Figure 6. Figure 6: Left: SKEL kinematic tree with learned anatomical joint locations. Right: SMPL’s kinematic tree. Middle: the superposition of both. In con￾trast to SMPL, which has axis-aligned rotation axes, SKEL’s rotation axes are bone-aligned. similarly to Loper et al. [2015] by le…
Figure 5
Figure 5. Figure 5: BioAMASS: examples of BSM fits to AMASS poses. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 8
Figure 8. Figure 8: Left: Rigging the skeleton to the regressed joints and posing them [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Illustration of SKEL’s degrees of freedom. The bone and body surface [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 11
Figure 11. Figure 11: On DFAUST female subjects, we predict the joint locations and show [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]
Figure 10
Figure 10. Figure 10: Anatomical joints regression error over the female DFAUST dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]
Figure 12
Figure 12. Figure 12: Average per vertex distance between SKEL and SMPL fit to the [PITH_FULL_IMAGE:figures/full_fig_p009_12.png]
Figure 13
Figure 13. Figure 13: SKEL can be fit to existing SMPL datasets to upgrade them with [PITH_FULL_IMAGE:figures/full_fig_p010_13.png]
Figure 15
Figure 15. Figure 15: Given an input skeleton, and a target weight, SKEL can generate [PITH_FULL_IMAGE:figures/full_fig_p010_15.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 17 canonical work pages

  1. [14]

    Matthew M

    SMPL: A skinned multi-person linear model.ACM Transactions on Graphics34, 6 (2015), 1–16. Matthew M. Loper, Naureen Mahmood, and Michael J. Black

    work page 2015

  2. [16]

    Laurie Needham, Murray Evans, Darren P Cosker, Logan Wade, Polly M McGuigan, James L Bilzon, and Steffi L Colyer

    DeepLabCut: markerless pose estimation of user-defined body parts with deep learning.Nature neuroscience 21, 9 (2018), 1281–1289. Laurie Needham, Murray Evans, Darren P Cosker, Logan Wade, Polly M McGuigan, James L Bilzon, and Steffi L Colyer

    work page 2018

  3. [18]

    Scientific reports10, 1 (2020), 1–12

    Efficient trajectory optimiza- tion for curved running using a 3D musculoskeletal model with implicit dynamics. Scientific reports10, 1 (2020), 1–12. Ahmed A. A. Osman, Timo Bolkart, and Michael J. Black

    work page 2020

  4. [19]

    InEuropean Conf

    STAR: Sparse Trained Articulated Human Body Regressor. InEuropean Conf. on Computer Vision (ECCV), Vol. LNCS 12355. Springer, 598–613. https://doi.org/10.1007/978-3-030-58539-6_36 Ahmed A A Osman, Timo Bolkart, Dimitrios Tzionas, and Michael J. Black

    doi:10.1007/978-3-030-58539-6_36

  5. [21]

    Shunsuke Saito, Zi-Ye Zhou, and Ladislav Kavan

    Full-body musculoskeletal model for muscle- driven simulation of human gait.IEEE Transactions on Biomedical Engineering63, 10 (2016), 2068–2079. Shunsuke Saito, Zi-Ye Zhou, and Ladislav Kavan

    work page 2016

  6. [22]

    Robert Schleicher, Marlies Nitschke, Jana Martschinke, Marc Stamminger, Björn Es- kofier, Jochen Klucken, and Anne Koelewijn

    Computational bodybuilding: Anatomically-based modeling of human bodies.ACM Transactions on Graphics34, 4 (2015), 1–12. Robert Schleicher, Marlies Nitschke, Jana Martschinke, Marc Stamminger, Björn Es- kofier, Jochen Klucken, and Anne Koelewijn

    work page 2015

  7. [23]

    InProceedings of the 16th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP)(2021-02-08/2021-02-10)

    BASH: Biomechanical Animated Skinned Human for Visualization of Kinematics and Muscle Activity. InProceedings of the 16th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP)(2021-02-08/2021-02-10). Springer, 25–36. https://doi.org/10.5220/0010210600250036 Ajay Seth, Ricardo Matias, António P...

    work page doi:10.5220/0010210600250036 2021

  8. [24]

    https://doi.org/10.1371/ journal.pone.0141028 Karthik Shetty, Annette Birkhold, Srikrishna Jaganathan, Norbert Strobel, Bernhard Egger, Markus Kowarschik, and Andreas Maier

    A Biomechanical Model of the Scapulothoracic Joint to Accurately Capture Scapular Kinematics during Shoulder Movements.PloS One11, 1 (01 2016), 1–18. https://doi.org/10.1371/ journal.pone.0141028 Karthik Shetty, Annette Birkhold, Srikrishna Jaganathan, Norbert Strobel, Bernhard Egger, Markus Kowarschik, and Andreas Maier

    work page 2016

Show all 30 references
  1. [25]

    https: //doi.org/10.1016/j.compbiomed.2023.107383 Shashank Tripathi, Lea Müller, Chun-Hao P

    BOSS: Bones, organs and skin shape model.Computers in Biology and Medicine165 (2023), 107383. https: //doi.org/10.1016/j.compbiomed.2023.107383 Shashank Tripathi, Lea Müller, Chun-Hao P. Huang, Omid Taheri, Michael Black, and Dimitrios Tzionas

    2023

  2. [26]

    https: //doi.org/10.1101/2022.07.07.499061 Timo Von Marcard, Roberto Henschel, Michael J Black, Bodo Rosenhahn, and Gerard Pons-Moll

    OpenCap: 3D human movement dynamics from smartphone videos.bioRxiv(2022). https: //doi.org/10.1101/2022.07.07.499061 Timo Von Marcard, Roberto Henschel, Michael J Black, Bodo Rosenhahn, and Gerard Pons-Moll

    2022 doi

  3. [28]

    https://doi.org/10.1145/3306346.3322983 Haoyang Wang, Riza Alp Güler, Iasonas Kokkinos, George Papandreou, and Stefanos Zafeiriou

    Hand Modeling and Simulation Using Stabilized Magnetic Resonance Imaging.ACM Transactions on Graphics38, 4, Article 115 (July 2019), 14 pages. https://doi.org/10.1145/3306346.3322983 Haoyang Wang, Riza Alp Güler, Iasonas Kokkinos, George Papandreou, and Stefanos Zafeiriou

    2019

  4. [29]

    Rapid bilevel optimization to concurrently solve mus- culoskeletal scaling, marker registration, and inverse kinematic problems for human motion reconstruction.bioRxiv(2022). https://doi.org/10.1101/2022.08.22.504896 arXiv:https://www.biorxiv.org/content/early/2022/08/23/2022....

    2022 doi

  5. [30]

    Adaptable Anatomical Models for Realistic Bone Motion Reconstruction.Comput. Graph. Forum34, 2 (2015), 459–471. https://doi.org/10.1111/cgf.12575

    2015 doi

  6. [1988]

    Bohan Wang, George Matcuk, and Jernej Barbič

    The effects of knee brace hinge design and placement on joint mechanics.Journal of Biomechanics21, 11 (1988), 965–974. Bohan Wang, George Matcuk, and Jernej Barbič

    1988

  7. [2000]

    http://mocap

    CMU Graphics Lab Motion Capture Database. http://mocap. cs.cmu.edu. Accessed: 2012-12-11. Scott L. Delp, Frank C. Anderson, Allison S. Arnold, Peter Loan, Ayman Habib, Chand T. John, Eran Guendelman, and Darryl G. Thelen

    2012

  8. [2003]

    https://doi.org/10.1145/882262.882311 Dragomir Anguelov, Praveen Srinivasan, Daphne Koller, Sebastian Thrun, Jim Rodgers, and James Davis

    The Space of Human Body Shapes: Reconstruction and Parameterization from Range Scans.ACM Transactions on Graphics22, 3 (July 2003), 587–594. https://doi.org/10.1145/882262.882311 Dragomir Anguelov, Praveen Srinivasan, Daphne Koller, Sebastian Thrun, Jim Rodgers, and James Davis

    2003

  9. [2005]

    https://doi.org/10.1145/1073204

    SCAPE: Shape Completion and Animation of People.ACM Transactions on Graphics24, 3 (July 2005), 408–416. https://doi.org/10.1145/1073204. 1073207 Marian Bittner, Wei-Tse Yang, Xucong Zhang, Ajay Seth, Jan van Gemert, and Frans CT van der Helm

    2005 doi

  10. [2007]

    Hao-Shu Fang, Jiefeng Li, Hongyang Tang, Chao Xu, Haoyi Zhu, Yuliang Xiu, Yong- Lu Li, and Cewu Lu

    OpenSim: Open-source soft- ware to create and analyze dynamic simulations of movement.IEEE Transactions on Biomedical Engineering54, 11 (2007), 1940–1950. Hao-Shu Fang, Jiefeng Li, Hongyang Tang, Chao Xu, Haoyi Zhu, Yuliang Xiu, Yong- Lu Li, and Cewu Lu

    2007

  11. [2010]

    Creating and animating subject- specific anatomical models.Comput. Graph. Forum29, 8 (Dec. 2010), 2340–2351. https://doi.org/10.1111/j.1467-8659.2010.01718.x Petr Kadleček, Alexandru-Eugen Ichim, Tiantian Liu, Jaroslav Křivánek, and Ladislav Kavan

    2010

  12. [2012]

    Marilyn Keller, Silvia Zuffi, Michael J

    Zygote body: A new interactive 3-dimensional didactical tool for teaching anatomy.Webmedcentral Anatomy(2012). Marilyn Keller, Silvia Zuffi, Michael J. Black, and Sergi Pujades

    2012

  13. [2013]

    2013), 8 pages

    Anatomy transfer.ACM Transactions on Graphics32, 6, Article 188 (Nov. 2013), 8 pages. https://doi.org/10.1145/2508363. 2508415 Brett Allen, Brian Curless, and Zoran Popović

    2013 doi

  14. [2014]

    2014), 220:1–220:13

    MoSh: Motion and Shape Capture from Sparse Markers.ACM Transactions on Graphics33, 6 (Nov. 2014), 220:1–220:13. https://doi.org/10.1145/2661229.2661273 Naureen Mahmood, Nima Ghorbani, Nikolaus F. Troje, Gerard Pons-Moll, and Michael J. Black

    2014

  15. [2015]

    InIEEE Conf

    Pose-Conditioned Joint Angle Limits for 3D Human Pose Reconstruction. InIEEE Conf. on Computer Vision and Pattern Recognition (CVPR). IEEE, 1446–1455. https://doi.org/10.1109/CVPR.2015.7298751 Dicko Ali-Hamadi, Tiantian Liu, Benjamin Gilles, Ladislav Kavan, François Faure, Oli...

    2015

  16. [2016]

    2016), 13 pages

    Reconstructing Personalized Anatomical Models for Physics-Based Body Animation.ACM Transactions on Graphics35, 6, Article 213 (Nov. 2016), 13 pages. https://doi.org/10.1145/2980179.2982438 Angjoo Kanazawa, Michael J. Black, David W. Jacobs, and Jitendra Malik

    2016

  17. [2018]

    InIEEE Conf

    End- to-end Recovery of Human Shape and Pose. InIEEE Conf. on Computer Vision and Pattern Recognition (CVPR). IEEE, 7122–7131. Manuel Kaufmann, Velko Vechev, and Dario Mylonopoulos. 2022.aitviewer. https: //doi.org/10.5281/zenodo.1234 Robi Kelc

    2022 doi

  18. [2019]

    The Virtual Caliper: Rapid Creation of Metrically Accurate Avatars from 3D Measurements.IEEE Trans. Vis. Comput. Graph.25, 5 (2019), 1887–1897. https://doi.org/10.1109/TVCG.2019.2898748 Apoorva Rajagopal, Christopher L Dembia, Matthew S DeMers, Denny D Delp, Jen- nifer L Hicks...

    2019

  19. [2020]

    InIEEE Conf

    VIBE: Video Inference for Human Body Pose and Shape Estimation. InIEEE Conf. on Computer Vision and Pattern Recognition (CVPR). IEEE, Piscataway, NJ, 5252–5262. https: //doi.org/10.1109/CVPR42600.2020.00530 Charles L Lawson and Richard J Hanson. 1995.Solving least squares prob...

    2020

  20. [2021]

    Marlies Nitschke, Eva Dorschky, Dieter Heinrich, Heiko Schlarb, Bjoern M Eskofier, Anne D Koelewijn, and Antonie J van den Bogert

    The accuracy of several pose estimation methods for 3D joint centre localisation.Scientific reports11, 1 (2021), 20673. Marlies Nitschke, Eva Dorschky, Dieter Heinrich, Heiko Schlarb, Bjoern M Eskofier, Anne D Koelewijn, and Antonie J van den Bogert

    2021

  21. [2022]

    Towards Single Camera Human 3D-Kinematics.Sensors23, 1 (2022),

    2022

  22. [2023]

    https://doi.org/10.1109/TPAMI.2022

    AlphaPose: Whole-Body Regional Multi-Person Pose Estimation and Tracking in Real-Time.IEEE Transactions on Pattern Analysis and Machine Intelligence45, 6 (2023), 7157–7173. https://doi.org/10.1109/TPAMI.2022. 3222784 Benjamin Gilles, Lionel Reveret, and Dinesh Pai

    2023 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.