Pith. sign in

REVIEW 3 major objections 5 minor 28 references

Representing and Detecting Label Ambiguity in IMU-Based Exercise Evaluation

T0 review · 3 major / 5 minor · reviewed 2026-07-11 · grok-4.5

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

Pith's one-line read Training IMU exercise classifiers on full label distributions flags borderline repetitions without hurting classification accuracy.

desk verdict Solid classification parity with a useful AGLD construction, but the headline ambiguity-detection gains are partly circular because AGLD is both train target and eval reference. read the letter →

arxiv 2607.04842 v1 pith:CWUBIYJC submitted 2026-07-06 cs.LG

Andreas Spilz , Heiko Oppel , Michael Munz This is my paper · ORCID
classification cs.LG
keywords LabelAmbiguityDistributionLearningPhysiotherapeuticExerciseEvaluationIMUKullback-LeiblerdivergenceborderlinerepetitionsFunctionalMovementScreen
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

Home physiotherapy systems that score exercises from wearable sensors currently force every repetition into one hard category, even when the movement sits near a boundary where trained raters disagree. This paper shows how to build a soft label distribution for each repetition automatically, by running a rule-based scorer many times with slightly varied thresholds that mimic rater noise. A network trained to match those full distributions equals or beats ordinary one-hot training when picking the main class on four IMU datasets, while also recovering which repetitions are ambiguous and which two classes compete. The practical payoff is feedback that can say a movement lies between two categories instead of issuing an arbitrary hard verdict that erodes patient trust.

What carries the argument

Automatically Generated Label Distributions (AGLD): one hundred simulated raters that share the same physiotherapist-defined criteria but draw independent thresholds from normals around fitted means, so relative class frequencies become a soft target that encodes borderline ambiguity without a large human rater pool.

What would settle it

Gather a multi-rater exercise set in which each rater also marks every repetition as ambiguous or clear; if AGLD entropy and the network's predicted entropy then fail to rank the marked-ambiguous cases above the clear ones, the claim that AGLD captures real ambiguity fails.

Watch

Extended reading notes

Core claim

A network trained with a Kullback-Leibler objective to reproduce automatically generated label distributions matches or exceeds a one-hot cross-entropy baseline on macro-F1 for the leading class across four IMU exercise datasets, and recovers ambiguous repetitions and their two competing classes more reliably from the predicted output distribution.

Load-bearing premise

The method assumes that sampling independent criterion thresholds from normal distributions with fixed heuristic widths produces label distributions whose entropy and top classes faithfully match the ambiguity human raters would actually perceive.

Editorial extensions

If this is right

  • Feedback systems can flag borderline repetitions and name both competing classes instead of forcing a single hard assignment.
  • Label-distribution training adds usable ambiguity information at no cost to classification of the leading class.
  • Predicted entropy can serve as a practical detector of ambiguous executions once a low threshold is chosen.
  • The same AGLD construction applies to both nominal fault categories and ordinal FMS scores.

Reading between the lines

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

  • Simulated-rater distributions could transfer to other criteria-based clinical scores that suffer boundary disagreement without needing multi-annotator pools.
  • Calibrating the threshold widths against real multi-rater data would turn AGLD entropy into a drop-in proxy for interrater reliability on new exercises.
  • Soft targets of this form may reduce the apparent arbitrariness patients feel when nearly identical movements receive opposite hard scores.
  • Extending evaluation past the top-two classes is needed before multi-way ambiguity is fully represented.
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

3 major / 5 minor

Summary. The paper proposes Automatically Generated Label Distributions (AGLD) for IMU-based exercise evaluation: a rule-based labeling pipeline is run with criterion thresholds sampled from N(μc, σc²) to produce a class distribution per repetition without a large human rater pool. A CNN is trained with Kullback–Leibler divergence to match the full AGLD (ambiguity approach) and compared to a one-hot cross-entropy baseline that collapses AGLD to its mode. On four datasets (RD, RGS, DS, HS), the ambiguity approach matches or exceeds baseline macro-F1 for the most relevant class (Table 1) and yields higher Famb1 and Ftop21 across intermediate entropy thresholds (Figs. 3–4), with hierarchical outcome trees at τ = 0.47 bit. The authors conclude that representing label distributions in the training target adds ambiguity information at no cost to classification.

Significance. If the AGLD is accepted as a useful proxy for borderline executions, the work is practically relevant for home-based physiotherapy feedback systems: it offers a concrete way to flag repetitions between categories and report competing classes rather than forcing a hard label. Strengths include evaluation on four real IMU datasets, participant/class/entropy-stratified five-fold CV, metrics reported across the full τ range rather than a single threshold, and an explicit hierarchical breakdown of detection errors. Classification parity (or better) under distributional training is a clean, useful empirical result. The main scientific contribution is the AGLD construction plus the demonstration that KLD training recovers distributional structure without harming mode classification; the clinical claim about human-perceived ambiguity is more provisional.

major comments (3)
  1. [§2.4–2.6, Eqs. (1)–(2); abstract/conclusion] Sections 2.4, 2.6 and Eqs. (1)–(2): Famb1 and Ftop21 treat AGLD entropy H(p) and top2(p) as ground truth, while the ambiguity model is trained with DKL(p ∥ q) to reproduce that same p. The baseline collapses p to one-hot. Higher Famb1/Ftop21 is therefore largely expected once the network approximates p; it does not independently show recovery of clinical borderline cases. The abstract and conclusion should state this dependency explicitly (e.g., “more reliably recovers the AGLD-defined ambiguity”) or add an evaluation that does not use AGLD as both target and reference.
  2. [§3.5, Figure 6; §4.4] Section 3.5 and Figure 6 provide the only external check of AGLD entropy against human rater disagreement (DS/HS only). Precision stays ~0.3–0.4 across the informative τ range, and raters never labeled ambiguity directly—only scores. This is too weak to support the claim that the system “detects ambiguity … more reliably” in the clinical sense. Either collect direct ambiguous/unambiguous rater labels on a subset, or substantially qualify the claim and treat Figure 6 as a consistency check rather than validation.
  3. [§2.2, §4.5, Figure 7] Section 2.2 and Limitations 4.5: σc is fixed heuristically (5°, 5 cm, 0.5 m/s) with independent draws per criterion, and no sensitivity analysis is reported. Because AGLD entropy (and thus Famb1/Ftop21 and the τ = 0.47 split) is a direct function of σc (illustrated in Fig. 7), the ambiguity-detection results are not shown to be robust to this free parameter. A modest sensitivity study (e.g., scale σc by 0.5× and 2× and re-report Famb1/Ftop21 and classification) is needed for the load-bearing half of the main claim.
minor comments (5)
  1. [Abstract; Table 1; §4.1] HS FMS 1 has only eight repetitions (Fig. 1); the large Fcls1 gain (0.32→0.63 in Table 1) is correctly flagged as unreliable in §4.1 but still appears in the abstract’s unqualified “matched or exceeded … on all four datasets.” Soften the abstract wording for HS.
  2. [§2.4, §3.4] The entropy threshold τ = 0.47 bit is motivated by a 0.9/0.1 two-class split (§2.4), but for K = 3 or 4 the same entropy can arise from other mass allocations. A short note on this would help readers interpret the outcome trees.
  3. [§1; §2.4] Related work on multi-annotator and noisy-label methods is discussed, but soft-label / label-smoothing baselines (training with a fixed soft target not derived from AGLD) are not compared. Even a brief discussion would clarify what is specific to AGLD versus any soft target.
  4. [Figure 5; Appendix A] Figure 5 and Appendix A trees are informative but dense; stating absolute counts of true positives for top-2 in the main text (not only percentages of parents) would aid comparison across datasets.
  5. [Figure 5; Appendix A] Minor: “ambivalence” appears in the outcome-tree node labels (Fig. 5, A1–A3) while the paper elsewhere uses “ambiguity”; unify terminology.

Circularity Check

2 steps flagged · score 6.0 of 10

Famb1/Ftop21 gains largely show the KLD network can reproduce its own synthetic AGLD target rather than independently recover clinical ambiguity; classification parity is non-circular.

  1. fitted input called prediction [Sections 2.4 (Eq. 2), 2.6 (Famb1 / Ftop21 definitions), Results 3.2–3.3]
    "The ambiguity approach retains the full label distribution and trains the network to reproduce it. It minimizes the Kullback-Leibler divergence from the AGLD to the network output, L(n)KLD = DKL(p(n) ∥ q(n)) ... Ambiguity detection measures how well an approach recovers from its softmax output whether a repetition is ambiguous. ... the reference is positive when H(p(n)) > τ and the prediction is positive when H(q(n)) > τ, and report the F1-score ... denoted Famb1. ... A reference repetition is positive when H(p(n)) > τ ... and a prediction counts as a true positive only when ... top2(q(n)) equ"

    The KLD objective forces q toward p. Famb1 and Ftop21 then score exactly the distributional properties (entropy and ordered top-2) of that same p. Any network that successfully approximates the training target will therefore outperform a one-hot baseline that never sees the distribution; the reported gains are statistically forced once approximation succeeds and do not constitute an independent external test of clinical ambiguity recovery.

  2. self definitional [Section 2.2 (AGLD generation) + 2.5–2.6 (AGLD as sole reference for all metrics)]
    "Each instance assigns a single class to a repetition, and the relative frequencies of these assignments across the 100 instances define its automatically generated label distribution (AGLD). ... The AGLD is the reference against which the trained network is evaluated in Section 2.6. ... We take the highest-probability class of the AGLD, arg max k p(n)k , as the reference ... a repetition is ambiguous when the entropy of its distribution exceeds a threshold τ. We apply this criterion to both distributions..."

    AGLD is constructed by the authors' own rule-based procedure (thresholds fitted in prior self-work [23], then sampled with heuristic σc). That same synthetic object is then used both as the KLD training target and as the sole ground-truth reference for Fcls1, Famb1 and Ftop21. There is no independent external label of 'ambiguous' for the primary metrics; the detection claim therefore reduces to 'the model recovers the synthetic distribution we defined and trained it on.'

full rationale

The paper's strongest claim has two halves. Classification of the AGLD mode (Fcls1, Table 1) is a fair head-to-head: both the one-hot baseline and the KLD model are scored against the same argmax(p), and the ambiguity approach matching or exceeding the baseline is a genuine empirical result. The second half—detecting ambiguity and the relevant classes more reliably—is not. Sections 2.4 and 2.6 define the ambiguity approach as minimizing DKL(p || q) so that the network output q approximates the full AGLD p, then define Famb1 and Ftop21 by treating H(p) and top2(p) as ground truth and H(q)/top2(q) as the prediction. Once the network can approximate p, superior recovery of entropy and top-2 is expected by construction; the one-hot baseline is deliberately denied the distributional target. The only external check (Section 3.5 / Figure 6, rater disagreement on DS/HS) shows low precision (~0.3–0.4) and is not used for the main Famb1/Ftop21 claims. AGLD construction itself (thresholds fitted in prior self-work then sampled with heuristic σc) is a methodological contribution, not a circular derivation, but it becomes the sole reference for the detection claim. Score 6 reflects partial circularity confined to the ambiguity-detection half of the central claim; the paper is otherwise self-contained and does not smuggle uniqueness theorems or rename known results.

Assumptions & free parameters 6 free parameters · 4 assumptions · 1 invented entities

The central claim rests on the AGLD construction and on treating AGLD entropy/top-2 as ground truth for ambiguity. Free parameters are the heuristic rater-noise widths, the 0.47-bit operating threshold, and inherited network hyperparameters. Domain axioms include the criteria-to-kinematics mapping and the assumption that independent Gaussian threshold noise approximates human rater disagreement. The invented entity is the AGLD itself, which has only partial external validation (mode agreement with original labels; weak precision against rater disagreement).

free parameters (6)
  • σc (angular criteria) = 5°
    Standard deviation of simulated rater thresholds for angular criteria; set heuristically to 5° with no calibration or ablation (Section 2.2). Directly controls AGLD entropy and therefore all ambiguity metrics.
  • σc (distance criteria) = 5 cm
    Heuristic rater-noise width for distance criteria; same role as angular σc.
  • σc (velocity criteria) = 0.5 m/s
    Heuristic rater-noise width for velocity criteria.
  • entropy threshold τ for unambiguous/ambiguous split = 0.47 bit
    Used for fold stratification and for the hierarchical outcome trees; chosen as the entropy of a 0.9/0.1 two-class distribution (0.47 bit) without formal optimization (Section 2.4).
  • μc (best-fit criterion thresholds) = per-criterion, from prior random search
    Means of the threshold sampling distributions, obtained by random search maximizing geometric-mean class-wise F1 against original labels in prior work; inherited and not re-fit here.
  • network hyperparameters (lr, weight decay, dropout, batch size, patience) = as listed in §2.4
    Adam lr 1e-4, weight decay 1e-3, dropout 0.2, batch 32, early-stop patience 30; chosen empirically in prior work and held fixed (Section 2.4).
assumptions (4)
  • ad hoc to paper Human rater disagreement on criteria-based exercise scores is adequately modeled by independent Gaussian noise on per-criterion kinematic thresholds.
    Core modeling choice of AGLD (Section 2.2). Limitations §4.5 note that real disagreement may be correlated across criteria or systematically biased; no independent calibration is provided for RD/RGS.
  • domain assumption Shannon entropy of the label distribution is a graded measure of clinical ambiguity, and a fixed τ separates ambiguous from unambiguous repetitions.
    Used for stratification, Famb1, Ftop21, and outcome trees (Sections 2.3, 2.6). No established clinical standard for τ is cited; authors treat low τ as safer (Section 4.3).
  • domain assumption The most frequent class of the AGLD is a valid single-label reference for classification F1.
    Defines Fcls1 (Section 2.6). Supported by moderate-to-high Forig1 against original labels (Table 2), but agreement is partly built-in because μc were fit to those labels.
  • domain assumption OpenSim inverse-kinematics from the given IMU setups yields kinematic quantities accurate enough for criterion evaluation.
    Inherited from the authors’ prior labeling/augmentation pipeline; required for every AGLD instance.
invented entities (1)
  • Automatically Generated Label Distribution (AGLD)
    purpose: Provide a full class distribution per repetition without a large human rater pool, serving as both training target and evaluation reference for ambiguity.
    Constructed by 100 threshold-perturbed rule-based labelers. Independent evidence is partial: mode agrees reasonably with original labels (Forig1 0.76–0.91), but entropy vs rater-disagreement precision is low (~0.3–0.4) and RD/RGS have no multi-rater reference at all.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Representing and Detecting Label Ambiguity in IMU-Based Exercise Evaluation." pith.science (2026). https://pith.science/paper/CWUBIYJC

@misc{pith2026260704842,
  author       = {Pith},
  title        = {Pith review of: Representing and Detecting Label Ambiguity in IMU-Based Exercise Evaluation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CWUBIYJC}},
  note         = {Machine review of arXiv:2607.04842}
}
read the original abstract

Home-based physiotherapy is performed without supervision, which leads to incorrect execution and motivates systems that assess movement automatically from inertial measurement units (IMUs). Such systems assign each repetition to a category, yet a relevant share of repetitions falls near a class boundary, where even trained raters disagree. Classifiers trained with one-hot labels collapse these borderline repetitions onto a single class and discard this ambiguity. We address this with a method that automatically generates a label distribution per repetition without a large rater pool. We train a network to reproduce the full distribution with a Kullback-Leibler objective, the ambiguity approach, and compare it against a one-hot cross-entropy baseline on four IMU exercise datasets. From the network output we further determine whether a repetition is ambiguous and which classes are relevant to it. The ambiguity approach matched or exceeded the baseline classification on all four datasets, and detected ambiguity and the relevant classes more reliably. Representing the label distribution in the training target therefore adds information about ambiguity at no cost to classification.

Figures

Figures reproduced from arXiv: 2607.04842 by the authors.

Figure 1
Figure 1. Number of repetitions per class for each dataset: (a) RD, (b) RGS, (c) DS, (d) HS. For RD the classes are CE, TL, SUP, and PRO, for RGS they are CE, IKF, IHA, and SH, and for DS and HS they are the FMS scores 1, 2, and 3 (Section 2.1). For DS and HS each bar is split into the count of repetitions on which the raters assigned the same score and the count on which they disagreed, given in parentheses as (agree / disag… view at source ↗
Figure 2
Figure 2. Distribution of the AGLD entropy across repetitions, shown separately for each dataset: (a) RD, (b) RGS, (c) DS, (d) HS. Entropy is given in bits. Within each bar the counts are stacked by the most frequent class of the AGLD, the class assigned the largest share across the 100 labeling instances. The per exercise classes are defined in Section 2.1. The maximum attainable entropy differs between datasets, with log2 4… view at source ↗
Figure 3
Figure 3. Ambiguity detection performance F amb 1 as a function of the entropy threshold τ in the five-fold cross￾validation, with one panel per dataset: (a) RD, (b) RGS, (c) DS, (d) HS. Solid lines show the baseline, dashed lines the ambiguity approach. Lines give the mean over the five folds and shaded bands the standard deviation. The threshold extends to the maximum attainable entropy of each dataset, log2 4 = 2 bit for R… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Combined ambiguity detection and top-2 performance F top2 1 as a function of the entropy threshold τ in the five-fold cross-validation, with one panel per dataset: (a) RD, (b) RGS, (c) DS, (d) HS. Solid lines show the baseline, dashed lines the ambiguity approach. Line…
Figure 5
Figure 5. Figure 5: Outcome tree for RGS at the entropy threshold τ = 0.47 bit (Section 2.4). Solid lines mark the Yes branch and dashed lines the No branch. Each node gives the share of its parent with the absolute count in parentheses. From the detection stage onward each node reports t…
Figure 6
Figure 6. Figure 6: Recall and precision of the AGLD-entropy-based ambiguity detection against the rater-based reference as a function of the entropy threshold τ, for DS and HS. A repetition is positive in the reference when its raters do not assign the same score, and positive in the pre…
Figure 7
Figure 7. Figure 7: Relation between the value a repetition reaches on one example criterion of the DS exercise and the resulting AGLD entropy (Section 2.2). The criterion is the angle of the femur relative to horizontal at the lowest point of the squat. Entropy is given in bits. Each lab…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 18 canonical work pages

  1. [1]

    Effectiveness of Individualized Home-Based Exercise on Turning and Balance Performance Among Adults Older than 50 yrs

    Ashari, A.; Hamid, T.A.; Hussain, M.R.; Hill, K.D. Effectiveness of Individualized Home-Based Exercise on Turning and Balance Performance Among Adults Older than 50 yrs. American Journal of Physical Medicine & Rehabilitation 2016, 95, 355–365. https://doi.org/10.1097/phm.0000000000000388

    work page doi:10.1097/phm.0000000000000388 2016

  2. [2]

    Effect of a Home-Based Exercise Program on Functional Recovery Following Rehabilitation After Hip Fracture: A Randomized Clinical Trial

    Latham, N.K.; Harris, B.A.; Bean, J.F.; Heeren, T.; Goodyear, C.; Zawacki, S.; Heislein, D.M.; Mustafa, J.; Pardasaney , P .; Giorgetti, M.; et al. Effect of a Home-Based Exercise Program on Functional Recovery Following Rehabilitation After Hip Fracture: A Randomized Clinical Trial. JAMA 2014, 311, 700–708. https://doi.org/10.1001/jama.2014.469

    work page doi:10.1001/jama.2014.469 2014

  3. [3]

    Gelaw, A.Y .; Janakiraman, B.; Gebremeskel, B.F.; Ravichandran, H. Effectiveness of Home-based rehabil- itation in improving physical function of persons with Stroke and other physical disability: A systematic review of randomized controlled trials. Journal of Stroke and Cerebrovascular Diseases: The Official Journal of National Stroke Association 2020, 29...

    work page doi:10.1016/j.jstrokecerebrovasdis.2020.104800 2020

  4. [4]

    Home-based prescribed exercise improves balance-related activities in people with Parkinson’s disease and has benefits similar to centre-based exer- cise: a systematic review

    Flynn, A.; Allen, N.E.; Dennis, S.; Canning, C.G.; Preston, E. Home-based prescribed exercise improves balance-related activities in people with Parkinson’s disease and has benefits similar to centre-based exer- cise: a systematic review. Journal of Physiotherapy 2019, 65, 189–199. https://doi.org/10.1016/j.jphys.2019 .08.003

    work page doi:10.1016/j.jphys.2019 2019

  5. [5]

    Patient Involvement With Home-Based Exercise Programs: Can Con- nected Health Interventions Influence Adherence? JMIR mHealth and uHealth 2018, 6, e47

    Argent, R.; Daly , A.; Caulfield, B. Patient Involvement With Home-Based Exercise Programs: Can Con- nected Health Interventions Influence Adherence? JMIR mHealth and uHealth 2018, 6, e47. https: //doi.org/10.2196/mhealth.8518

    work page doi:10.2196/mhealth.8518 2018

  6. [6]

    The majority are not per- forming home-exercises correctly two weeks after their initial instructionan assessor-blinded study

    Faber, M.; Andersen, M.H.; Sevel, C.; Thorborg, K.; Bandholm, T.; Rathleff, M. The majority are not per- forming home-exercises correctly two weeks after their initial instructionan assessor-blinded study . PeerJ 2015, 3, e1102. https://doi.org/10.7717/peerj.1102

    work page doi:10.7717/peerj.1102 2015

  7. [7]

    Do digital interventions increase adherence to home exercise rehabilitation? A systematic review of randomised controlled trials

    Lang, S.; McLelland, C.; MacDonald, D.; Hamilton, D.F. Do digital interventions increase adherence to home exercise rehabilitation? A systematic review of randomised controlled trials. Archives of Physiotherapy 2022, 12, 24. https://doi.org/10.1186/s40945-022-00148-z

    work page doi:10.1186/s40945-022-00148-z 2022

  8. [8]

    Automatic Assessment of Functional Movement Screening Exercises with Deep Learn- ing Architectures

    Spilz, A.; Munz, M. Automatic Assessment of Functional Movement Screening Exercises with Deep Learn- ing Architectures. Sensors 2023, 23, 5. https://doi.org/10.3390/s23010005

    work page doi:10.3390/s23010005 2023

Show all 28 references
  1. [9]

    Functional movement screening: the use of fundamen- tal movements as an assessment of function - part 1

    Cook, G.; Burton, L.; Hoogenboom, B.J.; V oight, M. Functional movement screening: the use of fundamen- tal movements as an assessment of function - part 1. International journal of sports physical therapy 2014, 9, 396–409

    2014

  2. [10]

    Functional movement screening: the use of funda- mental movements as an assessment of function-part 2

    Cook, G.; Burton, L.; Hoogenboom, B.J.; V oight, M. Functional movement screening: the use of funda- mental movements as an assessment of function-part 2. International journal of sports physical therapy 2014, 9, 549–63. 21 of 21

    2014

  3. [11]

    Test-Retest and Interrater Reliability of the Functional Movement Screen

    Shultz, R.; Anderson, S.C.; Matheson, G.O.; Marcello, B.; Besier, T. Test-Retest and Interrater Reliability of the Functional Movement Screen. Journal of Athletic T raining 2013, 48, 331–336. https://doi.org/10.4085/ 1062-6050-48.2.11

    2013

  4. [12]

    A systematic review of interrater reliability of pressure ulcer classification systems

    Kottner, J.; Raeder, K.; Halfens, R.; Dassen, T. A systematic review of interrater reliability of pressure ulcer classification systems. Journal of Clinical Nursing 2009, 18, 315–336. https://doi.org/10.1111/j.1365-2702.20 08.02569.x

    2009 doi

  5. [13]

    Histologic Grading of Breast Cancer: Linkage of Patient Outcome with Level of Pathologist Agreement

    Dalton, L.W.; Pinder, S.E.; Elston, C.E.; Ellis, I.O.; Page, D.L.; Dupont, W.D.; Blamey , R.W. Histologic Grading of Breast Cancer: Linkage of Patient Outcome with Level of Pathologist Agreement. Modern Pathology 2000, 13, 730–735. https://doi.org/10.1038/modpathol.3880126

    doi

  6. [14]

    Label Distribution Learning

    Geng, X. Label Distribution Learning. IEEE T ransactions on Knowledge and Data Engineering 2016, 28, 1734–

    2016

  7. [15]

    https://doi.org/10.1109/TKDE.2016.2545658

    2016 doi

  8. [16]

    Deep Label Distribution Learning With Label Ambiguity

    Gao, B.B.; Xing, C.; Xie, C.W.; Wu, J.; Geng, X. Deep Label Distribution Learning With Label Ambiguity. IEEE T ransactions on Image Processing 2017, 26, 2825–2838. https://doi.org/10.1109/TIP .2017.2689998

    2017 doi

  9. [17]

    Label Enhancement for Label Distribution Learning via Prior Knowledge

    Gao, Y .; Zhang, Y .; Geng, X. Label Enhancement for Label Distribution Learning via Prior Knowledge. In Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence (IJCAI-20), 2020, pp. 3223–3229. https://doi.org/10.24963/ijcai.2020/446

    2020 doi

  10. [18]

    Mitigating label noise through data ambiguation

    Lienen, J.; Hüllermeier, E. Mitigating label noise through data ambiguation. In Proceedings of the Thirty- Eighth AAAI Conference on Artificial Intelligence and Thirty-Sixth Conference on Innovative Applications of Artificial Intelligence and Fourteenth Symposium on Educational ...

    2024 doi

  11. [19]

    Beyond confusion matrix: learning from multiple annotators with awareness of instance features

    Li, J.; Sun, H.; Li, J. Beyond confusion matrix: learning from multiple annotators with awareness of instance features. Machine Learning 2023, 112, 1053–1075. https://doi.org/10.1007/s10994-022-06211-x

    2023 doi

  12. [20]

    Using Ranking-CNN for Age Estimation

    Chen, S.; Zhang, C.; Dong, M.; Le, J.; Rao, M. Using Ranking-CNN for Age Estimation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 5183–5192

    2017

  13. [21]

    Automatic Evaluation of Functional Movement Screening Based on Attention Mechanism and Score Distribution Prediction

    Lin, X.; Huang, T.; Ruan, Z.; Yang, X.; Chen, Z.; Zheng, G.; Feng, C. Automatic Evaluation of Functional Movement Screening Based on Attention Mechanism and Score Distribution Prediction. Mathematics 2023, 11, 4936. https://doi.org/10.3390/math11244936

    2023 doi

  14. [22]

    Automatic Evaluation Method for Functional Movement Screening Based on a Dual-Stream Network and Feature Fusion

    Lin, X.; Chen, R.; Feng, C.; Chen, Z.; Yang, X.; Cui, H. Automatic Evaluation Method for Functional Movement Screening Based on a Dual-Stream Network and Feature Fusion. Mathematics 2024, 12, 1162. https://doi.org/10.3390/math12081162

    2024 doi

  15. [23]

    Automatic Evaluation Method for Functional Move- ment Screening Based on Multi-Scale Lightweight 3D Convolution and an EncoderDecoder

    Lin, X.; Liu, Y .; Feng, C.; Chen, Z.; Yang, X.; Cui, H. Automatic Evaluation Method for Functional Move- ment Screening Based on Multi-Scale Lightweight 3D Convolution and an EncoderDecoder. Electronics 2024, 13, 1813. https://doi.org/10.3390/electronics13101813

    2024 doi

  16. [24]

    Boosting Automatic Exercise Evaluation Through Musculoskeletal Simulation-Based IMU Data Augmentation, 2025

    Spilz, A.; Oppel, H.; Munz, M. Boosting Automatic Exercise Evaluation Through Musculoskeletal Simulation-Based IMU Data Augmentation, 2025. arXiv:2505.24415 [cs], https://doi.org/10.48550/arXiv . 2505.24415

    2025 doi arXiv

  17. [25]

    GAITEX: Human motion dataset of impaired gait and rehabilitation exercises using inertial and optical sensors

    Spilz, A.; Oppel, H.; Werner, J.; Stucke-Straub, K.; Capanni, F.; Munz, M. GAITEX: Human motion dataset of impaired gait and rehabilitation exercises using inertial and optical sensors. Scientific Data 2025, 13, 11. https://doi.org/10.1038/s41597-025-06439-x

    2025 doi

  18. [26]

    Xsens MVN: Full 6DOF human motion tracking using miniature inertial sensors

    Roetenberg, D.; Luinge, H.; Slycke, P . Xsens MVN: Full 6DOF human motion tracking using miniature inertial sensors. Xsens Motion T echnol. BV T ech. Rep.2009, 3

    2009

  19. [27]

    OpenSim: open-source software to create and analyze dynamic simulations of movement

    Delp, S.L.; Anderson, F.C.; Arnold, A.S.; Loan, P .; Habib, A.; John, C.T.; Guendelman, E.; Thelen, D.G. OpenSim: open-source software to create and analyze dynamic simulations of movement. IEEE transactions on bio-medical engineering 2007, 54, 1940–1950. https://doi.org/10.11...

    2007 doi

  20. [28]

    Animating rotation with quaternion curves

    Shoemake, K. Animating rotation with quaternion curves. SIGGRAPH Comput. Graph. 1985, 19, 245–254. https://doi.org/10.1145/325165.325242. Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) a...

    1985 doi

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.