Pith. sign in

REVIEW 2 major objections 6 minor 30 references

Synchronized egocentric video and wearable IMUs give the strongest signals for simulated motor-impairment severity and action recognition.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A synchronized egocentric-video + multi-IMU benchmark of 19 simulated motor-impairment activities yields multimodal Macro-F1 of ~0.78 for severity and ~0.93 for action recognition.

T0 review reviewed 2026-07-11 challenge →

load-bearing objection Solid public multimodal benchmark for simulated motor impairment; multimodal fusion wins on their data, but clinical transfer is explicitly unproven. the 2 major comments →

arxiv 2607.03934 v1 pith:Y7PXNKV5 submitted 2026-07-04 cs.CV

EgoInertia-MI: A Multimodal Egocentric Vision and IMU Benchmark for Motor Impairment Assessment

Fatemah Alhamdoosh , Pietro Pala , Abduallah Mohamed , DK Arvind This is my paper
classification cs.CV
keywords egocentric visionwearable IMUmotor impairment assessmentmultimodal fusionaction recognitionseverity estimationhealthcare sensingbenchmark dataset
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

The paper introduces EgoInertia-MI, a public multimodal dataset of chest-mounted first-person video plus wrist, leg, and chest IMU streams from 17 healthy volunteers performing 19 upper- and lower-body activities under three instructed severity levels: none, mild, and severe. It defines two subject-disjoint benchmarks—three-class impairment severity estimation and 19-class action recognition—and shows that egocentric video already carries strong cues for both, while late fusion of video and IMU embeddings reaches the highest scores. The practical claim is that first-person vision plus wearables can support privacy-aware, everyday motor assessment without the identity leakage and constrained setups of third-person cameras. A sympathetic reader cares because clinical motor scoring is intermittent and sparse, while continuous, naturalistic sensing could track subtle day-to-day fluctuations that matter for care and research.

Core claim

On the EgoInertia-MI benchmark, multimodal late fusion of synchronized egocentric video and multi-site IMU signals reaches approximately 0.78 Macro-F1 for three-class motor-impairment severity estimation and 0.93 Macro-F1 for 19-class action recognition under subject-disjoint five-fold cross-validation, outperforming strong unimodal IMU and video baselines; egocentric video alone already supplies strong impairment cues.

What carries the argument

EgoInertia-MI: a synchronized multimodal dataset and protocol that pairs chest-mounted egocentric video with wrist, leg, and chest IMU streams across 19 activities and three instructed severity levels, evaluated with unimodal sequence/video models and late or cross-attention fusion under subject-disjoint folds.

Load-bearing premise

That motor patterns simulated by healthy people under instruction are close enough to real pathological impairments for the measured multimodal gains to transfer to clinical populations.

What would settle it

Train the same multimodal models on EgoInertia-MI, then test them without fine-tuning on a held-out cohort of real patients with clinically scored severity (e.g., validated tremor or bradykinesia scales); if severity Macro-F1 collapses near chance or far below the reported 0.78, the transfer claim fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Egocentric cameras can replace or complement third-person video for motor assessment while reducing facial and full-body identity exposure.
  • Fusing first-person vision with multi-site IMUs yields more reliable severity estimates than either modality alone on the reported protocol.
  • The public 19-activity, three-severity benchmark enables standardized comparison of future unimodal and multimodal models for impaired movement.
  • Wrist- and chest-centric sensing carry more discriminative signal than leg-only sensing for the simulated upper- and whole-body tasks studied here.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the simulated-to-clinical gap is modest, the same fusion recipe could support continuous home monitoring of medication ON/OFF fluctuations without clinic visits.
  • Privacy-preserving egocentric streams may lower the ethical barrier to collecting larger longitudinal motor datasets than third-person video currently allows.
  • Self-supervised or foundation-model pretraining on the released clips could further close the gap between frozen and fine-tuned video performance for severity estimation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. The paper introduces EgoInertia-MI, a synchronized multimodal dataset of chest-mounted egocentric video and wearable IMUs (wrist, leg, and camera-embedded chest) for motor-impairment analysis. Seventeen healthy volunteers perform 19 upper- and lower-body activities under three instructed severity levels (none, mild, severe), yielding ~1,150 segments (~10 h). Two subject-disjoint 5-fold CV benchmarks are defined: 3-class severity estimation and 19-class action recognition. Unimodal IMU baselines (TCN, LSTM, CNN, HARTransformer, DeepConvLSTM), video baselines (X3D variants, SlowFast, V-JEPA; frozen and fine-tuned), and two fusion strategies (late fusion, temporal cross-attention) are evaluated. Multimodal late fusion reports the best Macro-F1 of 0.784±0.049 (severity) and 0.934±0.013 (action), with fine-tuned SlowFast the strongest unimodal video model and DeepConvLSTM the strongest IMU model for severity.

Significance. If the reported ranking holds, the work supplies a missing public resource that pairs privacy-aware first-person video with quantitative inertial sensing for motor assessment, together with reproducible baselines and subject-independent protocols. The explicit release of code and data, the multi-architecture comparison (including frozen vs fine-tuned video models), and the sensor-ablation analysis are concrete contributions that other groups can build on. The central scientific claim is scoped to performance on this simulated benchmark; the authors themselves flag the simulation-to-clinic gap in §7. Within that scope the empirical result is useful for the community working on wearable and egocentric healthcare sensing.

major comments (2)
  1. §3.4 and §7: severity labels are produced solely by instructing healthy volunteers to mimic mild/severe reductions in speed/coordination plus added tremor/instability. The paper correctly notes that these levels do not map to validated clinical scales and that pathological motor signatures involve rigidity, fatigue, compensation, and medication state. Because the strongest claim (multimodal superiority on severity estimation) is measured against these labels, the manuscript should either (a) add a quantitative check that the simulated IMU/video features exhibit the expected spectral or kinematic signatures of the targeted impairments, or (b) restate the severity-task claim more narrowly as discrimination of instructed execution styles rather than clinical severity. Without one of these, transfer claims remain under-supported even though the internal ranking on EgoInertia-MI is sound.
  2. §5 and Table 1: clip-level training with 5 s windows and recording-level logit averaging is reasonable, yet no ablation is reported on clip duration, padding of short segments, or the effect of averaging versus majority vote / temporal pooling. Because free parameters (clip length, frame subsampling, resolution) are acknowledged and performance margins between late fusion and the best video model are modest (0.784 vs 0.750 Macro-F1), a short sensitivity analysis would strengthen that the multimodal gain is not an artifact of the chosen windowing scheme.
minor comments (6)
  1. Abstract and §1: the reported 0.78 / 0.93 Macro-F1 figures should be tied explicitly to late fusion (Table 1) so readers do not confuse them with cross-attention or unimodal numbers.
  2. Fig. 5 (left): sensor-combination bars would be clearer with numerical Macro-F1 values or error bars matching the fold statistics used elsewhere.
  3. §3.2: the 25.5 Hz IMU rate is justified by tremor Nyquist limits; a brief note on anti-aliasing / filtering would help reproducibility.
  4. Fig. 6: row-normalized confusion matrices are useful; adding absolute sample counts per class (or noting class imbalance) would aid interpretation of the severity off-diagonals.
  5. Minor typography: missing spaces after periods in several places (e.g., “Subtlealterations”, “Accurateassessmentof”) and inconsistent hyphenation of “privacy-aware” / “privacyaware”.
  6. Related work could more explicitly contrast EgoInertia-MI with existing clinical multimodal gait/tremor datasets (e.g., WearGait-PD, PADS) to sharpen the novelty claim around the egocentric + multi-severity design.

Circularity Check

0 steps flagged

No circular derivation: severity/action labels are protocol-assigned, and reported Macro-F1 scores are empirical measurements under subject-disjoint CV, not forced by construction or self-citation.

full rationale

EgoInertia-MI is a dataset-and-benchmark paper. Severity levels (0–2) and activity labels (0–18) are assigned by experimental protocol: healthy volunteers are instructed to perform natural, mild, or severe simulated impairment, then segments are annotated with those labels (Sections 3.4–3.5). Models (IMU CNNs/LSTMs/Transformers, video X3D/SlowFast/V-JEPA, late and cross-attention fusion) are trained and evaluated under subject-disjoint 5-fold CV; recording-level predictions average clip logits. The headline numbers (≈0.78 Macro-F1 severity, ≈0.93 Macro-F1 action) are therefore empirical held-out measurements on Table 1, not quantities obtained by fitting a free parameter that is then re-presented as a prediction, nor results forced by a uniqueness theorem or self-citation chain. Sensor citation of Respeck (Arvind et al.) only identifies hardware and is not load-bearing for the performance claim. The paper’s own §7 limitation (simulated vs clinical impairment) is a transfer/validity concern, not circularity. No equation equates an output to its defining input by construction.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 1 invented entities

This is an empirical dataset + baseline paper. The central performance claims rest on experimental design choices (simulation protocol, sensor placement, clip length, subject splits, model families) rather than free mathematical parameters or invented physical entities. The main load-bearing domain assumption is that instructed simulation by healthy volunteers is a useful proxy for studying multimodal motor-impairment cues.

free parameters (3)
  • clip_duration = 5 seconds
    All recordings are segmented into fixed 5-second clips (with last-frame padding); this design choice affects temporal context available to models and is not derived from first principles.
  • IMU_sampling_rate = 25.5 Hz
    Respeck IMUs record at 25.5 Hz (Nyquist 12.75 Hz), chosen to cover common tremor bands while remaining energy-efficient; rate is a hardware/protocol choice that bounds observable frequencies.
  • video_frame_subsampling_and_resolution = architecture-dependent (e.g. 32×256×256 for SlowFast)
    Architectures receive B∈{13,16,32} frames at F∈{182,256} after resize/crop from 2160×3840@30 FPS; these are standard but free input-configuration choices.
axioms (4)
  • domain assumption Healthy volunteers can produce motion patterns that usefully approximate graded motor impairment (no / mild / severe) under verbal instruction and demonstration.
    Core of the study design (§3.1, §3.4); without it the severity labels have no intended clinical meaning. Authors acknowledge the gap in §7.
  • domain assumption Subject-disjoint 5-fold cross-validation (3 test / 2 val / 12 train subjects) yields an unbiased estimate of generalization across people.
    Stated evaluation protocol (§4); standard but still an assumption given only 17 subjects and demographic skew (mean age 33).
  • domain assumption Temporal alignment via device start timestamps is accurate enough that video clips and T×18 IMU matrices can be treated as synchronized multimodal samples.
    §3.5 synchronization method; residual clock skew is not quantified.
  • standard math Standard deep sequence and video architectures (CNN/LSTM/TCN/DeepConvLSTM/HARTransformer; X3D/SlowFast/V-JEPA) plus late and cross-attention fusion are adequate baselines for the benchmark.
    §5; no claim of optimality, only representative performance.
invented entities (1)
  • EgoInertia-MI dataset / benchmark no independent evidence
    purpose: Provide a public synchronized egocentric-video + multi-IMU corpus with action and simulated-severity labels for motor-impairment research.
    The resource itself is the main contribution; it is an engineered artifact, not a physical postulate. Independent evidence will come from external reuse and eventual clinical transfer studies.

reviewed 2026-07-11 · how reviews work

0 comments
Cite this review

Pith. "Pith review of EgoInertia-MI: A Multimodal Egocentric Vision and IMU Benchmark for Motor Impairment Assessment." pith.science (2026). https://pith.science/paper/Y7PXNKV5

@misc{pith2026260703934,
  author       = {Pith},
  title        = {Pith review of: EgoInertia-MI: A Multimodal Egocentric Vision and IMU Benchmark for Motor Impairment Assessment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y7PXNKV5}},
  note         = {Machine review of arXiv:2607.03934}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Motor impairments, including tremor, bradykinesia, gait abnormalities, and postural instability, are common across many neurological and movement-related conditions. Conventional clinical assessments are often intermittent and may fail to capture subtle temporal variations in motor behavior. While wearable IMUs and third-person video have shown promise for objective motor assessment, third-person recordings raise privacy concerns and require constrained acquisition setups. In contrast, egocentric vision provides a more naturalistic and privacyaware alternative. In this work, we introduce EgoInertia-MI, a multimodal benchmark dataset combining synchronized egocentric video and wearable IMU signals for motor impairment analysis. The dataset contains 19 upper- and lower-body activities performed by healthy volunteers simulating varying levels of motor impairment severity levels: no impairment, mild impairment, and severe impairment. We establish two benchmark tasks: action recognition and motor impairment severity estimation, and evaluate multiple unimodal and multimodal baselines. Experimental results show that egocentric video provides strong cues for motor impairment assessment, while multimodal fusion achieves the best overall performance, reaching 0.78 Macro-F1 for severity estimation and 0.93 Macro-F1 for action recognition. These findings highlight the potential of combining egocentric vision and wearable sensing for ecologically valid and privacy-aware motor assessment. Code and data are available at:https://fatemah-alh.github.io/EgoInertia-MI-Page/.

Figures

Figures reproduced from arXiv: 2607.03934 by Abduallah Mohamed, DK Arvind, Fatemah Alhamdoosh, Pietro Pala.

Figure 1
Figure 1. Figure 1: Overview of EgoInertia-MI. Synchronized egocentric video and wearable IMU streams are processed by a multimodal model for movement understanding and im￾pairment severity estimation [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the EgoInertia-MI dataset, showing 19 upper- and lower-body activities with representative egocentric frames and activity labels [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Representative wrist IMU acceleration signals for the Hand Up/Down task across different impairment levels. 3.5 Annotation and Synchronization Each recording session lasts approximately 20 minutes and comprises multiple activities performed sequentially under the predefined protocol conditions. For every session, three synchronized data streams are acquired: (i) egocentric video with embedded camera IMU me… view at source ↗
Figure 4
Figure 4. Figure 4: Demographic and recording statistics of EgoInertia-MI dataset, including sex and age distributions, anthropometric measurements, and segment duration character￾istics. 3.6 Dataset Overview The proposed dataset comprises approximately 1,150 annotated multimodal seg￾ments collected from 17 participants (9 females and 8 males), corresponding to nearly 10 hours of synchronized recordings. Participants span a b… view at source ↗
Figure 5
Figure 5. Figure 5: (Right) further visualizes the relationship between both tasks across the evaluated approaches. IMU-based models achieved competitive severity es￾timation performance, reaching up to 0.689 ± 0.056 Macro-F1, indicating that inertial signals effectively capture fine-grained temporal motion dynamics and biomechanical patterns. However, IMU-based methods achieved lower action recognition performance (0.568±0.0… view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of normalized confusion matrices for action recognition and severity estimation using IMU-based and video-based architectures. Values are reported as row￾wise percentages. our instruction-based severity levels do not directly map to clinically validated rating scales. Therefore, while this benchmark provides a structured framework for studying movement variations, evaluation on authentic clinica… view at source ↗

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

30 extracted references · 1 canonical work pages

  1. [1]

    In: International Workshop on PRedictive Intelligence In MEdicine

    Acevedo Trebbau, G.T., Bandini, A., Guarin, D.L.: Video-based hand pose estima- tion for remote assessment of bradykinesia in parkinson’s disease. In: International Workshop on PRedictive Intelligence In MEdicine. pp. 241–252. Springer (2023)

    2023

  2. [2]

    Scientific Data (2026)

    Anderson, A.J., Eguren, D., Gonzalez, M.A., Caiola, M., Khan, N., Watkinson, S., Zuccaroli, I., Hirczy, S.S., Zabetian, C.P., Mills, K., et al.: Weargait-pd: An open- access wearables dataset for gait in parkinson’s disease and age-matched controls. Scientific Data (2026)

    2026

  3. [3]

    In: EAI International Conference on Body Area Networks

    Arvind, D., Fischer, D., Bates, C., Kinra, S.: Characterisation of breathing and physical activity patterns in the general population using the wearable respeck monitor. In: EAI International Conference on Body Area Networks. pp. 68–78. Springer (2019)

    2019

  4. [4]

    arXiv preprint arXiv:2506.09985 (2025)

    Assran, M., Bardes, A., Fan, D., Garrido, Q., Howes, R., Muckley, M., Rizvi, A., Roberts, C., Sinha, K., Zholus, A., et al.: V-jepa 2: Self-supervised video models enable understanding, prediction and planning. arXiv preprint arXiv:2506.09985 (2025)

    Pith/arXiv arXiv 2025

  5. [5]

    arXiv preprint arXiv:1803.01271 (2018)

    Bai, S., Kolter, J.Z., Koltun, V.: An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271 (2018)

    Pith/arXiv arXiv 2018

  6. [6]

    Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies8(4), 1–21 (2024)

    Bock, M., Kuehne, H., Van Laerhoven, K., Moeller, M.: Wear: An outdoor sports dataset for wearable and egocentric activity recognition. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies8(4), 1–21 (2024)

    2024

  7. [7]

    IEEE Transactions on Neural Systems and Reha- bilitation Engineering25(10), 1853–1863 (2017)

    Delrobaei, M., Baktash, N., Gilmore, G., McIsaac, K., Jog, M.: Using wearable technology to generate objective parkinson’s disease dyskinesia severity score: pos- sibilities for home monitoring. IEEE Transactions on Neural Systems and Reha- bilitation Engineering25(10), 1853–1863 (2017)

    2017

  8. [8]

    Sensors22(5), 1911 (2022)

    Dirgová Luptáková, I., Kubovčík, M., Pospíchal, J.: Wearable sensor-based human activity recognition with transformer model. Sensors22(5), 1911 (2022)

    1911

  9. [9]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Feichtenhofer, C.: X3d: Expanding architectures for efficient video recognition. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 203–213 (2020)

    2020

  10. [10]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Feichtenhofer, C., Fan, H., Malik, J., He, K.: Slowfast networks for video recog- nition. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 6202–6211 (2019)

    2019

  11. [11]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Gong, X., Mohan, S., Dhingra, N., Bazin, J.C., Li, Y., Wang, Z., Ranjan, R.: Mmg- ego4d: Multimodal generalization in egocentric action recognition. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 6481–6491 (2023)

    2023

  12. [12]

    In: Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition

    Grauman, K., Westbury, A., Byrne, E., Chavis, Z., Furnari, A., Girdhar, R., Ham- burger, J., Jiang, H., Liu, M., Liu, X., et al.: Ego4d: Around the world in 3,000 hours of egocentric video. In: Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition. pp. 18995–19012 (2022)

    2022

  13. [13]

    Inciarte, D.J., Torres-Russotto, D.R.: Movement Disorders, pp. 923–935. Springer International Publishing, Cham (2022). https://doi.org/10.1007/978-3-030-54441- 6_173

    work page doi:10.1007/978-3-030-54441- 2022

  14. [14]

    IEEE journal of biomedical and health informatics27(1), 190– 201 (2022) EgoInertia-MI: Egocentric Vision and IMU for Motor Impairment Analysis 15

    Kaur, R., Motl, R.W., Sowers, R., Hernandez, M.E.: A vision-based framework for predicting multiple sclerosis and parkinson’s disease gait dysfunctions—a deep learning approach. IEEE journal of biomedical and health informatics27(1), 190– 201 (2022) EgoInertia-MI: Egocentric Vision and IMU for Motor Impairment Analysis 15

    2022

  15. [15]

    In: European Conference on Computer Vision

    Liu, Q., Zhang, Y., Bai, S., Yuille, A.: Explicit occlusion reasoning for multi-person 3d human pose estimation. In: European Conference on Computer Vision. pp. 497–

  16. [16]

    In: International Conference on Medical Im- age Computing and Computer-Assisted Intervention

    Lu,M.,Poston,K.,Pfefferbaum,A.,Sullivan,E.V.,Fei-Fei,L.,Pohl,K.M.,Niebles, J.C., Adeli, E.: Vision-based estimation of mds-updrs gait scores for assessing parkinson’s disease motor severity. In: International Conference on Medical Im- age Computing and Computer-Assisted Intervention. pp. 637–647. Springer (2020)

    2020

  17. [17]

    PloS one12(2), e0171346 (2017)

    Moon, Y., McGinnis, R.S., Seagers, K., Motl, R.W., Sheth, N., Wright Jr, J.A., Ghaffari, R., Sosnoff, J.J.: Monitoring gait in multiple sclerosis with novel wearable motion sensors. PloS one12(2), e0171346 (2017)

    2017

  18. [18]

    Scientific data10(1), 918 (2023)

    Morgan, C., Tonkin, E.L., Masullo, A., Jovan, F., Sikdar, A., Khaire, P., Mirmehdi, M., McConville, R., Tourte, G.J., Whone, A., et al.: A multimodal dataset of real world mobility activities in parkinson’s disease. Scientific data10(1), 918 (2023)

    2023

  19. [19]

    In: International Conference on Learning Representations

    Narayanswamy, G., Liu, X., Ayush, K., Yang, Y., Xu, X., Liao, S., Garrison, J., Tailor, S., Sunshine, J., Liu, Y., et al.: Scaling wearable foundation models. In: International Conference on Learning Representations. vol. 2025, pp. 59382–59413 (2025)

    2025

  20. [20]

    Sensors16(1), 115 (2016)

    Ordóñez, F.J., Roggen, D.: Deep convolutional and lstm recurrent neural net- works for multimodal wearable activity recognition. Sensors16(1), 115 (2016). https://doi.org/10.3390/s16010115

    doi:10.3390/s16010115 2016

  21. [21]

    In: Proceedings of the European conference on computer vision (ECCV) workshops

    Pintea, S.L., Zheng, J., Li, X., Bank, P.J., van Hilten, J.J., van Gemert, J.C.: Hand- tremor frequency estimation in videos. In: Proceedings of the European conference on computer vision (ECCV) workshops. pp. 0–0 (2018)

    2018

  22. [22]

    IEEE Access13, 45321– 45339 (2025)

    Ranjan, R., Ahmedt-Aristizabal, D., Armin, M.A., Kim, J.: Computer vision for clinical gait analysis: A gait abnormality video dataset. IEEE Access13, 45321– 45339 (2025)

    2025

  23. [23]

    In: 2025 IEEE 35th International Workshop on Machine Learning for Signal Processing (MLSP)

    Robinson, D., Gupta, A., Qureshi, R., Fu, Q., Shah, M.: Strokevision-bench: A multimodal video and 2d pose benchmark for tracking stroke recovery. In: 2025 IEEE 35th International Workshop on Machine Learning for Signal Processing (MLSP). pp. 1–6. IEEE (2025)

    2025

  24. [24]

    Sensors & Diagnostics 1(3), 387–404 (2022)

    Sharma, A., Singh, A., Gupta, V., Arya, S.: Advancements and future prospects of wearable sensing technology for healthcare applications. Sensors & Diagnostics 1(3), 387–404 (2022)

    2022

  25. [25]

    PhysioNet (2024)

    Varghese, J., Brenner, A., Plagwitz, L., Varghese, J., Brenner, A., Fujarski, M., van Alen, C., Plagwitz, L., Warnecke, T., Goldberger, A., et al.: Pads-parkinsons disease smartwatch dataset. PhysioNet (2024)

    2024

  26. [26]

    Sensors 21(16), 5589 (2021)

    Vijayan, V., Connolly, J.P., Condell, J., McKelvey, N., Gardiner, P.: Review of wearable devices and data collection considerations for connected health. Sensors 21(16), 5589 (2021)

    2021

  27. [27]

    Scientific Data12(1), 1674 (2025)

    Voisard, C., Barrois, R., l’Escalopier, N.d., Vayatis, N., Vidal, P.P., Yelnik, A., Ricard, D., Oudre, L.: A dataset of clinical gait signals with wearable sensors from healthy, neurological, and orthopedic cohorts. Scientific Data12(1), 1674 (2025)

    2025

  28. [28]

    Gait & posture112, 95–107 (2024)

    Vun, D.S.Y., Bowers, R., McGarry, A.: Vision-based motion capture for the gait analysis of neurodegenerative diseases: A review. Gait & posture112, 95–107 (2024)

    2024

  29. [29]

    seen" versus

    Xiang, A.: Being" seen" versus" mis-seen": Tensions between privacy and fairness in computer vision. Harv. JL & Tech.36, 1 (2022)

    2022

  30. [30]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Yang, J., Liu, S., Guo, H., Dong, Y., Zhang, X., Zhang, S., Wang, P., Zhou, Z., Xie, B., Wang, Z., et al.: Egolife: Towards egocentric life assistant. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 28885–28900 (2025)

    2025

This paper was first reviewed by grok-4.5 on July 11, 2026.