Pith. sign in

REVIEW 3 major objections 5 minor 129 references

This paper claims that a purpose-built video model, ChildGait-Video, can score children's gait from standard RGB video, reaching 70–93% per-item accuracy on 34 clinical EVGS items.

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 new 1,185-video pediatric gait dataset with EVGS labels, plus a VideoMAE-based model, reaches ~84% average accuracy on 34 clinical gait items — far above MLLMs and prior gait models.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection The CGV dataset is the real contribution; the method's headline accuracy depends on human-corrected keypoints/masks, so the 'from standard RGB video' claim is currently overstated. the 3 major comments →

arxiv 2608.00371 v1 pith:5QY2KE6I submitted 2026-08-01 cs.CV

Decoding Children's Gait Behavior

Yifan Shen , Boyi Li , Meihuan Huang , Yuanzhe Liu , Xu Cao , Jinyang Jin , Zhengyuan Li , Anglin Liu
show 7 more authors
Junho Kim Jingyuan Zhu Lan Fangzhou Jianguo Cao Jintai Chen Ismini Lourentzou James Matthew Rehg
This is my paper
classification cs.CV
keywords children's gaitgait analysisEdinburgh Visual Gait ScoreEVGSpediatric gait datasetvideo action recognitionvideo foundation modelkinematic prompting
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 fine-grained pediatric gait analysis from ordinary RGB video as a new computer-vision problem and presents CGV, the first open-sourced children's gait video dataset, with 1,185 clips from 110 children and expert EVGS scores for 17 items per limb. Its central claim is that existing tools—multimodal language models and adult-focused gait recognition networks—fail on this task, while a domain-specific video model, ChildGait-Video, reaches 70–93% per-item accuracy (average 84%, F1 0.83), a large gain over the best baselines. If this holds, it would enable scalable, camera-only screening for gait disorders such as cerebral palsy without expensive motion-capture labs. The method works by overlaying skeletal keypoints as token-level prompts and pruning background patches using segmentation masks before feeding a video Transformer trained end-to-end.

Core claim

This paper establishes that automated pediatric gait scoring can be performed from standard RGB video by framing each EVGS item as a binary typical/atypical classification. It contributes the CGV dataset and the ChildGait-Video model, which combines a video masked-autoencoder backbone, token-level kinematic prompting (skeletal keypoints rendered onto each frame), and mask-guided patch pruning (keeping only patches that overlap the child's segmentation mask). On 34 bilateral scoring items, ChildGait-Video achieves 70–93% per-item accuracy and average F1 0.83, whereas multimodal language models and gait-recognition baselines remain near chance (50–60% and 45–56%, respectively). The authors int

What carries the argument

Token-level kinematic prompting: skeletal keypoints and their connections are drawn directly onto each RGB frame before patch embedding, so the model's attention can bind joint-level anatomical cues to appearance features. Mask-guided patch pruning: a binary segmentation mask of the child is used to deterministically drop background patches, forcing self-attention to focus on the subject and reducing computational load. Both are injected into a video masked-autoencoder Transformer fine-tuned end-to-end with a soft cross-entropy objective, with each EVGS item predicted as an independent binary classification.

Load-bearing premise

The 70–93% accuracies assume that the keypoints and masks used at inference are the same human-corrected annotations from the dataset, not automatically estimated ones; if automatic estimation is required, accuracy could drop substantially.

What would settle it

Run ChildGait-Video on the CGV test set with automatically detected keypoints and masks (no manual correction) and compare per-item accuracy to the reported 70–93% and F1 0.83. A drop of more than a few points would show the headline results depend on manual annotation rather than raw RGB video alone.

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

If this is right

  • ChildGait-Video outperforms all baselines: 84% average left/right accuracy and F1 0.83, versus 69%/72% for the best video foundation baseline and 50–60% for multimodal language models.
  • The CGV dataset establishes a benchmark for automatic pediatric gait scoring, with expert-validated EVGS labels (ICC 0.93) and synchronized pose and segmentation annotations.
  • General-purpose multimodal language models and adult-centric gait recognition networks are insufficient for fine-grained, phase-specific clinical gait analysis, indicating a need for domain-specific modeling.
  • The two proposed interventions—token-level kinematic prompting and mask-guided patch pruning—each improve accuracy over the base video model, and together produce the reported gains.
  • If the results generalize, clinics could screen for gait abnormalities with ordinary cameras instead of marker-based motion capture, lowering cost and accessibility barriers.

Where Pith is reading between the lines

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

  • The reported accuracies use human-corrected keypoints and masks at inference; if automatic pose and segmentation estimation is required instead, real-world accuracy could be materially lower than 70–93%.
  • The test set is a small, artificially balanced split (about one-seventh of 110 patients) without a disclosed seed or confidence intervals, so the benchmark's stability across clinics, cameras, and populations remains untested.
  • Binarizing EVGS by merging minor and major deviations discards severity information; the model may separate typical from atypical yet fail to distinguish mild from marked abnormalities, which is clinically relevant.
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

3 major / 5 minor

Summary. The paper introduces the Children Gait Video (CGV) dataset, consisting of 1,185 high-frame-rate videos from 110 pediatric subjects with expert EVGS annotations (17 items per limb) plus frame-level keypoints, masks, and bounding boxes. It evaluates zero-shot MLLMs and fine-tuned video/skeleton/gait-recognition baselines on a binarized EVGS classification task, and proposes ChildGait-Video, a VideoMAE-v2-based model that renders skeletal keypoints onto RGB frames and applies mask-guided patch pruning. The authors report per-item accuracies of 70–93%, average left/right accuracies of 84%, and F1 = 0.83, with a McNemar test (b,c) = (28,6), p = 2.3e-4 against VideoMAE v2. The dataset and benchmark are positioned as the first open-sourced children's gait video dataset and a new task for fine-grained pediatric gait assessment.

Significance. If the claims hold, the CGV dataset is a valuable resource: it targets an under-served pediatric population, uses a clinically validated scoring instrument (EVGS), reports high inter-rater agreement (ICC = 0.93) and a human baseline (93.8%), and provides publicly available annotations with patient-disjoint splits and a formal statistical comparison. The failure of current MLLMs and gait-recognition models on this task is a useful negative result for the community. However, the central performance claim is only established for inputs that include human-corrected keypoints and masks, not for raw RGB video as the abstract and Discussion assert. The claim that the method operates 'from standard RGB video' and is ready for scalable deployment therefore requires additional evidence, specifically an evaluation using automatically estimated keypoints and masks at test time. The dataset release also appears to make only pose sequences publicly available, which weakens the 'open-sourced video dataset' claim.

major comments (3)
  1. [§4, §7.2, Abstract, Discussion] The headline that ChildGait-Video works 'from standard RGB video' is not supported by the evaluation as described. Section 4 states keypoints were 'detected with Sapiens-2B and manually adjusted by human annotators' and masks were 'detected with SAM 3 and manually selected.' Section 7.2 then feeds these K_t and M_t into the model via token-level kinematic prompting and mask-guided patch pruning. There is no statement that at test time the keypoints/masks are re-estimated automatically. If the reported 84% L/R-AVG and per-item 70–93% (Table 4) depend on human-corrected annotations, the comparison with VideoMAE v2 and the MLLMs is not apples-to-apples, since those baselines receive only raw RGB. Please report an evaluation where K_t and M_t are obtained fully automatically at test time (e.g., raw Sapiens/SAM outputs before manual correction), and quantify the accuracy drop. Without this, t
  2. [§4, §7.2 Statistical Analysis] The test set is described only as 'randomly selected' and balanced, with a patient ratio of 6:1. No seed, number of test patients/videos, or per-item test sample sizes are provided, even though Table 1 shows per-item video counts varying from 261 to 647. The McNemar test pools (b,c) = (28,6) over all items, but the manuscript does not state how many paired instances are used or whether pooling across items is statistically appropriate. Please report the test-set composition, the random seed, per-item confidence intervals, and ideally results across multiple patient-disjoint splits. This is needed to assess the precision of the 70–93% and 84% figures.
  3. [§4, Abstract] The paper calls CGV 'the first open-sourced children's gait video dataset,' but Section 4 states that 'the public avliable data is the children's pose sequence' and that video data and annotations may not be used for training large foundation models. This suggests that the raw videos are not openly released under the CC BY-NC license. The open-source/video-dataset claim should be clarified: either provide the actual access mechanism for the videos, or adjust the claim to describe the publicly released component accurately. This is load-bearing because the dataset contribution is a central claim of the paper.
minor comments (5)
  1. [§4] Typo: 'public avliable data' should be 'publicly available data.'
  2. [§7.2] The term 'token-level kinematic prompting' is used for rendering keypoints onto RGB frames before patchification. It would help to clarify whether the rendered keypoints are always visible after patch embedding, or whether some keypoints can fall outside retained patches after mask-guided pruning.
  3. [§7.3 and Appendix A] The age-based split (younger vs older children) reports only average accuracies without confidence intervals or significance tests; adding these would strengthen the claim that gait analysis is harder for younger children.
  4. [Table 4] F1 scores are reported only as L/R-F1 averages. Per-item F1 or at least the range of per-item F1 would help identify which items drive the average.
  5. [Appendix E, Table 9] The module ablation shows a clear ordering (MPP > bbox mask > random mask), but there is no discussion of the computational cost of mask-guided pruning despite the claim of reduced overhead. A brief runtime comparison would be useful.

Circularity Check

0 steps flagged

No circular derivation: supervised benchmark with held-out patient split and independent ICC check; human-corrected keypoint/mask input is a deployment/generalization caveat, not a circular reduction.

full rationale

The paper is an empirical machine-learning benchmark rather than a derivation. ChildGait-Video is trained on a train split held out by patient ID to predict binarized EVGS items; the test predictions are not constructed from the labels, and no equation in the paper defines the output as a function of the input annotations. The EVGS scoring criteria in Table 7 are clinical rubrics applied to video, not the model's forward pass. Keypoints and masks are auxiliary inputs (Section 4: 'detected with Sapiens-2B ... and manually adjusted by human annotators' and 'detected with SAM 3 ... and manually selected'), while EVGS labels were produced by pediatricians with an independent second-pediatrician check (ICC = 0.93) and human baseline 93.8%, giving external grounding. The proposed model and its ablations use keypoints/masks, whereas VideoMAE v2 and MLLM baselines receive raw RGB only; this is a fairness/generalization limitation concerning the claim 'from standard RGB video' and scalable deployment, but it is not a circular reduction because the predicted score is not identical to the keypoint/mask input and no parameter is fitted to the test labels. Author self-citations (e.g., [13], [14], [15], [56], [86]) appear in related-work and motivation sections, but the central benchmark result is established by the paper's own controlled experiments, not by those citations. Hence there is no load-bearing self-citation chain, no ansatz smuggled in via citation, and no prediction that reduces by construction to its input.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The paper is empirical: it introduces data, labels, and a supervised model rather than a derivation. The main results rest on hand-chosen hyperparameters and on annotation pipelines whose outputs are manually corrected. No new physical or conceptual entities are postulated.

free parameters (5)
  • Learning rate eta = 1e-4
    Set by hand in the experimental setup; controls optimization and is not derived.
  • Weight decay = 0.05
    Chosen by hand with the optimizer settings in Section 4.
  • Temporal frames T = 16 (ablation at 8, 16, 32)
    Final model uses T=16; ablation shows 8->16 gives +12.9% and 16->32 only +3.7%, so the headline result is tied to this choice.
  • Mask foreground retention threshold = Not specified ('a sufficient proportion')
    Mask-guided patch pruning keeps patches only if the segmentation mask has enough foreground; the exact threshold is unstated and affects all reported results.
  • EVGS binarization = Scores 1 and 2 merged into 'atypical'
    The clinical 3-point scale is collapsed into binary labels, changing task difficulty and the meaning of all accuracy numbers.
axioms (4)
  • domain assumption EVGS scores provided by the author-clinicians are a valid and reliable ground truth for gait quality
    Section 4: annotations by 'experienced pediatricians in the author team' with ICC=0.93 against one additional pediatrician; no external gold-standard validation.
  • domain assumption Sapiens-2B keypoints and SAM3 masks, after manual correction, are accurate enough for clinical gait details
    Section 4 and 7.2: all model inputs rely on these annotations; errors in them transfer to the predicted scores.
  • domain assumption Pre-training on Kinetics-710 transfers to pediatric pathological gait
    Sections 7.1-7.2: ChildGait-Video and VideoMAE are initialized on adult action data; no pediatric pre-training is validated.
  • domain assumption 16 frames at 30 FPS capture the clinically relevant gait events
    Section 7.3: ablation supports sufficiency relative to 8 frames, but the final frame count remains an empirical modeling choice.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Decoding Children's Gait Behavior." pith.science (2026). https://pith.science/paper/5QY2KE6I

@misc{pith2026260800371,
  author       = {Pith},
  title        = {Pith review of: Decoding Children's Gait Behavior},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5QY2KE6I}},
  note         = {Machine review of arXiv:2608.00371}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We introduce a new problem domain for human action recognition: the fine-grained analysis of children's gait behaviors from standard RGB video. We specifically target the ambulatory patterns of children aged 3-17 years. Such behaviors arise naturally in the diagnosis and treatment of several critical developmental and neuromuscular disorders, such as cerebral palsy and hemiplegia. Despite their clinical value, current 3D sensor-based gait analysis systems are expensive, intrusive, and often impractical for young subjects. To address this, we introduce a new dataset comprising over 1,100 high-frame-rate (60 FPS) video sequences from 110 subjects, accompanied by synchronized, anonymized pose sequences. In each session, the child performs a 5-second "walk-around" task, capturing the gait cycle from multiple viewpoints. Crucially, we demonstrate that current state-of-the-art approaches, including gait foundation models and Multimodal Large Language Models (MLLMs), fail to effectively resolve these clinical nuances. We identify the key technical challenges in analyzing these erratic and subtle motor patterns and describe a unified end-to-end framework for decoding fundamental components of pediatric gait. Through comprehensive experimental results, we demonstrate the potential of this dataset to drive novel research questions and establish a rigorous baseline for automated child gait assessment.

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

129 extracted references · 13 linked inside Pith

  1. [1]

    Performance evaluation of various deep learning models in gait recognition using the casia-b dataset.Technologies, 12(12):264, 2024

    Nakib Aman, Md Rabiul Islam, Md Faysal Ahamed, and Mominul Ahsan. Performance evaluation of various deep learning models in gait recognition using the casia-b dataset.Technologies, 12(12):264, 2024

    2024

  2. [2]

    Weargait-pd: An open-access wearables dataset for gait in parkinson’s disease and age-matched controls.Scientific Data, 2026

    Anthony J Anderson, David Eguren, Michael A Gonzalez, Michael Caiola, Naima Khan, Sophia Watkinson, Isabella Zuccaroli, Siegfried S Hirczy, Cyrus P Zabetian, Kelly Mills, 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]

    Gait analysis in children with cerebral palsy

    Stéphane Armand, Geraldo Decoulon, and Alice Bonnefoy-Mazure. Gait analysis in children with cerebral palsy. EFORT open reviews, 1(12):448–460, 2016

    2016

  4. [4]

    Gait analysis methods in rehabilitation.Journal of neuroengineering and rehabilitation, 3(1):4, 2006

    Richard Baker. Gait analysis methods in rehabilitation.Journal of neuroengineering and rehabilitation, 3(1):4, 2006

    2006

  5. [5]

    Gait analysis: clinical facts

    Richard Baker, Alberto Esquenazi, Maria Grazia Benedetti, Kaat Desloovere, et al. Gait analysis: clinical facts. Eur. J. Phys. Rehabil. Med, 52(4):560–574, 2016

    2016

  6. [6]

    Quantitative gait analysis and prediction using artificial intelligence for patients with gait disorders.Scientific Reports, 13(1):23099, 2023

    Nawel Ben Chaabane, Pierre-Henri Conze, Mathieu Lempereur, Gwenolé Quellec, Olivier Rémy-Néris, Sylvain Brochard, Béatrice Cochener, and Mathieu Lamard. Quantitative gait analysis and prediction using artificial intelligence for patients with gait disorders.Scientific Reports, 13(1):23099, 2023

    2023

  7. [7]

    Correlations of the gait profile score and the movement analysis profile relative to clinical judgments.Gait & posture, 32(1):129–132, 2010

    Sarah Beynon, Jennifer L McGinley, Fiona Dobson, and Richard Baker. Correlations of the gait profile score and the movement analysis profile relative to clinical judgments.Gait & posture, 32(1):129–132, 2010

    2010

  8. [8]

    A gait imagery-based brain–computer interface with visual feedback for spinal cord injury rehabilitation on lokomat.IEEE Transactions on Biomedical Engineering, 72(1):102–111, 2024

    Cristian Felipe Blanco-Diaz, Ericka Raiane da Silva Serafini, Teodiano Bastos-Filho, André Felipe Oliveira de Azevedo Dantas, Caroline Cunha do Espirito Santo, and Denis Delisle-Rodriguez. A gait imagery-based brain–computer interface with visual feedback for spinal cord injury rehabilitation on lokomat.IEEE Transactions on Biomedical Engineering, 72(1):1...

    2024

  9. [9]

    Gait analysis in neurorehabilitation: from research to clinical practice.Bioengineering, 10(7):785, 2023

    Mirjam Bonanno, Alessandro Marco De Nunzio, Angelo Quartarone, Annalisa Militi, Francesco Petralito, and Rocco Salvatore Calabrò. Gait analysis in neurorehabilitation: from research to clinical practice.Bioengineering, 10(7):785, 2023

    2023

  10. [10]

    Spatio-temporal gait analysis in children with cerebral palsy using, foot-worn inertial sensors.Gait & posture, 39(1):436–442, 2014

    A Brégou Bourgeois, Benoît Mariani, Kamiar Aminian, PY Zambelli, and CJ Newman. Spatio-temporal gait analysis in children with cerebral palsy using, foot-worn inertial sensors.Gait & posture, 39(1):436–442, 2014

    2014

  11. [11]

    Rank consistent ordinal regression for neural networks with application to age estimation.Pattern Recognition Letters, 140:325–331, 2020

    Wenzhi Cao, Vahid Mirjalili, and Sebastian Raschka. Rank consistent ordinal regression for neural networks with application to age estimation.Pattern Recognition Letters, 140:325–331, 2020

    2020

  12. [12]

    Xu Cao and Jianguo Cao. Commentary: Machine learning for autism spectrum disorder diagnosis–challenges and opportunities–a commentary on schulte-rüther et al.(2022).Journal of Child Psychology and Psychiatry, 64 (6):966–967, 2023

    2022

  13. [13]

    Aggpose: Deep aggregation vision transformer for infant pose estimation.arXiv preprint arXiv:2205.05277, 2022

    Xu Cao, Xiaoye Li, Liya Ma, Yi Huang, Xuan Feng, Zening Chen, Hongwu Zeng, and Jianguo Cao. Aggpose: Deep aggregation vision transformer for infant pose estimation.arXiv preprint arXiv:2205.05277, 2022

    Pith/arXiv arXiv 2022

  14. [14]

    Vitasd: Robust vision transformer baselines for autism spectrum disorder facial diagnosis

    Xu Cao, Wenqian Ye, Elena Sizikova, Xue Bai, Megan Coffee, Hongwu Zeng, and Jianguo Cao. Vitasd: Robust vision transformer baselines for autism spectrum disorder facial diagnosis. InICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 1–5. IEEE, 2023

    2023

  15. [15]

    Workshop on ai for children: Healthcare, psychology, education

    Xu Cao, Jintai Chen, Wenqian Ye, Ana Jojic, Sheila Agyeiwaa Owusu, Sheng Li, Megan Coffee, Sicheng Zhao, and James Matthew Rehg. Workshop on ai for children: Healthcare, psychology, education. InICLR 2025 Workshop Proposals, 2025

    2025

  16. [16]

    Openpose: Realtime multi-person 2d pose estimation using part affinity fields.IEEE transactions on pattern analysis and machine intelligence, 43(1): 172–186, 2019

    Zhe Cao, Gines Hidalgo, Tomas Simon, Shih-En Wei, and Yaser Sheikh. Openpose: Realtime multi-person 2d pose estimation using part affinity fields.IEEE transactions on pattern analysis and machine intelligence, 43(1): 172–186, 2019

    2019

  17. [17]

    Sam 3: Segment anything with concepts

    Nicolas Carion, Laura Gustafson, Yuan-Ting Hu, Shoubhik Debnath, Ronghang Hu, Didac Suris, Chaitanya Ryali, Kalyan Vasudev Alwala, Haitham Khedr, Andrew Huang, et al. Sam 3: Segment anything with concepts. arXiv preprint arXiv:2511.16719, 2025

    Pith/arXiv arXiv 2025

  18. [18]

    Gaitpt: Skeletons are all you need for gait recognition

    Andy Catruna, Adrian Cosma, and Emilian Radoi. Gaitpt: Skeletons are all you need for gait recognition. In 2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition (FG), pp. 1–10. IEEE, 2024. 13

    2024

  19. [19]

    Gaitset: Regarding gait as a set for cross-view gait recognition

    Hanqing Chao, Yiwei He, Junping Zhang, and Jianfeng Feng. Gaitset: Regarding gait as a set for cross-view gait recognition. InProceedings of the AAAI conference on artificial intelligence, volume 33, pp. 8126–8133, 2019

    2019

  20. [20]

    Gaitset: Cross-view gait recognition through utilizing gait as a deep set.IEEE transactions on pattern analysis and machine intelligence, 44(7): 3467–3478, 2021

    Hanqing Chao, Kun Wang, Yiwei He, Junping Zhang, and Jianfeng Feng. Gaitset: Cross-view gait recognition through utilizing gait as a deep set.IEEE transactions on pattern analysis and machine intelligence, 44(7): 3467–3478, 2021

    2021

  21. [21]

    Age- inclusive 3d human mesh recovery for action-preserving data anonymization.arXiv preprint arXiv:2512.05259, 2025

    Georgios Chatzichristodoulou, Niki Efthymiou, Panagiotis Filntisis, Georgios Pavlakos, and Petros Maragos. Age- inclusive 3d human mesh recovery for action-preserving data anonymization.arXiv preprint arXiv:2512.05259, 2025

    arXiv 2025

  22. [22]

    Computer vision and machine learning-based gait pattern recognition for flat fall prediction.Sensors, 22(20):7960, 2022

    Biao Chen, Chaoyang Chen, Jie Hu, Zain Sayeed, Jin Qi, Hussein F Darwiche, Bryan E Little, Shenna Lou, Muhammad Darwish, Christopher Foote, et al. Computer vision and machine learning-based gait pattern recognition for flat fall prediction.Sensors, 22(20):7960, 2022

    2022

  23. [23]

    Steffi L Colyer, Murray Evans, Darren P Cosker, and Aki IT Salo. A review of the evolution of vision-based motion analysis and the integration of advanced computer vision methods towards developing a markerless system.Sports medicine-open, 4(1):24, 2018

    2018

  24. [24]

    Gait analysis alters decision-making in cerebral palsy.Journal of pediatric orthopaedics, 23(3):292–295, 2003

    Robert E Cook, Ingo Schneider, M Elizabeth Hazlewood, Susan J Hillman, and James E Robb. Gait analysis alters decision-making in cerebral palsy.Journal of pediatric orthopaedics, 23(3):292–295, 2003

    2003

  25. [25]

    Alterations in surgical decision making in patients with cerebral palsy based on three-dimensional gait analysis.Journal of Pediatric Orthopaedics, 17(5):608–614, 1997

    Peter A DeLuca, Roy B Davis, Sylvia Õunpuu, Sally Rose, and Robert Sirkin. Alterations in surgical decision making in patients with cerebral palsy based on three-dimensional gait analysis.Journal of Pediatric Orthopaedics, 17(5):608–614, 1997

    1997

  26. [26]

    Retinaface: Single-shot multi-level face localisation in the wild

    Jiankang Deng, Jia Guo, Evangelos Ververas, Irene Kotsia, and Stefanos Zafeiriou. Retinaface: Single-shot multi-level face localisation in the wild. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 5203–5212, 2020

    2020

  27. [27]

    Gait analysis in parkinson’s disease: An overview of the most accurate markers for diagnosis and symptoms monitoring.Sensors, 20(12):3529, 2020

    Lazzaro Di Biase, Alessandro Di Santo, Maria Letizia Caminiti, Alfredo De Liso, Syed Ahmar Shah, Lorenzo Ricci, and Vincenzo Di Lazzaro. Gait analysis in parkinson’s disease: An overview of the most accurate markers for diagnosis and symptoms monitoring.Sensors, 20(12):3529, 2020

    2020

  28. [28]

    Validation of a visual gait assessment scale for children with hemiplegic cerebral palsy.Gait & posture, 23(1):78–82, 2006

    Wendy E Dickens and Michael F Smith. Validation of a visual gait assessment scale for children with hemiplegic cerebral palsy.Gait & posture, 23(1):78–82, 2006

    2006

  29. [29]

    Brain-computer interface controlled robotic gait orthosis.Journal of neuroengineering and rehabilitation, 10(1):111, 2013

    An H Do, Po T Wang, Christine E King, Sophia N Chun, and Zoran Nenadic. Brain-computer interface controlled robotic gait orthosis.Journal of neuroengineering and rehabilitation, 10(1):111, 2013

    2013

  30. [30]

    An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    Pith/arXiv arXiv 2010

  31. [31]

    Revisiting skeleton-based action recognition

    Haodong Duan, Yue Zhao, Kai Chen, Dahua Lin, and Bo Dai. Revisiting skeleton-based action recognition. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 2969–2978, 2022

    2022

  32. [32]

    Gaitpart: Temporal part-based model for gait recognition

    Chao Fan, Yunjie Peng, Chunshui Cao, Xu Liu, Saihui Hou, Jiannan Chi, Yongzhen Huang, Qing Li, and Zhiqiang He. Gaitpart: Temporal part-based model for gait recognition. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 14225–14233, 2020

    2020

  33. [33]

    Exploring deep models for practical gait recognition

    Chao Fan, Saihui Hou, Yongzhen Huang, and Shiqi Yu. Exploring deep models for practical gait recognition. arXiv preprint arXiv:2303.03301, 2023

    Pith/arXiv arXiv 2023

  34. [34]

    Learning gait representation from massive unlabelled walking videos: A benchmark.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45 (12):14920–14937, 2023

    Chao Fan, Saihui Hou, Jilong Wang, Yongzhen Huang, and Shiqi Yu. Learning gait representation from massive unlabelled walking videos: A benchmark.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45 (12):14920–14937, 2023

    2023

  35. [35]

    Opengait: Revisiting gait recognition towards better practicality

    Chao Fan, Junhao Liang, Chuanfu Shen, Saihui Hou, Yongzhen Huang, and Shiqi Yu. Opengait: Revisiting gait recognition towards better practicality. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 9707–9716, 2023

    2023

  36. [36]

    Skeletongait: Gait recognition using skeleton maps

    Chao Fan, Jingzhe Ma, Dongyang Jin, Chuanfu Shen, and Shiqi Yu. Skeletongait: Gait recognition using skeleton maps. InProceedings of the AAAI conference on artificial intelligence, volume 38, pp. 1662–1669, 2024

    2024

  37. [37]

    Opengait: A comprehensive benchmark study for gait recognition towards better practicality.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    Chao Fan, Saihui Hou, Junhao Liang, Chuanfu Shen, Jingzhe Ma, Dongyang Jin, Yongzhen Huang, and Shiqi Yu. Opengait: A comprehensive benchmark study for gait recognition towards better practicality.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025. 14

    2025

  38. [38]

    Alphapose: Whole-body regional multi-person pose estimation and tracking in real-time.IEEE transactions on pattern analysis and machine intelligence, 45(6):7157–7173, 2022

    Hao-Shu Fang, Jiefeng Li, Hongyang Tang, Chao Xu, Haoyi Zhu, Yuliang Xiu, Yong-Lu Li, and Cewu Lu. Alphapose: Whole-body regional multi-person pose estimation and tracking in real-time.IEEE transactions on pattern analysis and machine intelligence, 45(6):7157–7173, 2022

    2022

  39. [39]

    Slowfast networks for video recognition

    Christoph Feichtenhofer, Haoqi Fan, Jitendra Malik, and Kaiming He. Slowfast networks for video recognition. InProceedings of the IEEE/CVF international conference on computer vision, pp. 6202–6211, 2019

    2019

  40. [40]

    The spinal cord injury functional ambulation inventory (sci-fai).Journal of rehabilitation medicine, 33(4):177–181, 2001

    Edelle C Field-Fote, Gerard G Fluet, Scott D Schafer, Eric M Schneider, Robin Smith, Pamela A Downey, and Carla D Ruhl. The spinal cord injury functional ambulation inventory (sci-fai).Journal of rehabilitation medicine, 33(4):177–181, 2001

    2001

  41. [41]

    Gpgait: Generalized pose-based gait recognition

    Yang Fu, Shibei Meng, Saihui Hou, Xuecai Hu, and Yongzhen Huang. Gpgait: Generalized pose-based gait recognition. InProceedings of the IEEE/CVF International Conference on Computer Vision, pp. 19595–19604, 2023

    2023

  42. [42]

    Cut out the middleman: Revisiting pose-based gait recognition

    Yang Fu, Saihui Hou, Shibei Meng, Xuecai Hu, Chunshui Cao, Xu Liu, and Yongzhen Huang. Cut out the middleman: Revisiting pose-based gait recognition. InEuropean Conference on Computer Vision, pp. 112–128. Springer, 2024

    2024

  43. [43]

    Gemini 3 pro: the frontier of vision ai

    Google. Gemini 3 pro: the frontier of vision ai. Online, 2025. URLhttps://blog.google/innovation-and-ai/ technology/developers-tools/gemini-3-pro-vision/

    2025

  44. [44]

    Gait lateral network: Learning discriminative and compact representations for gait recognition

    Saihui Hou, Chunshui Cao, Xu Liu, and Yongzhen Huang. Gait lateral network: Learning discriminative and compact representations for gait recognition. InEuropean conference on computer vision, pp. 382–398. Springer, 2020

    2020

  45. [45]

    Posture-based infant action recognition in the wild with very limited data

    Xiaofei Huang, Lingfei Luan, Elaheh Hatamimajoumerd, Michael Wan, Pooria Daneshvar Kakhaki, Rita Obeid, and Sarah Ostadabbas. Posture-based infant action recognition in the wild with very limited data. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4912–4921, 2023

    2023

  46. [46]

    Infantnet: A large scale dataset for infant body pose and shape estimation

    Zhipeng Hui, Jiahao Wang, Xun Dong, Guofeng Zhang, Xingrui Wang, Jiawei Peng, Qihao Liu, Xiaoding Yuan, Yi Zhang, Junjie Oscar Yin, et al. Infantnet: A large scale dataset for infant body pose and shape estimation. 2025

    2025

  47. [47]

    The new freezing of gait questionnaire: unsuitable as an outcome in clinical trials?Movement disorders clinical practice, 7(2):199–205, 2020

    Femke Hulzinga, Alice Nieuwboer, Bauke W Dijkstra, Martina Mancini, Carolien Strouwen, Bastiaan R Bloem, and Pieter Ginis. The new freezing of gait questionnaire: unsuitable as an outcome in clinical trials?Movement disorders clinical practice, 7(2):199–205, 2020

    2020

  48. [48]

    Haruyuki Iwama, Mayu Okumura, Yasushi Makihara, and Yasushi Yagi. The ou-isir gait database comprising the large population dataset and performance evaluation of gait recognition.IEEE Transactions on Information Forensics and Security, 7(5):1511–1521, 2012

    2012

  49. [49]

    Rtmpose: Real-time multi-person pose estimation based on mmpose.arXiv preprint arXiv:2303.07399, 2023

    Tao Jiang, Peng Lu, Li Zhang, Ningsheng Ma, Rui Han, Chengqi Lyu, Yining Li, and Kai Chen. Rtmpose: Real-time multi-person pose estimation based on mmpose.arXiv preprint arXiv:2303.07399, 2023

    Pith/arXiv arXiv 2023

  50. [50]

    On denoising walking videos for gait recognition

    Dongyang Jin, Chao Fan, Jingzhe Ma, Jingkai Zhou, Weihua Chen, and Shiqi Yu. On denoising walking videos for gait recognition. InProceedings of the Computer Vision and Pattern Recognition Conference, pp. 12347–12357, 2025

    2025

  51. [51]

    Concurrent assessment of gait kinematics using marker-based and markerless motion capture.Journal of biomechanics, 127:110665, 2021

    Robert M Kanko, Elise K Laende, Elysia M Davis, W Scott Selbie, and Kevin J Deluzio. Concurrent assessment of gait kinematics using marker-based and markerless motion capture.Journal of biomechanics, 127:110665, 2021

    2021

  52. [52]

    Anipose: A toolkit for robust markerless 3d pose estimation.Cell reports, 36(13), 2021

    Pierre Karashchuk, Katie L Rupp, Evyn S Dickinson, Sarah Walling-Bell, Elischa Sanders, Eiman Azim, Bingni W Brunton, and John C Tuthill. Anipose: A toolkit for robust markerless 3d pose estimation.Cell reports, 36(13), 2021

    2021

  53. [53]

    Evaluation of the performance of accelerometer-based gait event detection algorithms in different real-world scenarios using the marea gait database.Gait & posture, 51: 84–90, 2017

    Siddhartha Khandelwal and Nicholas Wickström. Evaluation of the performance of accelerometer-based gait event detection algorithms in different real-world scenarios using the marea gait database.Gait & posture, 51: 84–90, 2017

    2017

  54. [54]

    Sapiens: Foundation for human vision models

    Rawal Khirodkar, Timur Bagautdinov, Julieta Martinez, Su Zhaoen, Austin James, Peter Selednik, Stuart Anderson, and Shunsuke Saito. Sapiens: Foundation for human vision models. InEuropean Conference on Computer Vision, pp. 206–228. Springer, 2024. 15

    2024

  55. [55]

    Winnie WT Lam, Yuk Ming Tang, and Kenneth NK Fong. A systematic review of the applications of markerless motion capture (mmc) technology for clinical measurement in rehabilitation.Journal of NeuroEngineering and Rehabilitation, 20(1):57, 2023

    2023

  56. [56]

    The 1st ai children challenge

    Boyi Li, Yifan Shen, Houze Yang, Xu Cao, Guojun Yun, Li Gao, Turong Chen, Long Xu, Jianguo Cao, and Meihuan Huang. The 1st ai children challenge. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 5564–5570, 2026

    2026

  57. [57]

    Uniformerv2: Spatiotemporal learning by arming image vits with video uniformer.arXiv preprint arXiv:2211.09552, 2022

    Kunchang Li, Yali Wang, Yinan He, Yizhuo Li, Yi Wang, Limin Wang, and Yu Qiao. Uniformerv2: Spatiotemporal learning by arming image vits with video uniformer.arXiv preprint arXiv:2211.09552, 2022

    Pith/arXiv arXiv 2022

  58. [58]

    Qwen3-vl-embedding and qwen3-vl-reranker: A unified framework for state-of-the- art multimodal retrieval and ranking.arXiv preprint arXiv:2601.04720, 2026

    Mingxin Li, Yanzhao Zhang, Dingkun Long, Keqin Chen, Sibo Song, Shuai Bai, Zhibo Yang, Pengjun Xie, An Yang, Dayiheng Liu, et al. Qwen3-vl-embedding and qwen3-vl-reranker: A unified framework for state-of-the- art multimodal retrieval and ranking.arXiv preprint arXiv:2601.04720, 2026

    Pith/arXiv arXiv 2026

  59. [59]

    The potential and limitations of vision-language models for human motion understanding: A case study in data-driven stroke rehabilitation.arXiv preprint arXiv:2511.17727, 2025

    Victor Li, Naveenraj Kamalakannan, Avinash Parnandi, Heidi Schambra, and Carlos Fernandez-Granda. The potential and limitations of vision-language models for human motion understanding: A case study in data-driven stroke rehabilitation.arXiv preprint arXiv:2511.17727, 2025

    arXiv 2025

  60. [60]

    An in-depth exploration of person re-identification and gait recognition in cloth-changing conditions

    Weijia Li, Saihui Hou, Chunjie Zhang, Chunshui Cao, Xu Liu, Yongzhen Huang, and Yao Zhao. An in-depth exploration of person re-identification and gait recognition in cloth-changing conditions. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 13824–13833, 2023

    2023

  61. [61]

    Multi-view large population gait database with human meshes and its performance evaluation.IEEE Transactions on Biometrics, Behavior, and Identity Science, 4(2): 234–248, 2022

    Xiang Li, Yasushi Makihara, Chi Xu, and Yasushi Yagi. Multi-view large population gait database with human meshes and its performance evaluation.IEEE Transactions on Biometrics, Behavior, and Identity Science, 4(2): 234–248, 2022

    2022

  62. [62]

    Evaluation and accurate diagnoses of pediatric diseases using artificial intelligence.Nature medicine, 25(3):433–438, 2019

    Huiying Liang, Brian Y Tsui, Hao Ni, Carolina CS Valentim, Sally L Baxter, Guangjian Liu, Wenjia Cai, Daniel S Kermany, Xin Sun, Jiancong Chen, et al. Evaluation and accurate diagnoses of pediatric diseases using artificial intelligence.Nature medicine, 25(3):433–438, 2019

    2019

  63. [63]

    Gait recognition via effective global-local feature representation and local temporal aggregation

    Beibei Lin, Shunli Zhang, and Xin Yu. Gait recognition via effective global-local feature representation and local temporal aggregation. InProceedings of the IEEE/CVF international conference on computer vision, pp. 14648–14656, 2021

    2021

  64. [64]

    Disentangling and unifying graph convolutions for skeleton-based action recognition

    Ziyu Liu, Hongwen Zhang, Zhenghao Chen, Zhiyong Wang, and Wanli Ouyang. Disentangling and unifying graph convolutions for skeleton-based action recognition. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 143–152, 2020

    2020

  65. [65]

    Visual gait analysis: the development of a clinical assessment and scale

    SE Lord, PW Halligan, and DT Wade. Visual gait analysis: the development of a clinical assessment and scale. Clinical rehabilitation, 12(2):107–119, 1998

    1998

  66. [66]

    2d/3d pose estimation and action recognition using multitask deep learning

    Diogo C Luvizon, David Picard, and Hedi Tabia. 2d/3d pose estimation and action recognition using multitask deep learning. InProceedings of the IEEE conference on computer vision and pattern recognition, pp. 5137–5146, 2018

    2018

  67. [67]

    Multi-task deep learning for real-time 3d human pose estimation and action recognition.IEEE transactions on pattern analysis and machine intelligence, 43(8):2752–2764, 2020

    Diogo C Luvizon, David Picard, and Hedi Tabia. Multi-task deep learning for real-time 3d human pose estimation and action recognition.IEEE transactions on pattern analysis and machine intelligence, 43(8):2752–2764, 2020

    2020

  68. [68]

    Karel GB Maathuis, Cees P Van Der Schans, Andries Van Iperen, Hans S Rietman, and Jan HB Geertzen. Gait in children with cerebral palsy: observer reliability of physician rating scale and edinburgh visual gait analysis interval testing scale.Journal of Pediatric Orthopaedics, 25(3):268–272, 2005

    2005

  69. [69]

    The ou-isir gait database comprising the treadmill dataset.IPSJ Transactions on Computer Vision and Applications, 4:53–62, 2012

    Yasushi Makihara, Hidetoshi Mannami, Akira Tsuji, Md Altab Hossain, Kazushige Sugiura, Atsushi Mori, and Yasushi Yagi. The ou-isir gait database comprising the treadmill dataset.IPSJ Transactions on Computer Vision and Applications, 4:53–62, 2012

    2012

  70. [70]

    Expanded and independent validation of the movement disorder society–unified parkinson’s disease rating scale (mds-updrs).Journal of neurology, 260(1):228–236, 2013

    Pablo Martinez-Martin, Carmen Rodriguez-Blazquez, Mario Alvarez-Sanchez, Tomoko Arakaki, Alberto Bergareche-Yarza, Anabel Chade, Nelida Garretto, Oscar Gershanik, Monica M Kurtis, Juan Carlos Martinez- Castrillo, et al. Expanded and independent validation of the movement disorder society–unified parkinson’s disease rating scale (mds-updrs).Journal of neur...

    2013

  71. [71]

    Young labeled faces in the wild (ylfw): a dataset for children faces recognition

    Iurii Medvedev, Farhad Shadmand, and Nuno Gonçalves. Young labeled faces in the wild (ylfw): a dataset for children faces recognition. In2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition (FG), pp. 1–10. IEEE, 2024. 16

    2024

  72. [72]

    Emoreact: a multimodal approach and dataset for recognizing emotional responses in children

    Behnaz Nojavanasghari, Tadas Baltrušaitis, Charles E Hughes, and Louis-Philippe Morency. Emoreact: a multimodal approach and dataset for recognizing emotional responses in children. InProceedings of the 18th acm international conference on multimodal interaction, pp. 137–144, 2016

    2016

  73. [73]

    Reliability and validity of the edinburgh visual gait score for cerebral palsy when used by inexperienced observers.Gait & posture, 28(2):323–326, 2008

    AML Ong, SJ Hillman, and JE Robb. Reliability and validity of the edinburgh visual gait score for cerebral palsy when used by inexperienced observers.Gait & posture, 28(2):323–326, 2008

    2008

  74. [74]

    Introducing gpt-5.2

    OpenAI. Introducing gpt-5.2. Online, 2025. URLhttps://openai.com/index/introducing-gpt-5-2/

    2025

  75. [75]

    3d human pose estimation in video with temporal convolutions and semi-supervised training

    Dario Pavllo, Christoph Feichtenhofer, David Grangier, and Michael Auli. 3d human pose estimation in video with temporal convolutions and semi-supervised training. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 7753–7762, 2019

    2019

  76. [76]

    Gait analysis and clinical correlations in early parkinson’s disease.Functional neurology, 32(1):28, 2017

    Michele Pistacchi, Manuela Gioulis, Flavio Sanson, Ennio De Giovannini, Giuseppe Filippi, Francesca Rossetto, and Sandro Zambito Marsala. Gait analysis and clinical correlations in early parkinson’s disease.Functional neurology, 32(1):28, 2017

    2017

  77. [77]

    Lhm: Large animatable human reconstruction model for single image to 3d in seconds

    Lingteng Qiu, Xiaodong Gu, Peihao Li, Qi Zuo, Weichao Shen, Junfei Zhang, Kejie Qiu, Weihao Yuan, Guanying Chen, Zilong Dong, et al. Lhm: Large animatable human reconstruction model for single image to 3d in seconds. InProceedings of the IEEE/CVF International Conference on Computer Vision, pp. 14184–14194, 2025

    2025

  78. [78]

    Computer vision for clinical gait analysis: A gait abnormality video dataset.IEEE Access, 13:45321–45339, 2025

    Rahm Ranjan, David Ahmedt-Aristizabal, Mohammad Ali Armin, and Juno Kim. Computer vision for clinical gait analysis: A gait abnormality video dataset.IEEE Access, 13:45321–45339, 2025

    2025

  79. [79]

    Observational gait assessment tools in paediatrics–a systematic review.Gait & posture, 40(2):279–285, 2014

    Chandrasekar Rathinam, Andrew Bateman, Janet Peirson, and Jane Skinner. Observational gait assessment tools in paediatrics–a systematic review.Gait & posture, 40(2):279–285, 2014

    2014

  80. [80]

    Edinburgh visual gait score for use in cerebral palsy.Journal of pediatric orthopaedics, 23(3):296–301, 2003

    Heather S Read, M Elizabeth Hazlewood, Susan J Hillman, Robin J Prescott, and James E Robb. Edinburgh visual gait score for use in cerebral palsy.Journal of pediatric orthopaedics, 23(3):296–301, 2003

    2003

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.