ChildGait-MV

Dataset Description

Gait abnormalities are common in children with neurological and musculoskeletal disorders. However, conventional three-dimensional gait analysis is costly, technically demanding, and not widely accessible. Computer vision-based markerless gait assessment offers a promising alternative, but its development and validation are hindered by the limited availability of clinically annotated pediatric gait datasets.

Here, we introduce ChildGait-MV, a multi-view pediatric gait dataset comprising 145 children with diverse gait disorders and 1,728 gait recordings originally collected from sagittal and coronal views using RGB cameras and smartphones. To protect participant privacy, raw gait videos are not publicly released. Instead, the dataset provides machine-readable two-dimensional (2D) human pose sequences, pose-derived gait motion parameters, human segmentation masks, bounding boxes, and comprehensive clinical metadata.

Clinical annotations encompass diagnostic information, cerebral palsy (CP) gait patterns, Gross Motor Function Classification System (GMFCS) levels, and detailed Edinburgh Visual Gait Scale (EVGS) scores covering 17 parameters for each lower limb. Technical validation demonstrated high inter-rater agreement and annotation reliability, supporting the consistency and quality of the clinical annotations.

ChildGait-MV provides an open resource for developing and benchmarking computer vision-based pediatric gait assessment methods, pathological gait recognition algorithms, and clinically interpretable markerless gait analysis systems.

Dataset Overview

Characteristic	Description
Dataset name	ChildGait-MV
Participants	145 children
Original gait recordings	1,728
Original camera views	Sagittal and coronal
Acquisition devices	RGB cameras and smartphones
Publicly released raw videos	No
2D human pose sequences	Yes
Pose-derived gait motion parameters	Yes
Human segmentation masks	Yes
Bounding boxes	Yes
Clinical metadata	Diagnosis, CP gait patterns, GMFCS levels, and EVGS scores
Clinical assessment	17 EVGS parameters per lower limb

Data Contents

The publicly available dataset consists of structured data derived from the original gait recordings. It includes:

• 2D human pose sequences: Machine-readable skeletal keypoints extracted from the original gait videos.
• Gait motion parameters: Pose-derived parameters characterizing gait movement.
• Human segmentation masks: Segmentation annotations derived from the original recordings.
• Bounding boxes: Spatial localization information for human subjects.
• Clinical metadata: Diagnostic information and clinical gait assessment results.

Note: Original gait videos are not included in the public release. The publicly available data are derived representations and annotations of the original recordings.

Clinical Annotations

The dataset includes clinically relevant annotations to facilitate the development and evaluation of clinically interpretable gait analysis methods.

• Diagnosis: Clinical diagnostic categories of the participating children.
• CP gait patterns: Clinical classification of gait patterns in children with cerebral palsy.
• GMFCS: Gross Motor Function Classification System levels for children with cerebral palsy.
• EVGS: Edinburgh Visual Gait Scale scores covering 17 parameters for each lower limb.

EVGS provides structured clinical descriptions of gait deviations across different anatomical regions and gait characteristics, supporting item-level analysis and clinically meaningful model evaluation.

Data Collection

Original gait recordings were collected from children with diverse gait disorders using RGB cameras and smartphones. Videos were acquired from sagittal and coronal views to capture complementary aspects of gait movement.

The publicly released dataset contains structured representations and clinical annotations derived from these recordings rather than the original video files.

Potential Applications

ChildGait-MV is intended to support research in:

• Computer vision-based pediatric gait assessment.
• Human pose estimation and gait representation learning.
• Markerless motion analysis.
• Recognition and classification of pathological gait patterns.
• Automated and interpretable EVGS assessment.
• Multi-view gait analysis.
• Clinical decision support for pediatric rehabilitation.
• Development and benchmarking of pose-based gait analysis algorithms.

Relevant Publication

A subset of the original recordings included in ChildGait-MV was used in our previous study published at ECCV 2026:

Shen, Y., Li, B., Huang, M., Liu, Y., Cao, X., Jin, J., Li, Z., Liu, A., Kim, J., Zhu, J., et al. (2026). Decoding children's gait behavior. In Proceedings of the European Conference on Computer Vision (ECCV) (pp. 113–136). Springer.

Authors and Affiliations

Meihuan Huang¹˒²†, Boyi Li³†, Long Xu⁴, Yifan Shen⁵, Turong Chen¹, Xue Bai¹, Zhenkun Gu⁶, Jianguo Cao¹, Meizhen Huang², Richard Huan Xu², and Xu Cao⁵˒⁷*

¹ Department of Rehabilitation Medicine, Shenzhen Children's Hospital, Shenzhen, China

² Department of Rehabilitation Science, The Hong Kong Polytechnic University, Hong Kong, China

³ Department of Electrical and Computer Engineering, University of Illinois Urbana-Champaign, USA

⁴ Department of Pediatric Rehabilitation, Xiangya Boai Rehabilitation Hospital, Changsha, China

⁵ Department of Computer Science, University of Illinois Urbana-Champaign, USA

⁶ Department of Orthopedics, Shenzhen Children's Hospital, Shenzhen, China

⁷ PediaMed AI, USA

† These authors contributed equally to this work.

• Corresponding authors:

• Xu Cao: xucao@pediamed.ai

License

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

Users are required to provide appropriate credit to the dataset creators and indicate if changes were made.

Citation

If you use ChildGait-MV in your research, please cite the dataset as follows:

@misc{meihuan_huang_2026,
    author       = {Meihuan Huang and Boyi Li and Long Xu and Yifan Shen and Turong Chen and Xue Bai and Zhenkun Gu and Jianguo Cao and Meizhen Huang and Richard Huan Xu and Xu Cao},
    title        = {ChildGait-MV (Revision 835d17c)},
    year         = 2026,
    url          = {https://huggingface.co/datasets/PediaMedAI/ChildGait-MV},
    doi          = {10.57967/hf/11529},
    publisher    = {Hugging Face}
}

Please also cite the related ECCV 2026 publication when using data overlapping with the previously published study.