Care-PD
Bibliographic record
Abstract
<!DOCTYPE html> Overview Please read carefully the terms and conditions and any accompanying documentation at neurips2025.care-pd.ca/terms-of-use before you download and/or use the CARE-PD dataset. Project page: https://neurips2025.care-pd.ca/ CARE-PD is the largest publicly available archive of 3D mesh gait data for Parkinson's Disease (PD) and the first to include data collected across multiple sites. The dataset aggregates 9 cohorts from 8 clinical sites, including 362 participants spanning a range of disease severity. All recordings—whether from RGB video or motion capture—are unified into anonymized SMPL body gait meshes through a curated harmonization pipeline. This dataset enables two main benchmarks: Supervised clinical score prediction: Estimating UPDRS gait scores from 3D meshes Unsupervised motion pretext tasks for Parkinsonian gait representation learning Dataset Contents CARE-PD consists of 9 harmonized datasets: 3DGait – Clinical gait recordings with UPDRS scores BMCLab – Gait recordings with medication status and UPDRS scores (original license: CC BY 4.0) DNE – Contains healthy, Parkinson's, and other neurological conditions (original license: CC BY 4.0) E-LC – Medication status (on/off) and PD subtypes KUL-DT-T – Freezer/non-freezer subtypes PD-GaM – Clinical gait recordings with UPDRS scores T-SDU – Ambient walking recordings T-SDU-PD – PD patient walking with UPDRS scores T-LTC – Ambient walking recordings Canonicalized SMPL files *_canonical.pkl files in the Canonicalized_SMPL_pickles folder keep the same nested dataset format as the original pickles. The canonical versions change only the motion coordinates: pose/trans are rotated so the motion uses a shared coordinate system. x = lateral y = up z = forward They also preprocess translation so: The first frame starts at x=0, z=0 The body stands/walks on y=0 Note: KUL-DT-T and E-LC are the only datasets that are not purely straight walking sequences, so for these two datasets the subject is canonicalized to start facing z+ in the first frame. Data Structure The main SMPL datasets are provided in a standardized format: { "anonymized_subject_id": { "anonymized_walk_id": { "pose": array, # SMPL pose parameters (shape varies by dataset) "trans": array, # Translation data "beta": array, # Body shape parameters (zeros for privacy) "fps": int, # Frames per second (standardized) "UPDRS_GAIT": int, # Clinical score (0-3) or None if unavailable "medication": str, # Medication status or None if unavailable "other": str # Additional labels or None if unavailable } } } Additionally, we provide h36m, HumanML3D, and SMPL_6D formats. Getting Started Please refer to https://github.com/TaatiTeam/CARE-PD for getting started with the dataset. Benchmarks CARE-PD includes data splits to test generalization: 6-Fold (split per subject) Leave-one-subject-out Fixed train-test splits (split per subject) The former two are only provided for the supervised clinical score prediction task. Terms of Use By accessing and using this database (the "Database"), users ("Users") acknowledge and agree to comply with the following conditions: License and Attribution The Database is publicly released under a Creative Commons Attribution-NonCommercial (CC BY-NC 4.0) license. Users must provide appropriate attribution by citing the Database and the original publications associated with each dataset accessed from the Database. Data Privacy and Ethics Users must not attempt to identify, contact, or otherwise compromise the anonymity of any individuals whose data is included in the Database. All use of the data must comply with applicable ethical guidelines and legal regulations, including privacy laws (e.g., GDPR, HIPAA, PIPEDA). Data Handling and Security Users must maintain appropriate data security measures to prevent unauthorized access, sharing, or use of the data. Users are encouraged, but not required, to direct third parties to the original Database URL rather than re-hosting the data. Intellectual Property Notice Copyright and other rights remain with the original data providers. Disclaimer of Warranty Use of the Database is subject to Section 5 (Disclaimer of Warranties and Limitation of Liability) of the CC BY-NC 4.0 licence. By using the Database, Users expressly acknowledge and agree to abide by these Terms of Use. Citation If you use CARE-PD in your research, please cite: Adeli V, Klabučar I, Rajabi J, Filtjens B, Mehraban S, Wang D, Seo H, Hoang T-H, Do MN, Muller C, Neves de Oliveira C, Boari Coelho D, Ginis P, Gilat M, Nieuwboer A, Spildooren J, McKay JL, Kwon H, Clifford G, Esper CD, Factor SA, Genias I, Dadashzadeh A, Shum L, Whone A, Mirmehdi M, Iaboni A, Taati B. CARE-PD: A Multi-Site Anonymized Clinical Dataset for Parkinson’s Disease Gait Assessment. In: Advances in Neural Information Processing Systems (NeurIPS); 2025. Additionally, please cite the relevant datasets: 3DGait Diwei Wang, Chaima Zouaoui, Jinhyeok Jang, Hassen Drira, and Hyewon Seo. 2023. Video-Based Gait Analysis for Assessing Alzheimer’s Disease and Dementia with Lewy Bodies. In Applications of Medical Artificial Intelligence: Second International Workshop, AMAI 2023, Held in Conjunction with MICCAI 2023, Vancouver, BC, Canada, October 8, 2023, Proceedings. Springer-Verlag, Berlin, Heidelberg, 72–82. https://doi.org/10.1007/978-3-031-47076-9_8 BMCLab Shida TKF, Costa TM, de Oliveira CEN, de Castro Treza R, Hondo SM, Los Angeles E, Bernardo C, Dos Santos de Oliveira L, de Jesus Carvalho M, Coelho DB. A public data set of walking full-body kinematics and kinetics in individuals with Parkinson's disease. Front Neurosci. 2023 Feb 16;17:992585. doi: 10.3389/fnins.2023.992585. PMID: 36875659; PMCID: PMC9978741. DNE Hoang TH, Zallek C, Do MN. Smartphone-Based Digitized Neurological Examination Toolbox for Multi-test Neurological Abnormality Detection and Documentation. IEEE J Biomed Health Inform. 2024 Aug 26;PP. doi: 10.1109/JBHI.2024.3439492. Epub ahead of print. PMID: 39186431. Hoang TH, Zehni M, Xu H, Heintz G, Zallek C, Do MN. Towards a Comprehensive Solution for a Vision-Based Digitized Neurological Examination. IEEE J Biomed Health Inform. 2022 Aug 26(8):4020-4031. doi: 10.1109/JBHI.2022.3167927. Epub 2022 Aug 11. PMID: 35439148; PMCID: PMC9707344. E-LC Lucas McKay J, Goldstein FC, Sommerfeld B, Bernhard D, Perez Parra S, Factor SA. Freezing of Gait can persist after an acute levodopa challenge in Parkinson's disease. NPJ Parkinsons Dis. 2019 Nov 22;5:25. doi: 10.1038/s41531-019-0099-z. PMID: 31799377; PMCID: PMC6874572. Kwon H, Clifford GD, Genias I, Bernhard D, Esper CD, Factor SA, McKay JL. An Explainable Spatial-Temporal Graphical Convolutional Network to Score Freezing of Gait in Parkinsonian Patients. Sensors (Basel). 2023 Feb 4;23(4):1766. doi: 10.3390/s23041766. PMID: 36850363; PMCID: PMC9968199. KUL-DT-T Spildooren J, Vercruysse S, Desloovere K, Vandenberghe W, Kerckhofs E, Nieuwboer A. Freezing of gait in Parkinson's disease: the im
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.008 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.003 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.003 | 0.002 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.313 | 0.307 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".