Context The data includes 2D human pose estimates of Parkinson's patients performing a variety of tasks (e.g. communication, drinking from a cup, leg agility). Pose estimates were produced using Convolutional Pose Machines (CPM, https://arxiv.org/abs/1602.00134). The goal of this project was to use features derived from videos of Parkinson's assessment to predict the severity of parkinsonism and dyskinesia based on clinical rating scales. Content Data was acquired as part of a study to measure the minimally clinically important difference in Parkinson's rating scales. Participants received a two hour infusion of levodopa followed by up to two hours of observation. During this time, they were assessed at regular intervals and assessments were video recorded for post-hoc ratings by neurologists. There were between 120-130 videos per task. The data includes all movement trajectories (extracted frame-by-frame) from the videos of Parkinson's assessments using CPM, as well as confidence values produced by CPM. Ground truth ratings of parkinsonism and dyskinesia severity are included using the UDysRS, UPDRS, and CAPSIT rating scales. Camera shake has been removed from trajectories (see paper for more details). No other preprocessing has been performed. Files are saved in JSON format. For information on how to deal with files, see data_import_demo.ipynb or view online at https://github.com/limi44/Parkinson-s-Pose-Estimation-Dataset. Acknowledgements We would like to acknowledge the staff and patients at Toronto Western Hospital for their time and assistance in this study. Citations If you use this dataset in your work, please cite the following reference: **M.H. Li, T.A. Mestre, S.H. Fox, and B. Taati, Vision-based assessment of parkinsonism and levodopa-induced dyskinesia with pose estimation, Journal of NeuroEngineering and Rehabilitation, vol. 15, no. 1, p. 97, Nov. 2018. doi:10.1186/s12984-018-0446-z.** You may also find the following paper useful. In this paper, we evaluated the responsiveness of features to clinically relevant changes in dyskinesia severity: **M.H. Li, T.A. Mestre, S.H. Fox, B. Taati, Automated assessment of levodopa-induced dyskinesia: Evaluating the responsiveness of video-based features, Parkinsonism & Related Disorders. (2018). doi:10.1016/j.parkreldis.2018.04.036.** Inspiration In our study, we aimed to evaluate the readiness of off-the-shelf human pose estimation and deep learning for clinical applications in Parkinson's disease. We hope that others may find this dataset useful for furthering progress in technology-based monitoring of neurological disorders. License This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. ## Banner Photo by jesse orrico on Unsplash.