Skip to content

Latest commit

 

History

History
55 lines (44 loc) · 1.98 KB

README.md

File metadata and controls

55 lines (44 loc) · 1.98 KB

Automatic Labeling of Parkinson’s Disease Gait Videos with Weak Supervision

Open In Colab

Implementation of "Automatic Labeling of Parkinson’s Disease Gait Videos with Weak Supervision".

.

Environment

Required packages:

  • pytorch
  • pytorch3d
  • snorkel

To install pytorch3d please follow the instructions at
https://github.com/facebookresearch/pytorch3d/blob/master/INSTALL.md
To install snorkel please follow the instructions at
https://www.snorkel.org/get-started/

3D Human Pose

First, train the 3D pose estimator on the Human3.6M dataset. Please first download the data from here and put in the /data directory.

python train.py

Then you need to fine-tune the pose estimator on your multi-view data. We can not provide our PD data due to privacy issues. You can use your data to train the network. Instructions on preparing custom data will be added.

python train_PD.py

PD Gait Score

Using the fine-tuned 3D Pose model you can use the PD_labeling.py to estimate gait scores.

python PD_labeling.py

Citation:

@article{GHOLAMI2023102871,
title = {Automatic labeling of Parkinson’s Disease gait videos with weak supervision},
journal = {Medical Image Analysis},
volume = {89},
pages = {102871},
year = {2023},
issn = {1361-8415},
doi = {https://doi.org/10.1016/j.media.2023.102871},
url = {https://www.sciencedirect.com/science/article/pii/S1361841523001317},
author = {Mohsen Gholami and Rabab Ward and Ravneet Mahal and Maryam Mirian and Kevin Yen and Kye Won Park and Martin J. McKeown and Z. Jane Wang},
}