[FG 2024] Benchmarking Skeleton-based Motion Encoder Models for Clinical Applications

Introduction

This project is created as part of the research for the paper titled "Benchmarking Skeleton-based Motion Encoder Models for Clinical Applications: Estimating Parkinson’s Disease Severity in Walking Sequences" accepted at IEEE international conference on automatic face & gesture recognition (FG 2024).

Installation

git clone https://github.com/TaatiTeam/MotionEncoders_parkinsonism_benchmark.git
cd MotionEncoders_parkinsonism_benchmark
pip install -r requirements.txt

Data

Dataloaders will be added soon.

Demo

Demo will be added soon.

Leaderboard

Model	F1 Score	Paper/Source
MixSTE	0.41	Link
MotionAGFormer	0.42	Link
MotionBERT-LITE	0.43	Link
POTR	0.46	Link
MotionBERT	0.47	Link
PD STGCN	0.48	Link
PoseFormerV2	0.59	Link
PoseFormerV2-Finetuned	0.62	Link

For detailed rankings, visit the Paperswithcode Leaderboard.

Acknowledgement

Special thanks to the creators of the dataset for making their clinical data publicly available:

• A public data set of walking full-body

Our code also refers to the following repositories. We thank the authors for releasing their codes.

• PoseFormerV2
• MotionBERT
• MixSTE
• POTR
• MotionAGFormer
• stgcn_parkinsonism_prediction

Citation

Please cite our paper if this library helps your research:

@inproceedings{PDmotionBenchmark2024,
  title     =   {Benchmarking Skeleton-based Motion Encoder Models for Clinical Applications: Estimating Parkinson’s Disease Severity in Walking Sequences}, 
  author    =   {Vida Adeli, Soroush Mehraban, Yasamin Zarghami, Irene Ballester, Andrea Sabo, Andrea Iaboni, Babak Taati},
  booktitle =   {2024 18th IEEE international conference on automatic face & gesture recognition (FG 2024)},
  year      =   {2024}
}