[ARIAL@IJCAI2025] Python code and dataset repository for paper "Rehabilitation Exercise Quality Assessment and Feedback Generation Using Large Language Models with Prompt Engineering"
[ArXiv Paper], [Springer Nature]
Pipeline Diagram:
create a virtual env to install the requirements:
pip install -r requirements.txtExerciseLLM is a categorized and described rehabilitation exercise movement dataset that originates from UI-PRMD and REHAB24-6 dataset, which can be downloaded directly from the linked sites.
For UI-PRMD, only the segmented files from Kinect are used. For REHAB24-6, only the 2d_joints_segmented data are used. A zip file containing 2d_joints_segmented data is included in the repository.
Ensure the downloaded dataset follows the format below:
./dataset/
├── REHAB24-6
├── 2d_joints_segmented
├── Ex1-segmented
├── PM_000_c18_rightarm-1-rep1-1.npy
├── PM_000_c18_rightarm-1-rep2-1.npy
├── ...
└── PM_122_c18_rightarm-7-rep10-0.npy
├── Ex2-segmented/
...
└── Ex6-segmented/
└── annotations.csv
└── UI-PRMD
├── correct
├── kinect
├── angles/
├── positions
├── m01_s01_e01_positions.txt
├── m01_s01_e02_positions.txt
├── ...
└── m10_s10_e10_positions.txt
└── incorrect/
Run the following Python scripts to generate the exercise-specific features data.
UI-PRMD
python generate_UIPRMD.py
This will generate a 'features' and 'absolutes' folder under both `UI-PRMD/correct' and 'UI-PRMD/incorrect'
REHAB24-6
python generate_REHAB24-6.py
The final data directory tree follows:
./dataset/
├── REHAB24-6
├── 2d_joints_segmented
├── Ex1-segmented/
├── Ex2-segmented/
...
└── Ex6-segmented/
├── features
├── Ex1-segmented/
├── PM_000_c18_rightarm-1-rep1-1_features.csv
...
└──PM_122_c18_rightarm-7-rep10-0_features.csv
├── Ex2-segmented/
...
└── Ex6-segmented/
└── annotations.csv
└── UI-PRMD
├── correct
├── kinect
├── absolutes
├── m01_s01_e01_absolutes.csv
├── m01_s01_e02_absolutes.csv
├── ...
└── m10_s10_e10_absolutes.csv
├── angles/
├── positions/
└── features
├── m01_s01_e01_features.csv
├── m01_s01_e02_features.csv
├── ...
└── m10_s10_e10_features.csv
└── incorrect/
Visualize UI-PRMD movements by running UI-PRMD_visualization.ipynb
This research was funded by the New Frontiers in Research Fund, Canada, and the TRANSFORM HF Undergraduate Summer Research Program, Canada.
Contributors are Jessica Tang, Ali Abedi, Tracey Colella, Shehroz Khan at KITE Research Institute.