This project develops an automated system for assessing physical rehabilitation exercises using RGB data.
This project develops an automated system for assessing physical rehabilitation exercises using RGB data. The end-to-end deep learning approaches (i.e. C3D, 3D-ResNet) and the feature extraction based approaches (i.e. LR, MLP, KNN) were implemented. KiMoRe dataset is used for training and testing the models.
- python 3.7
- Clone the repo. Note the colab branch is the main branch.
git clone -b colab https://github.com/claraguoguo/TeleRehab.git
- Install common ML libraries i.e. scipy, pandas, numpy, matplotlib, seaborn, ffmpeg...
- config.cfg: config file for running code locally
- colab_config.cfg: config file for running code on Google Golab
- n_repetition = 5
- n_repetition = 10
- n_repetition = 15
- n_repetition = 10
- n_repetition = 10
- train.py: train and test deep learning models (cnn, resnet, c3d)
- train_LSTM.py: train and test LSTM model
- train_NN.py: train and test MLP and linear regression models
- train_NN_sklearn.py: train and test sklearn models
- train_weighted_loss.py: train and test deep learning models (cnn, resnet, c3d) with weighted loss implementation
python train.py --config config.cfg --model_name c3d
-
Code to extract features from skeletal data can be found in TeleRehab_Utilities
-
(DEPRECATED) Code used to extract skeletal joints with openpose-COCO model can be found at demo_video_KIMORE.py. This code is developed on top of pytorch-openpose.
- openpose
- pytorch-openpose - pytorch implementation of openpose including Body and Hand Pose Estimation (this version works on Apple M1 chip)
Clara Guo - [email protected]