This is the code for the paper Rahil Mehrizi et. al., Automatic Health Problem Detection from Gait Videos Using Deep Neural Networks, https://arxiv.org/abs/1906.01480v1

Python 2.7
Pytorch v0.4.0 or Pytorch v0.1.12

1- Watch our video: https://drive.google.com/file/d/11wHq3zmBUM-z-OWonfXmMqLv8lHVzL5T/view?usp=sharing
2- Clone this repository and get the dataset. We provide the Human3.6M dataset in ground truth 3d points, camera rotation matrix, and estimated 2d points from Stacked Hourglass network alonge with corresponding confidence extracted from heatmaps.
3- Edit genlist4view.py to set path-to-data and generate training/validation lists based on the data location.

git clone https://github.com/rmehrizi/multi-view-pose-estimation.git
cd multi-view-pose-estimation
wget https://drive.google.com/file/d/1CQJGSCDxRLiPEw-azTjcV7h6bx6I1xwI/view?usp=sharing
unzip dataset.zip
rm dataset.zip
python genlist4view.py

1- For a quick demo, edit train4view.py to set path-to-data and change "demo_mode" to "True" and visualize the results on randomly selested images

cd multi-view-pose-estimation
python train4view.py

1- Edit train4view.py to set path-to-data and change "validation_mode" to "True" and check tracking results in multi-view-pose-estimation/result/.

cd multi-view-pose-estimation
python train4view.py

Training is performed in two steps:

1- Edit genlist.py to set path-to-data and generate training/validation lists for single view images.
2- Edit train.py to set path-to-data and run it to train the model (2 blocks) and save the best model in exp/single-view

cd multi-view-pose-estimation
python genlist.py
python train.py

1- Edit options/train-options to set "lr" to 0.0001 and "nEpochs" to 5 .
2- Edit train4view.py to set path-to-data and change both "demo_mode" and "validation_mode" to "False".
3- Run train4view.py to train the model (4 blocks) and save the best model in exp/multi-view

cd multi-view-pose-estimation
python train4view.py

If you find this code useful in your research, please consider citing our work: https://arxiv.org/vc/arxiv/papers/1906/1906.01480v1.pdf