Requires python 3.6+
Install dependencies
pip install -r requirements.txt
To run on a single video:
python predict.py --from_video <video-path>
To run on a directory of images
python predict.py --from_dir <video-dir-path>
After running inference, there should be two additional files for each input video. One file *_landmarks.csv contains the decoded pose landmarks stored in a csv file. The other file *_network_output.mp4 contains the original video annotated with the landmark information.
NOTE: Inference should be fast and use very little memory, so you should easily be able to run this on your laptop