Authors: Patrick Tourniaire & Justin Regef
This is a project for automatically detecting if a person in a video is performing a squat or a pull-up somewhere in the video. Which is used to then give local feedback on the execution of the exercise(s) performed.
First you have to install the poetry dependencies and implicitly setup a venv with those deps.
$ pip install poetry
$ poetry installThen you have to set8up MMPose to be able to load and infer on the pose estimation models.
$ poetry run mim install mmengine "mmcv>=2.0.1" "mmdet>=3.1.0" "mmpose>=1.1.0"After running the above setup, you should be able to simply provide an input video to the main.py file using the --input_video argument.
Note: the processing time for extracting the pose skeletons can take a long time, therefore, we also have some example data of us performing exercises under data/examples/* which has cached pickle files for the computed skeletons.
$ poetry run python main.py --input_video <path_to_video>If you want to run based on a cached sequence of skeletons then you can simply run this command, and refer to a specific pickle file which you want to test.
$ poetry run python main.py --input_video cache/<video_name>.pickle # Without the _kpts_hands or _kpts_joints endingIf you experience any issues when running this project, feel free to contact us:
Justin Regef: justin.regef@polytechnique.edu
Patrick Tourniaire: patrick.tourniaire@polytechnique.edu
This project relies on the pypipeline package which is to publicly available yet but was developed by Patrick Tourniaire, and is available in the deps/ folder as a wheel file. However, running poetry install will take care of this, but if you decide not to use poetry then it is important to also install this package.