A figure skating jump recognition software using IMU data as input.
CLI usage:
pip install requirements.txt
python3 mai.py -p <session name> | -t
-p is used for the dataset generation. -t is used for training.
A fastAPI is available inside the api/api.py file. It only has one endpoint.
Add a new dataset in constants.sessions. 'path' is the path were all csvs of the session are. IMU acquisition has to be synchronized.
'sample_time_fine_synchro' corresponds to the sample_time_fine synchro where video begins (so your annotation ready data is in sync with the video).
Current model has been trained with a dataset of roughly 500 annodated jumps. Because of privacy concerns, the dataset is not disclosed.
Made by the S2M for Patinage Quebec.