About the recognition of push up

[crapper]

Hello, I have currently read the article from aicurious and I am wondering how this application is using MobileNetV2 to classify the bone signal in a binary questions of "is the input frame push up". I can only saw you have briefly explained you have used it for it but not further explanation. May I have a more detailed information for the recognition on pushup? Are you doing the recognition by collecting all possible bones signal of human activities through feeding the MPII dataset to MobileNetV2?
Thank you.