YouTube TrAIner uses Google’s TensorFlow platform and machine learning model called PoseNet to turn YouTube’s huge library of content into a whole new world of training videos.
PoseNet allows for real-time human pose estimation in a browser and through either a webcam or smartphone camera. In this case a basic webcam camera and laptop powered the experience allowing us to see how closely people were able to replicate a cricket shot by analyzing their stance, movement and swing and then comparing it to the IPL video they were copying. Through this comparison and their visible PoseNet skeleton we were able to calculate their accuracy score to help them improve their shots.
Off the back of this we also developed a Web App for the 2019 Indian Premier League season. Star Sports and Google developed a similar A.I. powered activation that toured the country with the IPL trophy.
Fans could compete against the IPL heroes and compare their posture, form and movements as they tried to copy their hero.
Codroit# Posenet Model