In any sports such as Tennis or badminton, one of the main ways for sportsmen to improve their skills is through constant feedback and support from their coaches, which is currently unmatched by any other means imaginable. This led us to come up with an innovative way to provide the feedback and
What it does
The model helps users become better sports players by providing feedback on their recorded videos and simulating alternative reality where they could have done better.
How we built it
Step1: We preprocessed the videos by converting them into frames and manually annotating them by indicating the time positions where the player could have done better. Step2: We used pre-trained weights from a pose estimation algorithm to identify the players on the field/frame. Step3: We built two LSTM models, one of which acted as a predictor of the timestep where the player made a mistake and the other of which acted as a generative model simulating frames for alternate reality which could have led to the player winning.
Challenges we ran into
Annotating and creating a huge amount of datasets for the machine learning model to learn a basic function representation of the model.
Accomplishments that we're proud of
We built a basic working model which sufficiently puts forth the idea mentioned.
What we learned
Handling huge amounts of data and combing outputs from different models.
What's next for Coach.IO
We believe this idea can act as a major hallmark in the way people learn sports. So, we intend to take this project forward by using the model we built at the hackathon as the baseline for a hopefull amazing future project.