sense

Badminton Shot Recognition

Everyone has played badminton at least once in their lifetime. To play, we need to learn different shots and for that we take help from coaches, tournament videos, books and last we do practice every day.

But, after lots of practice, if you want to know that how perfect are your shots or what shot that international player played in the video, how will you do that?

So, at that time, this application Badminton Shot Recognition can be used, which will provide realtime feedback by recognizing your badminton shots.

Model Training

The model for badminton shot recognition was trained on the model StridedInflatedEfficientNet, by fine-tuning 17 layers of the model. To get the weights for the badminton shot recognition model click here.

Dataset Breakdown

There are 17 different classes (badminton shots + background activity + no racket actions) on which this model is trained using transfer learning on top of the weights provided by sense repository.

Number of total videos: 278
Duration of videos: 4 to 5 seconds
Dataset distribution: (approx) 70% train, 30% valid (Each class)

To download, the dataset click here.

Examples of classes

Forehand underhead shot	Backhand underhead shot	Forward backhand shot	Forward forehand shot
Backhand backcourt shot	Forehand backcourt shot	Forehand serve	Backhand serve	Backhand side shot
Forehand side shot	Forehand overhead shot	Backhand overhead shot	With racket	Without racket

Output