Game of Tensors repo for the #PoweredByTF2 Challenge. Contains game server, game client (in Unity), mobile client UI, and the code for model training (which includes the dataset).
Very simple Flask server which waits for a socket connection (desktop UI) and then serves endpoints for the mobile clients to push action notifications. It then forwards the actions to the desktop UI.
Fabulous game UI in Unity, which connects to the game server and displays the game.
Simple mobile web page which runs the pre-trained model using TensorFlow.js and notifies the game server whenever a gesture has been recognised.
Model training code
TensorFlow 2 model, including training and test set, to recognise the model. Model’s architecture:
Model performance, which includes for each class: the correlation between actual score and predicted score (top row) and the event timeline, with peaks for the actual and predicted confidence scores (bottom row):
Try out the model
We created a simple web page (hosted on Firebase) to showcase the gesture model. It currently works for Android only, and it’s far from perfect (worked for around 70% of devices we tried). Go on, give it a try at https://tensor-game.firebaseapp.com/gesturedemo.html ! (Use Google Chrome on Android with your volume up)