1. Take one or more images for each direction, by clicking on the areas on the right.
2. Press Train to train the neural net on your images.
3. Press Start to play!
Each frame, the current webcam view goes into to a pretrained neural network model called mobilenet. Mobilenet was trained on millions of images. The output of mobilenet is then given into our neural net (which just trained on your example images). For the final output, our network makes a prediction about which category of your example images, it thinks, it is currently seeing. The snake then moves in the direction linked to the predicted category.
Take multiple, slightly different pictures of the same camera view to improve the recognition accuracy!