Deep Learning: Color Brightness Classifier

Simple neural network trying to classify the brightness of the canvas background color. The rgb-values (red/green/blue) of the shown color are fed as inputs into the neural network which then outputs its guess. The first few guesses are more or less random, but you can train the network by clicking either 'dark' or 'bright', depending on the brightness of the displayed color. After around 50 - 100 'training sessions' the results get pretty accurate. If you watch closely, you can see the neural network transform slowly over time (watch the red/blue lines).

The needle shows the neural networks guess and its certainty - the more the needle points to the left/right, the more certain the neural network is with its decision. The lamp is green when the network has guessed correctly and red, when it's been wrong.

Most important source was Daniel Shiffman's Youtube Playlist on neural networks. Also 3Blue1Brown's Youtube Playlist gives valuable information on how neural networks work. The idea for this little project comes from this video.


  • press 'Dark' or 'Bright' to train the neural network
  • R - reset