Smallest CNN possible

Smallest CNN taking 3x3 pictures as inputs and able to recognize 4 classes

This very simple network shows an example of a CNN learning to distinguishv between 4 simple classes. The input image is only 3x3 pixels and the convolutional filter is 2x2 pixels. The network is composed of one convolutional layer and 1 fully connected layer. The small size of the network helps reducing computation time so as to confirm functionality. I've added comments to explain the basics of how it works, if that can help anyone.