[tiger10guy]

I'm pretty sure there's not an autoencoder involved, it just looks like a vanilla conv net.

This is the implementation: http://torontodeeplearning.github.io/convnet/