how convolutional autoencoder learns multiple class? (ex digit data)

Question

I know how CNN autoencoder works, but suddenly I feel weird. Digit data has 10 class, it means that autoencoder can be learned not only binary but multiple classes. However, I think autoencoder only can be learned one class... Is there anybody to explain this? :)

cnn autoencoder example(digit data) : https://blog.keras.io/building-autoencoders-in-keras.html

Answer 1

In this example, the classes in the data set are not relevant. The autoencoder simply attempts to make the output image as similar to the input image as possible.