There are a few key steps when understanding how to take an image and produce a latent space-like representation:
- Create an encoder that takes and compresses an image down to an encoding space—the following block diagram describes the process:
A block diagram showing how an encoder produces a compressed representation of an image
- Create a decoder that takes that encoded representation and recreates the image:
A block diagram showing how decoder takes a compressed representation and reproduces an image
- Combine the two models to create an autoencoder and train it:
A block diagram showing how an autoencoder can train both an encoder and decoder at the same time
- Save the encoder model to use for our future chapter development.
See the next recipe to see how we practically build an encoder model.