Encoder decoder architectures

A common approach to extracting spatial features from images are so-called encoder-decoder architectures.

While classification networks aim to take the information in an image and compress it down to a perhaps a single number that represents an image class, other tasks such as image segmentation instead look to label every pixel in the image in some way.

A common way of achieving this via deep learning is through the use of encoder-decoder networks, as we will see in this video.