TensorFlow 1.x Deep Learning Cookbook

How it works...

Neural networks store abstractions of the training images: lower layers memorize features such as lines and edges, while higher layers memorize more sophisticated images features such as eyes, faces, and noses. By applying a gradient ascent process, we maximize the loss function and facilitate the discovery of a content image of patterns similar to the ones memorized by higher layers. This results into a dreaming where the network sees trippy images.