Perception Engine is an ongoing project exploring neural network's abilities to create abstract representations of real-world items by reverse engineering classifying systems. Each piece from the project starts with the system generating a random geometric composition of shapes and lines that are then fed into a classifying neural network to rearrange the components to represent a real-world item. This reverse engineering of systems allows for a unique and insightful look into how machines develop their own algorithmic view of our world.
Bee, from the Perception Engines 2 Series, 2021.Digital master created by neural networks based on the ImageNet category "be". The boot-up sequence reveals the process by which computer perception manipulates shapes and colors to create the final printable layers.