This collection is an exploration of AI-based vision systems with the goal of giving the viewer a sense of [pareidolia](https://en.wikipedia.org/wiki/Pareidolia). The code used to generate these dreams was created through my research into visual perception. By exploring these networks, I hope to gain an understanding of our own visual perception systems. Each piece is unique 1/1. There will only ever be 36 unique pieces created.

[Deep dream](https://en.wikipedia.org/wiki/DeepDream) is one of the earliest techniques that was used to interpret what deep convolutional neural networks trained for computer vision tasks "dream" of. Additionally, there has been [ample](https://www.nature.com/articles/s42003-018-0110-y) recent [research](https://pubmed.ncbi.nlm.nih.gov/31260392/) suggesting these models do learn to ["perceive"](https://arxiv.org/ftp/arxiv/papers/2001/2001.07092.pdf) the world similarly to our own visual system. This collection is largely based on such research.

Deep Pareidolia

Even the most mundane visuals have such a deep underlying complexity that can be revealed through gradient ascent 🌼💐🌸🌺🏵️💮🌷

Revealing the Complexity

What is the value?

Details