This repository is composed of the Code (Neural Transfer Function.ipynb), the training data scraped from Google, and citations for the Duke AI Art Competition.
America the Beautiful: National Parks + Modern Art with a Neural Transfer Model
In this series, I wanted to explore the juxtaposition between ancient and modern, between the wild and the calculated, between natural and artificial. This prompts the re-discovery of the natural beauty of the national parks with artificial intelligence. The national parks are a blackbox of nature- much of their land mass untouched by humans. This is not dissimilar to the blackbox of an artificial neural network. The intrigue of both national parks and neural networks pairs nicely to provide a unique perspective into the artifical redefining the natural.
For this series, I developed a neural transfer model using a VGG19 CNN model trained on Imagenet data and layered randomized content and style layers scraped from the web with keywords "national park" and "modern art", respectively. This melded the content of the national parks with the style of modern art. Drawing from paintings, sculptures, and other mediums, the art created by this model is starkly modern while also being surprisingly familiar. We see common trends from modern art including cubism, expressionism, and a touch of surrealism.