Learning to dream: Supergan

Super high resolution images (16384×16384 pixels, i.e. 268 Megapixel) generated from two artificial neural networks, recursively upscaling itself.

Both networks trained on the Google Art dataset (conceptual motivations and technical information on the dataset can be read at Learning to see). Due to the recursive nature of the algorithm, the higher resolution images start producing fractal-like effects.

The images on this page are only 4 megapixels (2048×2048), to view the full 268 megapixel (16384×16384) images, click here for an interactive map-like interface.



Created during my PhD at Goldsmiths, funded by the EPSRC UK.