Results of the custom-built 2D fractal brownian motion generator which can generate ridged and turbulence noise were interpreted as red, green and blue color values. Different parameter combinations allow movement in the value landscape which yields several different and interesting patterns.
The outcome of these rgb fBM interpretations is fed into a deep dream neural network originally developed for image classification, which is used to amplify a certain structure in a neuron of the network’s layers in the image. This amplification step creates intricate sub patterns on the numerical landscape of the fBM generator and results in often unforeseen transformations.
They are organized into six series of nine images with 3200*3200 pixels each. These groups were selected by mapping thousands of possible generations into a 2d embedding using uniform manifold approximation. This allows the selection of similarly structured images within the huge possibility space.
