Implementation of Deep Dream and Neural Style Transfer Algorithm using Python

Abstract

Deep Dream is an artistic algorithm where a pretrained CNN feds an image and optimizes it to amplify the features it "sees" in the image. Depending on the neural network layer, the features amplified will either be low-level (like edges, certain geometric patterns, etc.) or high-level (like dog snouts, eyes, etc.), which heavily depends on the dataset on which the net was pretrained! As a result, mimicking phenomenological features of altered states without these other more widespread consequences offers an essential experimental tool for research into consciousness and psychiatry. Here, we discuss this device, which we refer to as the hallucination machine. Deep convolutional neural networks (DCNNs) and panoramic footage of natural surroundings, watched immersively through a head-mounted display, make up the innovative combination (panoramic VR). Neural style transfer is a technique in computer vision that enables the creation of artistic images by combining the content of one image with the style of another.