Computer Vision Art Gallery

ICCV 2019, Seoul, Korea

Daniel Ambrosi

Japanese Tea Garden - Grand Scale Dreamscape Detail, (2019)


My initial experiments in computational photography were driven by my desire to create photo-based depictions of the world that better convey the feeling of a place and the way we really experience it: not just visually, but also viscerally and cognitively. My grand format landscape images that result from these experiments are inspired by the 19th century master paintings of the Hudson River School and by the great romantic European landscape paintings that preceded them. Like those works, some of which reached ten feet in width, I’ve endeavored to create uncannily immersive and idyllic scenic experiences that deliver both breadth and detail. Now, capitalizing on recent technological developments in deep learning and artificial intelligence, along with the hard work and ingenuity of my generous engineering colleagues, Joseph Smarr (Google) and Chris Lamb (NVIDIA), I’ve been able to push my artwork in an intriguing new direction. Thanks to custom enhancements made to Google’s DeepDream software, it has become possible for me to imbue my giant landscape images with a stunning degree of unexpected form and content. My latest work submitted here further investigates the possibilities of human-AI hybrid art by exploring multi-level “dreaming” effects. These works begin with a detail of one of my panoramic images to which multiple styles of neural network “hallucination” are applied at multiple scales. Unlike my previous “full scene” Dreamscapes, which appear completely photographic from a distance, these large works are clearly hallucinatory upon first impression, yet still contain intricate details that defy expectations.