Relative Pose Constraint (2024)
Concept
“It is a kind of naiveté and arrogance to think that this thing stays forever, for eternity. All these projects have this strong dimension of missing, of self-effacement… they will go away, like our childhood, our life. They create a tremendous intensity when they are there for a few days” – The artist Christo commenting on the transience of the canonical ‘Wrapped Reichstag’ installation.
My piece interrogates the fundamental premise of Christo’s quote; that ‘Wrapped Reichstag’, and by extension any event which seems fundamentally tied to a particular geography, time, and social context, is not functionally eternal. By retrieving pictures from the internet, and then recreating the nearly 30 year old installation in 3D space from only the few available views using a neural radiance field, I hoped to question the way monumental events are preserved, recreated, and recontextualized constantly by digital tools and digital access paradigms like search engines and social media, particularly when this access is mediated by deep learning. Furthermore, the political and social context of ‘Wrapped Reichstag’, specifically its significance in the reunification of Germany through the visually symbolic rebirth of the German people as a new entity with a new relationship to their own past, is intriguing in our contemporary context when these same digital tools seem to more clearly demarcate the dividing line between two opposing views of what the West is, how we should view our history, and what defines a nation. Aesthetically, I tried to bias the representation learned by the NeRF to suggest the blurred and colorfully expressive squeegee work of Gerhard Richter, whose own oeuvre is intimately tied to the national divide and eventual reunification of the German state, and the idea of shared historical perspective (particularly his paintings ‘November 1989’ and ‘Betty’ which bracket the fall of the Berlin wall). By retrieving Christo and Jeanne-Claude’s installation from the annals of history and forcing its reconsideration in an alien digital format, this piece also examines the way we as a culture retrieve historical events and wield them like objects, outside of their context, while using them to justify our competing narratives of who we are.
Techniques
This piece was made using a combination of recently developed deep learning techniques. The base image dataset was scraped from google images and flickr, before being curated and preprocessed using the cv2 and PIL python libraries in various ways to create saturation distortions, false colorations, and more general glitch-like artifacts to help guide the generation towards an aesthetic in conversation with some of the prior mentioned artists. The processed images are then passed to a custom neural radiance field designed to work without camera pose priors which I extended from NoPE NeRF (CVPR 2023). After several iterations of training and tweaking the NeRF, I sampled a collection of novel view images by traversing the implied 3D scene. I then interpolated between the generated images using the pretrained FILM neural network (ECCV 2022), and converted these interpolations to video.
Thank you for viewing my work!
More about me: http://www.alexrusnak.com/