Self-supervised learning unveils urban change from street-level images

Steven Stalder, Michele Volpi, Nicolas Büttner, Stephen Law, Kenneth Harttgen, Esra Suel

Abstract: Highlights•First application of unsupervised and self-supervised methods to measure neighborhood change from street-level images.•We adapt the Barlow Twins strategy to learn representations from street-level images, embedding features of urban structures.•We develop a method to detect relevant changes in urban structures by comparing learned features at different time points.•We demonstrate the feasibility of our method for tracking neighborhood-level urban change in London between 2008 and 2018.