Domain adaptation using transformer models for automated detection of exterior cladding materials in street view images

Description

Wang teaches a model to name the outer wall material of buildings from street photos in twenty twenty-five. City surveys need wall types for energy and safety work, but labeling photos by hand is slow. Fully labeled sets from Scotland and partly labeled ones from London trained the system. The model had to adapt across the two places, a task called domain adaptation.

Six wall types were targeted, from brick and stone to glass and mixed. The best model, an axial transformer fed extra varied training images, scored about eighty-two in a hundred overall. Stone and brick scored highest, glass lowest. The author says the method can scale wall surveys, with more polish to come.

Catalog record

Source: https://europepmc.org/articles/PMC12824198