Bridging Geographic Bias in Urban Streetscape Inference via Lifelong Learning with Visual-Semantic Pivoting

The article presents HVSP-LL, a lifelong learning framework designed to mitigate geographic bias in urban streetscape inference by integrating a visual-semantic pivoting module with an equity-aware rehearsal mechanism. It utilizes a three-tier ontology to organize landscape concepts and achieves a Spearman correlation of 0.834 on a benchmark from twelve cities, marking a 6.1 point improvement over existing continual learning baselines while reducing inter-city perception gaps by 38%. This framework is significant for practitioners as it enhances the robustness of urban perception models, enabling more equitable landscape planning and decision-making across diverse urban environments.