A Generative Adversarial Network Framework for Vertiport Location: A Case Study in Toronto

The accessibility of building function is essential for urban research. We reviewed the recent work and concluded three limitations: few open-source datasets, coarse building function categories, and poor model interpretability with inadequate multimodal feature fusion. Thus, we created BuildingSense with fine-grained categories and multimodal data, and proved that the large model can be used for improving the interpretability of results, with three directions for enhancing their performance.