Fig. 9: Methodology overview.

First, we take a subset of image tiles representative of the urban forms present in the study area to be used in the data collection. We then collect pairwise comparisons from this subset of image tiles through local community workshops. The collected pairwise comparisons are used to train an AI-voter to make pairwise comparisons of other unseen images from the rest of the study area. These are then used to calculate perceived urban livability scores. This is done for each user group (DUA residents, planners, combined). The AI-voter selection behavior is compared to its respective user group behavior vis-à-vis physical features like green space and density.