Answers are grounded in this paper and cite their sources.
The amalgamation of buildings is a key operation in cartographic generalization for producing clear and coherent multiscale maps. Existing learning-based approaches either process rasterized representations, leading to geometric precision loss during raster-to-vector conversion, or employ discriminative models that frame amalgamation as classification or component selection, resulting in topologically inconsistent outputs when matching discrete geometric elements. In this study, we proposed CartoGenT, a generative Transformer framework for amalgamating vector-based building footprints. CartoGenT encodes building groups with coordinate-level spatial features and a scale-adaptive abstraction module to capture contextual dependencies across scales. It then reconstructs amalgamated polygons by jointly predicting vertex coordinates and their connectivity through matching, supporting connectivity-aware vector polygon generation. Experiments using Great Britain vector data showed that CartoGenT achieved an SBR IoU of 0.812, an IoU of 0.754, a TFD of 0.448 radians, and a VCR of 0.901, while producing no null predictions in the main experiment. Results demonstrated that generative modeling enables direct synthesis of connectivity-aware vector geometries, overcoming the precision loss of raster-to-vector conversion and the selection dilemma of discriminative approaches, thereby offering a promising paradigm for cartographic generalization.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: