Abstract
Some automated map generation systems may produce geometrically inconsistent outputs from visual data as they can lack an inherent geometric grammar. This can be addressed by reframing cartography as a structured language modeling problem. A specialized polylines language may be used to represent geographic features as discrete token sequences. A multi-modal model can be trained to autoregressively generate these sequences from inputs, such as satellite imagery, effectively translating visual data into a geometrically structured format. This process can embed geometric and topological logic into the generation, which may improve the likelihood of producing consistent map data, inferring occluded features, and enabling dynamic correction through text prompts.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Kligvasser, Idan; Intrator, Yotam; Rivlin, Ehud; and Cohen, Regev, "Vector Map Generation and Correction Using a Geometric Language Modeling Approach", Technical Disclosure Commons, (March 09, 2026)
https://www.tdcommons.org/dpubs_series/9476