Abstract
Static machine learning models applied in automated cartography may produce inaccurate map features or may not adapt to localized guidelines without potentially costly retraining. A post-processing framework can utilize a multi-modal, text-guided generative model to perform adaptive corrections. For example, a system can receive a visual representation of a map segment, such as a raster image of road markings, and a set of natural language guidelines as input. The generative model can then process these inputs to synthesize a new visual representation where map features may be altered to conform with the specified guidelines. This approach can decouple correction logic from initial map generation, which may enable flexible updates to digital map data while potentially reducing the need for model retraining or extensive manual validation.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Intrator, Yotam; Kligvasser, Idan; Rivlin, Ehud; and Livne, Amir, "Adaptive Correction of Digital Map Features Using a Text-Guided Generative Model", Technical Disclosure Commons, (January 29, 2026)
https://www.tdcommons.org/dpubs_series/9238