Textual Supervision Enhances Geospatial Representations in Vision-Language Models

Marcelo Sartori Locatelli, Fernando Tonucci, Jea Kwon, Luiz Felipe Vecchietti, Bryan Nathanael Wijaya, Cheng Yaw Low, Virgilio Almeida, Meeyoung Cha
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:80447-80475, 2026.

Abstract

Geospatial understanding is a critical yet underexplored dimension in the development of machine learning systems for tasks such as image geolocation and spatial reasoning. In this work, we analyze the geospatial representations acquired by three model families: vision-only architectures (e.g., ViT), vision-language models (e.g., CLIP), and large-scale multimodal foundation models (e.g., LLaVA, Qwen, and Gemma). By evaluating across image clusters, including people, landmarks, and everyday objects, grouped based on the degree of localizability, we reveal systematic gaps in spatial accuracy and show that textual supervision enhances the learning of geospatial representations. Our findings suggest the role of language as an effective complementary modality for encoding spatial context and multimodal learning as a key direction for advancing geospatial AI.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-locatelli26a, title = {Textual Supervision Enhances Geospatial Representations in Vision-Language Models}, author = {Locatelli, Marcelo Sartori and Tonucci, Fernando and Kwon, Jea and Vecchietti, Luiz Felipe and Wijaya, Bryan Nathanael and Low, Cheng Yaw and Almeida, Virgilio and Cha, Meeyoung}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {80447--80475}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/locatelli26a/locatelli26a.pdf}, url = {https://proceedings.mlr.press/v306/locatelli26a.html}, abstract = {Geospatial understanding is a critical yet underexplored dimension in the development of machine learning systems for tasks such as image geolocation and spatial reasoning. In this work, we analyze the geospatial representations acquired by three model families: vision-only architectures (e.g., ViT), vision-language models (e.g., CLIP), and large-scale multimodal foundation models (e.g., LLaVA, Qwen, and Gemma). By evaluating across image clusters, including people, landmarks, and everyday objects, grouped based on the degree of localizability, we reveal systematic gaps in spatial accuracy and show that textual supervision enhances the learning of geospatial representations. Our findings suggest the role of language as an effective complementary modality for encoding spatial context and multimodal learning as a key direction for advancing geospatial AI.} }
Endnote
%0 Conference Paper %T Textual Supervision Enhances Geospatial Representations in Vision-Language Models %A Marcelo Sartori Locatelli %A Fernando Tonucci %A Jea Kwon %A Luiz Felipe Vecchietti %A Bryan Nathanael Wijaya %A Cheng Yaw Low %A Virgilio Almeida %A Meeyoung Cha %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-locatelli26a %I PMLR %P 80447--80475 %U https://proceedings.mlr.press/v306/locatelli26a.html %V 306 %X Geospatial understanding is a critical yet underexplored dimension in the development of machine learning systems for tasks such as image geolocation and spatial reasoning. In this work, we analyze the geospatial representations acquired by three model families: vision-only architectures (e.g., ViT), vision-language models (e.g., CLIP), and large-scale multimodal foundation models (e.g., LLaVA, Qwen, and Gemma). By evaluating across image clusters, including people, landmarks, and everyday objects, grouped based on the degree of localizability, we reveal systematic gaps in spatial accuracy and show that textual supervision enhances the learning of geospatial representations. Our findings suggest the role of language as an effective complementary modality for encoding spatial context and multimodal learning as a key direction for advancing geospatial AI.
APA
Locatelli, M.S., Tonucci, F., Kwon, J., Vecchietti, L.F., Wijaya, B.N., Low, C.Y., Almeida, V. & Cha, M.. (2026). Textual Supervision Enhances Geospatial Representations in Vision-Language Models. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:80447-80475 Available from https://proceedings.mlr.press/v306/locatelli26a.html.

Related Material