GeoChef: A Data-Centric Guide to Tailoring Vision-Language Models for Remote Sensing

During the data-driven AI era, innovations in remote sensing vision models primarily focused on optimizing network architectures for specific tasks and conducting end-toend training. However, with the advent of the vision-language models (VLMs), constrained by substantial computational resource demands, the mainstream research paradigm has shifted towards adapting general-purpose VLMs using remote sensing vertical domain data for fine-tuning, rather than modifying their core architectures. This transition marks a change in the innovation pathway from model-centric to data-centric. Despite the increasing importance of data, the academic community still lacks a systematic, data-centric survey to guide researcheremote in constructing high-quality remote sensing vision-language datasets for various application scenarios and in conducting effective model fine-tuning and performance evaluation based on these datasets. To address this gap, this paper presents a systematic review of the development status of remote sensing datasets tailored for VLMs. It further provides a comprehensive practical guide for remote sensing researchers in the large model era, covering the entire pipeline from data construction to fine-tuning and evaluation. This paper aims to encourage the academic community to build more diverse and higherquality vision-language datasets for specialized remote sensing sub-fields. Ultimately, the goal is to converge these efforts into a comprehensive remote sensing vision-language corpus that offers extensive task coverage and a wide geographical distribution, thereby laying a solid data foundation for advancing towards general artificial intelligence in remote sensing. The project will be released at https://github.com/zytx121/Awesome-RS-VL-Data