Data-Driven Vision-Language Models for Remote Sensing: A Survey
This survey argues that the central bottleneck for advancing remote sensing vision-language models has shifted from architectural innovation to data curation. It introduces GeoChef, an end-to-end framework that maps dataset taxonomy, construction pipelines, fine-tuning recipes, and evaluation protocols onto a culinary metaphor. By cataloging over 120 datasets and launching a community-driven warehouse, the authors aim to turn small academic teams into proactive data chefs rather than passive consumers of foundation models.
The survey is timely and well-structured, sensibly reframing RS VLM research around data scarcity and curation quality. Its input-output taxonomy and the detailed leakage analysis in evaluation are genuine contributions. However, the work is synthetic by nature, and its heavy reliance on very recent preprints—especially from 2025—risks rapid obsolescence.
Rather than classifying by model capability, the authors organize datasets by input-output form—image captioning, VQA, and visual grounding—which yields a stable taxonomy that does not require被动扩张 when new tasks appear. The practical guidance on semi-automatic construction, the redundancy-check table, and the explicit warning about pre-training and metadata leakage in benchmarks give readers concrete tools they can use immediately.
The survey’s breadth comes at the cost of depth: because it cites an explosion of 2025 preprints, many of its cataloged datasets may be revised or superseded within months, making the compendium fragile. The GeoChef metaphor simplifies the pipeline but also flattens technical nuance; for instance, the discussion of FPFT versus PEFT is largely qualitative and omits wall-clock budgets or convergence curves.
More critically, the evaluation tables stitch together published results from papers that used different base models, hyperparameters, and prompt templates. When the authors note that 'no single model establishes absolute dominance,' they acknowledge inconsistency, yet they do not systematically flag that performance gaps on ultra-high-resolution benchmarks may reflect data contamination or experimental confounders rather than true architectural limits.
As a survey, the evidence is appropriately bibliographic: the authors fairly contrast their data-centric scope with prior model-centric reviews. Nevertheless, by aggregating metrics such as $BLEU$-$4$, $CIDEr$, and $IoU@0.5$ from original papers without re-running controlled experiments, the tables invite readers to compare incomparable settings. The authors implicitly admit this limitation when stating that model performance depends heavily on the alignment between training objectives and test distributions, but they do not provide the standardization needed to turn that observation into a fair horse race.
Reproducibility is partially served by the open-source Awesome-RS-VL-Data warehouse and the richly annotated dataset tables, which lower the barrier to finding and citing RS vision-language data. However, the paper does not release unified training code, standardized prompts, or the exact hyperparameters for the fine-tuning strategies it surveys. Without these, independent groups cannot replicate the aggregated benchmark numbers or directly validate the recommended SFT and RFT recipes, leaving the GeoChef pipeline as a conceptual guide rather than a reproducible protocol.
During the deep learning era, innovations in remote sensing (RS) vision models primarily focused on optimizing network architectures for specific tasks and conducting end-to-end training. However, with the advent of the Vision-Language Models (VLMs) era, constrained by substantial computational resource demands, the mainstream research paradigm has shifted towards adapting general-purpose VLMs using RS 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 researchers in constructing high-quality RS 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 RS datasets tailored for VLMs. It further provides a comprehensive practical guide for RS 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 higher-quality vision-language datasets for specialized RS sub-fields. Ultimately, the goal is to converge these efforts into a comprehensive RS 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 RS.
Pick a starting point or write your own. Challenges run in the background, so you can keep reading while the AI investigates.
No challenges yet. Disagree with the review? Ask the AI to revisit a specific claim.