When Large Vision-Language Model Meets Large Remote Sensing Imagery: Coarse-to-Fine Text-Guided Token Pruning
Junwei Luo, Yingying Zhang, Xue Yang, Kang Wu, Qi Zhu, Lei Liang, Jingdong Chen, Yansheng Li
2025-03-13
Summary
This paper talks about a smart way to help AI understand huge satellite images by focusing only on important parts, guided by text instructions, to save time and computing power.
What's the problem?
AI models either miss details in massive satellite images or use too much computer power to analyze them fully, making it hard to answer questions accurately.
What's the solution?
The method uses text clues to find key areas in the image, then breaks the image into smaller pieces and focuses on those areas step-by-step, skipping less important parts.
Why it matters?
This helps analyze satellite images faster for tasks like disaster response or environmental monitoring, making AI tools more practical for real-world use.
Abstract
Efficient vision-language understanding of large Remote Sensing Images (RSIs) is meaningful but challenging. Current Large Vision-Language Models (LVLMs) typically employ limited pre-defined grids to process images, leading to information loss when handling gigapixel RSIs. Conversely, using unlimited grids significantly increases computational costs. To preserve image details while reducing computational complexity, we propose a text-guided token pruning method with Dynamic Image Pyramid (DIP) integration. Our method introduces: (i) a Region Focus Module (RFM) that leverages text-aware region localization capability to identify critical vision tokens, and (ii) a coarse-to-fine image tile selection and vision token pruning strategy based on DIP, which is guided by RFM outputs and avoids directly processing the entire large imagery. Additionally, existing benchmarks for evaluating LVLMs' perception ability on large RSI suffer from limited question diversity and constrained image sizes. We construct a new benchmark named LRS-VQA, which contains 7,333 QA pairs across 8 categories, with image length up to 27,328 pixels. Our method outperforms existing high-resolution strategies on four datasets using the same data. Moreover, compared to existing token reduction methods, our approach demonstrates higher efficiency under high-resolution settings. Dataset and code are in https://github.com/VisionXLab/LRS-VQA.