GeoLLaVA-8K: Scaling Remote-Sensing Multimodal Large Language Models to 8K Resolution

Di Wang (Wuhan University) · Bo Du (Wuhan University) · Jing Zhang (The University of Sydney) · Wenjing Yang (National University of Defense Technology) · Fengxiang Wang (National University of Defense Technology) · Yulin Wang (Tsinghua University, Tsinghua University) · Mingshuo Chen (Beijing University of Posts and Telecommunications) · Haotian Wang (National University of Defense Technology) · Hongzhen Wang (Tsinghua University, Tsinghua University) · Long Lan (National University of Defense Technology, Tsinghua University) · Yueying Li (National University of Defense Technology) · Zonghao Guo (Tsinghua University) · Zefan Wang (Tsinghua University) · Shan Boqi (Beijing University of Aeronautics and Astronautics)

anchored token selectionbackground token pruningdata scarcitygeollava-8khighrs-vqakey semanticsmemory footprint multimodal foundation modelsobject-centric tokensremote sensingsuperrs-vqatoken explosionultra-high-resolutionvision-language datasetsxlrs-bench

Ultra-high-resolution (UHR) remote sensing (RS) imagery offers valuable data for Earth observation but pose challenges for existing multimodal foundation models due to two key bottlenecks: (1) limited availability of UHR training data, and (2) token explosion caused by the large image size. To address data scarcity, we introduce **SuperRS-VQA** (avg. 8,376$\times$8,376) and **HighRS-VQA** (avg. 2,000$\times$1,912), the highest-resolution vision-language datasets in RS to date, covering 22 real-world dialogue tasks. To mitigate token explosion, our pilot studies reveal significant redundancy in RS images: crucial information is concentrated in a small subset of object-centric tokens, while pruning background tokens (e.g., ocean or forest) can even improve performance. Motivated by these findings, we propose two strategies: *Background Token Pruning* and *Anchored Token Selection*, to reduce the memory footprint while preserving key semantics. Integrating these techniques, we introduce **GeoLLaVA-8K**, the first RS-focused multimodal large language model capable of handling inputs up to 8K$\times$8K resolution, built on the LLaVA framework. Trained on SuperRS-VQA and HighRS-VQA, GeoLLaVA-8K sets a new state-of-the-art on the XLRS-Bench. Datasets and code were released at https://github.com/MiliLab/GeoLLaVA-8K.