GeoScope：无需训练的程序化执行方法突破 VLM 超高分辨率遥感图像计数上限
GeoScope: Training-Free Program Execution beyond the Counting Horizon for Ultra-High-Resolution Remote Sensing Question Answering
遥感图像上数目标这件事，所有开源 VLM 一屏超过约 5 个物体就开始数不准。这篇 arXiv 论文让 LLM 先把问题拆成空间程序，VLM 只在能力范围内数数，不用训练就在 XLRS-Bench 区域计数上提高 16.3 分，超过已训练方法。做遥感分析或多模态评测的人值得看这套实测上限再切视图的思路。
GeoScope 是一个无需训练的超高分辨率遥感图像问答方法，论文实测 8 个开源 VLM 在 DOTA 航空图像上的表现，发现单视图物体数量超过约 5 个后计数准确率急剧下降，且这一计数上限随视图放大、物体变小而进一步收窄。论文用幂律响应模型预测每次视图切分能挽回多少漏计、增加多少误计，据此只在超出计数上限的视图才进行切分。
方法上，GeoScope 用纯文本 LLM 把问题解析为带类型的空间程序，先固定目标区域和分辨率，再让 Qwen3-VL-8B 等 VLM 只在其可操作范围内做定位和计数，方向关系则由定位框显式计算，不依赖 VLM 直接判断。这是首个按实测 VLM 极限来切分视图的超高分辨率方法。
效果上，GeoScope 在 XLRS-Bench 上让 Qwen2-VL-8B 区域计数提升 16.3 分、空间关系提升 10.5 分，区域计数准确率超过所有已训练方法；在 UHR-Micro 上全局和区域计数分别提升 16.8 和 36.2 分，在 MME-RealWorld 计数上提升 11.4 分，并指出其中一个几乎从不正确却常被选中的选项扭曲了该基准的比较结果。
GeoScope: Training-Free Program Execution beyond the Counting Horizon for Ultra-High-Resolution Remote Sensing Question Answering
In ultra-high-resolution (UHR) remote sensing images, the objects that questions count in a region or relate by direction often span only a few pixels. Existing methods mainly decide where a vision-language model (VLM) should look and trust its answer in the chosen view. We show that this assumption fails for counting, and measure VLMs as one calibrates a sensor. All eight open VLMs tested fall increasingly short once a view holds more than a certain number of objects, about five for most on DOTA aerial images. This counting horizon narrows as views grow and objects appear smaller. An empirical power-law response model predicts the undercount a split recovers and the spurious counts it adds, and supports the rule of splitting only views beyond the horizon. In GeoScope, a text-only language model parses the question into a typed spatial program that fixes the region and its resolution. The VLM only grounds and counts within its operating range, and directions are computed explicitly from the grounded boxes. To our knowledge, it is the first UHR method to split views by a measured limit of the VLM. Without training, it improves Qwen3-VL-8B by 16.3 points on regional counting and 10.5 on spatial relations in XLRS-Bench, and has the highest reported accuracy, trained methods included, on regional counting and on the whole benchmark by question. On UHR-Micro it improves global and regional counting by 16.8 and 36.2 points, 15.6 of the former from splitting. On MME-RealWorld counting it gains 11.4 points, and we show that an option that is almost never correct yet often chosen distorts comparisons there. Project page: https://avalon-s.github.io/GeoScope/
本条情报归属 AITOP「论文」核心追踪栏目。在跨区域产业观察坐标中，该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线，反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括：GeoScope、Qwen3-VL、XLRS-Bench、遥感，读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。
信源核验说明：本文原始线索采编自「arXiv cs.AI」（@Guohui Deng, Yuhang Jiang, Chengjun Wang, Zhongwen Hu, Zhaocong Wu, Linsheng Huang, Jinling Zhao, Chao Ruan, Miaozhong Xu, Chi Chen），经 AITOP 多维评分管线结构化提炼。