WeaveEarth: Structured Evidence Construction and Reasoning for Training-Free UHR Remote Sensing Understanding

2026-07-11

AI Summary

This work addresses the challenge of ultra-high-resolution remote sensing image understanding, which requires balancing global scene layout and critical local details under limited computational resources. Existing approaches often suffer from performance degradation due to information redundancy or error accumulation. To overcome these limitations, the authors propose a training-free, general-purpose framework that reframes the task as structured evidence construction and reasoning under global contextual constraints. Rather than acquiring more data, the method organizes existing information to construct a compact, low-redundancy, and spatially complementary minimal supporting evidence set. Joint reasoning is performed by integrating local visual evidence, spatial metadata, and relative topological relationships. The framework flexibly adapts to various frozen vision-language model backbones and achieves significant performance gains over state-of-the-art methods across multiple ultra-high-resolution remote sensing benchmarks, demonstrating its effectiveness and broad applicability.

Abstract

Ultra-High-Resolution (UHR) remote sensing image understanding requires Vision-Language Models (VLMs) to capture both the global scene layout and sparse yet task-critical local details under limited computational budgets. Existing methods mainly follow two paradigms. One is passive perception, which relies on resolution expansion or token compression and may therefore discard fine-grained details. The other is active perception, which depends on multi-round zooming and search, but suffers from high latency, contextual fragmentation, and error accumulation. We argue that a more effective path toward UHR understanding lies not in accessing more, but in organizing better. To this end, we propose WeaveEarth, a training-free framework that reformulates UHR understanding as a problem of structured evidence construction and reasoning under global context constraints. Specifically, WeaveEarth first employs Global-Aware Evidence Construction to select a compact, low-redundancy, and spatially complementary Minimal Support Evidence Set. It then introduces Structured Evidence Reasoning, which weaves local evidence, spatial metadata, and relative topology into a unified reasoning interface, thereby enhancing the VLM's ability to perform global-local joint reasoning. Extensive experiments show that WeaveEarth consistently outperforms strong baselines and existing UHR methods across multiple UHR remote sensing benchmarks and multiple frozen VLM backbones. Code is available at https://github.com/XianZhi-Ma/WeaveEarth.

Problem

• Ultra-High-Resolution
• remote sensing
• Vision-Language Models
• evidence construction
• global-local reasoning

Innovation

• structured evidence reasoning
• training-free framework
• ultra-high-resolution remote sensing
• global-local joint reasoning
• minimal support evidence set