ZoomEarth: Active Perception for Ultra-High-Resolution Geospatial Vision-Language Tasks

2025-11-15

AI Summary

To address redundancy arising from passive perception paradigms in ultra-high-resolution (UHR) remote sensing image understanding, this paper proposes an active perception paradigm: adaptively cropping and dynamically scaling regions of interest to focus on information-dense areas, while introducing a region-guided reward mechanism to enhance fine-grained vision-language reasoning. Key contributions include: (1) the first large-scale remote sensing benchmark—LRS-GRO—designed specifically for active perception; (2) ZoomEarth, a scalable framework integrating supervised fine-tuning (SFT), grouped relative policy optimization (GRPO), and a dynamic cropping-scaling architecture; and (3) plug-and-play support for downstream tasks including cloud removal, denoising, and segmentation. ZoomEarth achieves state-of-the-art performance on LRS-GRO and demonstrates superior zero-shot transferability across three public UHR benchmarks, validating both effectiveness and generalization capability.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Active Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Ultra-high-resolution (UHR) remote sensing (RS) images offer rich fine-grained information but also present challenges in effective processing. Existing dynamic resolution and token pruning methods are constrained by a passive perception paradigm, suffering from increased redundancy when obtaining finer visual inputs. In this work, we explore a new active perception paradigm that enables models to revisit information-rich regions. First, we present LRS-GRO, a large-scale benchmark dataset tailored for active perception in UHR RS processing, encompassing 17 question types across global, region, and object levels, annotated via a semi-automatic pipeline. Building on LRS-GRO, we propose ZoomEarth, an adaptive cropping-zooming framework with a novel Region-Guided reward that provides fine-grained guidance. Trained via supervised fine-tuning (SFT) and Group Relative Policy Optimization (GRPO), ZoomEarth achieves state-of-the-art performance on LRS-GRO and, in the zero-shot setting, on three public UHR remote sensing benchmarks. Furthermore, ZoomEarth can be seamlessly integrated with downstream models for tasks such as cloud removal, denoising, segmentation, and image editing through simple tool interfaces, demonstrating strong versatility and extensibility.

Problem

• Active perception paradigm for ultra-high-resolution remote sensing
• Overcoming redundancy in fine-grained visual information processing
• Adaptive framework for revisiting information-rich regions in geospatial imagery

Innovation

• Active perception paradigm revisits information-rich regions
• Adaptive cropping-zooming framework with Region-Guided reward
• Trained via supervised fine-tuning and Group Relative Policy Optimization