Search papers, labs, and topics across Lattice.
This paper reformulates open-world geo-localization as a perceive-then-verify reasoning task, addressing issues of perceptual hallucination and context drift in existing models. The authors introduce GeoPAVE, a bi-level framework that integrates perception-based hypothesis generation with verification-based evidence grounding to enhance decision-making processes. A novel dataset, PAVED, is also introduced to facilitate rigorous evaluation, featuring real-world user check-in data and comprehensive reasoning trajectories for multi-step queries and tool invocations.
By framing geo-localization as a human-like reasoning challenge, GeoPAVE significantly reduces perceptual errors and enhances decision accuracy in ambiguous visual contexts.
Open-world geo-localization requires models to reason over ambiguous visual cues through multi-step reasoning and external knowledge grounding. While recent large vision-language models exhibit strong multimodal reasoning capabilities, existing approaches still suffer from perceptual hallucination and context drift due to the lack of explicit evidence-grounded verification. In this work, we reformulate geo-localization as a human-like perceive-then-verify reasoning problem and propose GeoPAVE (Geo-localization Perception-and-Verification-Engine), a bi-level agentic framework that contains perception-based hypothesis generation via single-pass rollouts and verification-based evidence grounding for decision actions: support, refute, and refine. To support rigorous evaluation, we further introduce PAVED, a novel dataset derived from real-world user check-in data, equipped with comprehensive reasoning trajectories featuring multi-hop queries, multi-round tool invocations, and structured perception-verification traces. The dataset and code are available at https://github.com/Arandinglv/GeoPAVE.
