DialogueVPR Enables Conversational Visual Place Recognition.

Current language-guided geo-localization methods often struggle with the inherent ambiguity and incompleteness of real-world natural language descriptions because they rely on static, one-shot retrieval. DialogueVPR proposes a fundamental shift, framing visual place recognition as an interactive, conversational reasoning task. This new approach allows for dynamic clarification and refinement of spatial information. To support this, the researchers developed DlgQuest-Cities, the first large-scale dialogue-based benchmark for place recognition. They also introduced DQ-pilot, a unified reasoning framework that combines a cross-modal multi-level retriever with an intelligent questioner. DQ-pilot is trained through a curriculum involving supervised fine-tuning and reinforcement refinement, guided by task-aligned metrics. Experiments confirm that this reasoning-based method significantly outperforms traditional baselines.