What Transfers across Vision-in-the-Loop Search Studies? Lessons from Agent Memory and Multimodal Retrieval

Abstract

Three strands anchor this examination of vision-in-the-loop search: Query-by-visual-search: multimodal framework for content-based image retrieval; Multimodal clothing recognition for semantic search in unconstrained surveillance imagery; and Efficient Search in a Panoramic Image Database for Long-term Visual Localization. The synthesis centers on agent memory, multimodal retrieval, and cross-modal registration, separating recurring design principles from application-specific choices. It also considers what information is required for an independent reader to reproduce or audit the findings. Taken together, the references support a conditional design framework rather than a universal method ranking. The proposed agenda calls for explicit failure analysis, comparable baselines, and reporting that makes scope and uncertainty visible.