From Multimodal Retrieval to Cross-Modal Registration: Evidence Pathways in Vision-in-the-Loop Search

Abstract

Reading Universal Multimodal Neural Machine Translation Via Image Retrieval from Search Engines alongside Visual Image Browsing and Exploration (Vibe): User Evaluations of Image Search… and ISE: Interactive Image Search using Visual Content reveals that vision-in-the-loop search is as much an evaluation problem as a modeling problem. This article organizes those references around multimodal retrieval, cross-modal registration, and observation timing. It compares problem definitions, data assumptions, validation choices, and the limits placed on each study's conclusions. The synthesis closes with research questions about robustness, transfer, and accountability. Addressing them would connect the cited scholarship to experiments that are easier to reproduce, audit, and extend.