AI RESEARCH
Matching with Deliberation: Test-Time Evolutionary Hierarchical Multi-Agents for Zero-Shot Compositional Image Retrieval
arXiv CS.CV
•
ArXi:2605.22478v1 Announce Type: new Zero-Shot Compositional Image Retrieval (ZS-CIR) requires both preserving the visual continuity of the reference image and faithfully executing the semantic variables specified in the modification text, which constitutes the core challenge of the task. Existing methods often suffer from Perception Myopia in a single space, or fall into Logic Drift in iterative collaboration due to the perception ceiling of the underlying retriever.