BiView-Match: A Dual-View Object Association Mechanism for Extreme Appearance Variations in Millimeter-Wave Security Imaging

Active millimeter-wave (AMMW) security screening struggles with cross-view target association due to extreme appearance variations. This paper introduces BiView-Match, a novel mechanism that leverages spatial localization cues via its SameLocationHead module and an asymmetric cross-attention network to establish robust target correspondences, bypassing reliance on appearance similarity. Experiments on the BiMMW dataset demonstrate BiView-Match significantly improves association accuracy under severe appearance changes. This work offers a viable solution for dual-view object association in AMMW imaging challenged by drastic appearance differences.