OneFocus: Enabling Real-World X-ray Security Screening with a Unified Vision-Language Model

A research team has introduced OneFocus, a unified vision-language model designed to improve automated contraband detection in X-ray security screening, alongside a new benchmark dataset of 52,124 image-caption pairs spanning 28 contraband classes. The model, detailed in a paper submitted to arXiv on June 14, 2026, is built to address a persistent weakness in conventional X-ray detectors: an inability to adapt to novel contraband types and a lack of deeper visual comprehension [1]. The researchers argue that while vision-language models (VLMs) offer stronger generalization, their application in this domain has been limited by a scarcity of high-quality training data [2]. To bridge this gap, the team constructed MMXray, a curated benchmark containing 52,124 image-caption pairs across 28 fine-grained contraband categories [1]. They supplemented this with CleanDET, a synthesis dataset providing clean foreground contraband images and background images with varied density levels, and AnyContraSyn, a controllable method for generating realistic occlusion patterns [2]. An extensible data curation pipeline called OnePipe was also developed to systematize the process [1]. OneFocus, trained on this foundation, is designed to perform four core tasks: visual question answering, contraband localization, classification, and image understanding [2]. The paper reports that the model achieves state-of-the-art performance on these tasks and demonstrates robust cross-domain generalization, establishing what the authors call a strong vision-language baseline for security screening [1]. The work comes amid a broader surge in generative AI applications, driven by advances in deep neural networks and transformer architectures that have enabled models to generate and interpret complex data, from text and images to video [3][4]. The security screening sector has increasingly sought such tools to handle the volume and sophistication of threats in logistics and transportation, where X-ray contraband detection is a critical chokepoint [1].