Search papers, labs, and topics across Lattice.
This paper introduces Falcon, a multimodal framework designed for compositional threat reasoning in X-ray baggage screening, addressing the limitations of conventional object-centric vision-language models. By modeling risk as a relational property of spatially dispersed components rather than individual detections, Falcon enables a structured representation of safety states that incorporates component presence and functional compatibility. Experimental results demonstrate that Falcon significantly enhances functional grounding and threat assessment coherence compared to existing models, establishing a new evaluation paradigm for multimodal systems in safety-critical applications.
Falcon redefines threat assessment in X-ray screening by shifting from object detection to relational reasoning about component interactions.
Conventional vision-language models are largely object-centric, focusing on detecting and describing individual entities. In safety-critical X-ray baggage screening, however, threat often emerges not from a single object but from the functional compatibility of spatially dispersed components, such as batteries, detonators, and explosive charges. We formalize this setting as \emph{compositional threat reasoning}, where risk is modeled as a relational property of grounded regions rather than an independent detection outcome. We introduce \textbf{Falcon}, a multimodal framework that abstracts segmentation-aware region features into a structured safety state capturing component presence, pairwise functional compatibility, and scene-level risk. This structured representation is injected into the language model as an explicit intermediate interface, encouraging relationally consistent and safety-aware reasoning. To evaluate this problem, we present \textbf{Falcon-X}, a benchmark that unifies dense grounding with structured supervision over component completeness and risk inference in cluttered X-ray imagery. Experiments show that while existing multimodal models adapt to appearance, they struggle with compositional safety reasoning. Falcon improves functional grounding and produces more coherent threat assessments, establishing compositional safety reasoning as a distinct evaluation paradigm for multimodal systems.
