CutPaste&Find is a novel, efficient approach for detecting hallucinations in Large Vision-Language Models (LVLMs) without expensive API calls or iterative validation.

• Addresses object hallucination where models fabricate non-existent objects or incorrect attributes
• Employs a visual-aid knowledge base approach that's more efficient than existing solutions
• Enables practical deployment for large-scale or offline use cases
• Significantly reduces computational costs while maintaining robust detection performance

Security Impact: By effectively identifying when AI systems generate false information about visual content, this framework enhances the trustworthiness and safety of multimodal AI systems in critical applications.

CutPaste&Find: Efficient Multimodal Hallucination Detector with Visual-aid Knowledge Base