Preprint
Sep 2026
Detecting Object Hallucinations in Large Vision-Language Models via Cross-Modal Attention Drifts and Mask-Based Verification
CADMP is proposed, a lightweight object hallucination detection framework that combines adjacent-layer cross-modal attention drift with prediction sensitivity to targeted visual masking, and achieves consistently competitive detection performance.
Xu Wen, Boxu Chen, Le Yang et al.
· 0 citations