Search papers, labs, and topics across Lattice.
This paper introduces CADMP, a novel framework for detecting object hallucinations in large vision-language models by analyzing cross-modal attention drifts across adjacent layers and employing mask-based verification. The method quantifies distributional changes in attention maps during decoding to identify significant shifts in visual grounding, while also measuring prediction probability changes after masking visually relevant regions. Experimental results show that CADMP consistently outperforms existing detection methods, highlighting the importance of both attention stability and prediction sensitivity in identifying hallucinated outputs.
Object hallucinations in LVLMs can be effectively detected by analyzing attention drifts and prediction changes, revealing critical insights into model reliability.
Despite recent advances in large vision-language models (LVLMs), object hallucination remains a major barrier to their reliable deployment. Existing detection methods often characterize visual grounding using attention from individual layers, leaving its evolution across layers underexplored. We propose CADMP, a lightweight object hallucination detection framework that combines adjacent-layer cross-modal attention drift with prediction sensitivity to targeted visual masking. During decoding, CADMP quantifies distributional changes between consecutive cross-modal attention maps to capture abrupt transitions in visual grounding. It then selects the transition with the largest drift, locates the corresponding visually relevant regions, and measures the change in prediction probability after masking these regions. These two signals provide complementary evidence: attention drift characterizes the stability of internal visual grounding, while probability variation verifies whether a prediction truly depends on the identified visual evidence. A lightweight detector integrates both signals to identify hallucinated predictions. Experiments on multiple benchmarks and representative open-source LVLMs demonstrate that CADMP achieves consistently competitive detection performance. Ablation studies further confirm the complementary contributions of adjacent-layer drift modeling and mask-based grounding verification.
