AI RESEARCH
STEAR: Layer-Aware Spatiotemporal Evidence Intervention for Hallucination Mitigation in Video Large Language Models
arXiv CS.CV
•
ArXi:2604.03045v1 Announce Type: new Video Large Language Models (Video-LLMs) remain prone to spatiotemporal hallucinations, often generating visually uned details or incorrect temporal relations. Existing mitigation methods typically treat hallucination as a uniform decoding failure, applying globally shared correction rules. We instead observe that decoder layers contribute differently to visual grounding and later linguistic composition, indicating that intervention must be layer-aware. Based on this insight, we propose STEAR, a layer-aware spatiotemporal evidence intervention framework.