一作ACL 2026Findings· Findings of ACL
Inject to Heal: Alleviating hallucination in LVLM s via Context Embedding Injection
Mehrdad Fazli · Bowen Wei · Ziwei Zhu
摘要
Hallucinations—generating responses inconsistent with the visual input—remain a critical limitation of large vision-language models (LVLMs), especially in open-ended tasks such as image captioning and visual reasoning. In this work, we probe the layer-wise generation dynamics that drive hallucina...