Less is More: Mitigating Multimodal Hallucination from an EOS Decision Perspective

arXiv:2402.14545v1 Announce Type: cross
Abstract: Large Multimodal Models (LMMs) often suffer from multimodal hallucinations, wherein they may create content that is not present in the visual inputs. In this paper, we explore a new angle of this issue: overly detailed training data hinders the model's ability to timely terminate generation, leading to continued outputs beyond visual perception limits. By investigating how the model decides to terminate generation with EOS, the special end-of-sentence token, we find that the model assesses the …

abstract arxiv cs.cl cs.cv data decision explore hallucination hallucinations inputs issue large multimodal models lmms multimodal multimodal models paper perspective training training data type visual