
OPERA introduces a novel decoding method to alleviate hallucination issues in multi-modal large language models by implementing an Over-trust Penalty and Retrospection-Allocation strategy. The approach aims to reduce hallucinations without the need for additional data, knowledge, or training.


coremsg

Alleviating-Hallucination-in-Multi-Modal-Large-Language-Models-with-OPERA


Alleviating Hallucination in Multi-Modal Large Language Models with OPERA


title_rewrite


OPERA introduces a novel decoding method to reduce hallucinations in MLLMs without additional data or training.


opera-alleviating-hallucination-in-multi-modal-large-language-models


OPERA: Alleviating Hallucination in Multi-Modal Large Language Models
