2 papers across 1 session
Contrastive decoding fails to mitigate object hallucinations in MLLMs—apparent improvements stem from misleading factors rather than genuine effectiveness.