Feed-forward neurons in LVLMs encode whether a text token is visually grounded, and a detector built on these neurons can reduce hallucination by overriding or replacing ungrounded tokens.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Unveiling the Response of Large Vision-Language Models to Visually Absent Tokens
Feed-forward neurons in LVLMs encode whether a text token is visually grounded, and a detector built on these neurons can reduce hallucination by overriding or replacing ungrounded tokens.