A null-image diagnostic shows VLM attribute hallucinations track the visual logit margin, not the language prior, and a routed Calib/Abstain/Adapt framework reduces them.
V-DPO: Mitigating hallucination in large vision language models via vision-guided direct preference optimization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
When Visual Signals Mislead: A Mechanistic Study of Attribute Hallucination in Vision-Language Models
A null-image diagnostic shows VLM attribute hallucinations track the visual logit margin, not the language prior, and a routed Calib/Abstain/Adapt framework reduces them.