Selecting the intermediate layer where image-conditioned and text-only predictions diverge most, and adding that layer's contrastive visual signal back to the final logits, reduces object hallucinations in four large multimodal models.
Improved baselines with visual instruction tuning,
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
-
Extracting Visual Facts from Intermediate Layers for Mitigating Hallucinations in Multimodal Large Language Models
Selecting the intermediate layer where image-conditioned and text-only predictions diverge most, and adding that layer's contrastive visual signal back to the final logits, reduces object hallucinations in four large multimodal models.