CoF improves multimodal LLM benchmark scores by having the model locate an answer region, then reweighting attention toward that region during inference.
Vicuna: An open-source chatbot impressing gpt-4 with 90%* chatgpt quality,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2024 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
CoF: Coarse to Fine-Grained Image Understanding for Multi-modal Large Language Models
CoF improves multimodal LLM benchmark scores by having the model locate an answer region, then reweighting attention toward that region during inference.