MRGD guides MLLM decoding with a learned hallucination reward and a detector-based recall reward, allowing users to trade off object precision, recall, and test-time compute while reducing object hallucinations on CHAIR and AMBER.
Args: Alignment as reward-guided search
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
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Controlling Multimodal LLMs via Reward-guided Decoding
MRGD guides MLLM decoding with a learned hallucination reward and a detector-based recall reward, allowing users to trade off object precision, recall, and test-time compute while reducing object hallucinations on CHAIR and AMBER.