HALC: Object Hallucination Reduction via Adaptive Focal-Contrast Decoding

arXiv:2403.00425v1 Announce Type: cross
Abstract: While large vision-language models (LVLMs) have demonstrated impressive capabilities in interpreting multi-modal contexts, they invariably suffer from object hallucinations (OH). We introduce HALC, a novel decoding algorithm designed to mitigate OH in LVLMs. HALC leverages distinct fine-grained optimal visual information in vision-language tasks and operates on both local and global contexts simultaneously. Specifically, HALC integrates a robust auto-focal grounding mechanism (locally) to correct hallucinated tokens on the fly, and a specialized beam search algorithm (globally) …

abstract algorithm arxiv capabilities contrast cs.ai cs.cv cs.lg decoding decoding algorithm fine-grained hallucination hallucinations information language language models modal multi-modal novel tasks type via vision vision-language models visual