Omni-Woodpecker: Hallucination Correction via Evidence-Prior Rebalancing for Omni LLMs

One-line takeaway

Omni-Woodpecker corrects hallucinations via evidence-prior rebalancing, improving performance across benchmarks with strong cross-architecture generalization.

Omni-Woodpecker通过证据优先重平衡机制，纠正多模态大模型中的幻觉问题，在多个基准上显著提升性能且具备跨架构泛化能力。

Key points

Omni-Woodpecker addresses cross-modal hallucinations in Omni LLMs by rebalancing fused representations based on evidence-prior conflicts. It identifies modality-specific conflicts, decomposes hidden states into grounding evidence and competing priors, and adaptively suppresses misleading priors while restoring evidence before decoding. The method achieves significant performance gains across multiple benchmarks and generalizes well to different architectures without requiring retraining.

• OWP identifies cross-modal conflicts from modality-specific views to detect hallucination sources.
• It uses orthogonal decomposition to separate grounding evidence from competing priors in hidden states.
• The framework adaptively rebalances fused representations by suppressing misleading priors and restoring evidence.
• OWP achieves strong performance gains on CMM, AVHBench, and four general OLLM benchmarks across architectures.

Written only from the official arXiv abstract by an open-source model, then checked automatically (every number and name must appear in the abstract). The paper is authoritative.