Aligning with Your Own Voice: Self-Corrected Preference Learning for Hallucination Mitigation in LVLMs
For LVLM developers, this reduces reliance on proprietary models and distribution mismatch, offering a more efficient alignment method.
The paper tackles hallucination in Large Vision-Language Models (LVLMs) by proposing AVES-DPO, a self-corrected preference learning framework that uses in-distribution data from the model's own knowledge, achieving state-of-the-art hallucination mitigation with only 5.2k samples.
Large Vision-Language Models (LVLMs) frequently suffer from hallucinations. Existing preference learning-based approaches largely rely on proprietary models to construct preference datasets. We identify that this reliance introduces a distributional mismatch between the proprietary and target models that hinders efficient alignment. To address this, we propose Alignment via VErified Self-correction DPO (AVES-DPO), a framework that aligns LVLMs using in-distribution data derived from the model's intrinsic knowledge. Our approach employs a consensus-based verification mechanism to diagnose diverse hallucinations and guides the model to self-correct, thereby generating preference pairs strictly compatible with its internal distribution. Extensive experiments demonstrate that AVES-DPO surpasses existing baselines in hallucination mitigation while requiring only 5.2k samples.