Aligning with Your Own Voice: Self-Corrected Preference Learning for Hallucination Mitigation in LVLMs

TL;DR

This paper introduces AVES-DPO, a novel self-correction framework for LVLMs that effectively reduces hallucinations by leveraging in-distribution data and a consensus verification mechanism.

Contribution

The paper presents AVES-DPO, a new preference learning method that aligns LVLMs using intrinsic knowledge, avoiding reliance on proprietary models and reducing hallucinations.

Findings

AVES-DPO outperforms existing methods in hallucination mitigation.

Requires only 5.2k samples for effective alignment.

Uses a consensus-based verification to diagnose and correct hallucinations.

Abstract

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.