Revealing and Enhancing Core Visual Regions: Harnessing Internal Attention Dynamics for Hallucination Mitigation in LVLMs

TL;DR

This paper introduces PADE, a training-free method that uses internal attention dynamics to identify and enhance core visual regions in LVLMs, significantly reducing hallucinations and improving reasoning accuracy.

Contribution

The paper proposes PADE, a novel attention intervention leveraging internal positive attention dynamics to mitigate hallucinations in LVLMs without additional training or computational overhead.

Findings

PADE improves visual grounding in LVLMs.

PADE reduces hallucination rates across benchmarks.

PADE maintains long-term output consistency.

Abstract

LVLMs have achieved strong multimodal reasoning capabilities but remain prone to hallucinations, producing outputs inconsistent with visual inputs or user instructions. Existing training-free methods, including contrastive decoding and auxiliary expert models, which incur several times more computational overhead and may introduce potential interference, as well as static internal signal enhancement, are often vulnerable to the attention sink phenomenon. We find that internal Positive Attention Dynamics (PAD) in LVLMs naturally reveal semantically core visual regions under the distortions of attention sinks. Based on this, we propose Positive Attention Dynamics Enhancement (PADE), a training-free attention intervention that constructs a PAD map to identify semantically core visual regions, applies per-head Median Absolute Deviation Scaling to adaptively control the intervention strength, and leverages System-Token Compensation to maintain attention to complex user instructions and support long-term output consistency. Experiments on multiple LVLMs and benchmarks show that PADE improves visual grounding and reduces hallucinations, validating the effectiveness of leveraging internal attention dynamics for reliable multimodal reasoning.