让视觉模型在长文本生成中持续记住图像，提升复杂推理能力。
Persistent Visual Memory: Sustaining Perception for Deep Generation in LVLMs
- 引入轻量可学习的视觉记忆模块，直接提供图像特征供调用。
- 在4B和8B模型上均显著提升准确率，尤其在长序列推理中表现更好。
- 适合需要长时间保持视觉感知的生成任务，如多步视觉推理。
尽管自回归大视觉语言模型在多模态任务中表现出色，但在深度生成过程中面临“视觉信号稀释”问题：随着文本历史累积，注意力分区函数扩大，导致视觉注意力随生成长度增加而反向衰减。为此，我们提出持久视觉记忆（PVM），一种轻量级可学习模块，旨在增强对视觉证据的持续、按需访问。PVM作为并行分支集成于LVLM的前馈网络（FFN）旁，建立与距离无关的检索路径，直接提供视觉嵌入以增强视觉感知，从而从结构上缓解深度生成中的信号抑制。在Qwen3-VL模型上的大量实验表明，PVM在参数开销极低的情况下带来显著性能提升，在4B和8B规模模型上均实现一致的平均准确率增益，尤其在依赖持久视觉感知的复杂推理任务中效果明显。深入分析显示，PVM提升了长生成任务的鲁棒性，并加速了内部预测收敛。
原文摘要 · Abstract (English)
While autoregressive Large Vision-Language Models (LVLMs) demonstrate remarkable proficiency in multimodal tasks, they face a "Visual Signal Dilution" phenomenon, where the accumulation of textual history expands the attention partition function, causing visual attention to decay inversely with generated sequence length. To counteract this, we propose Persistent Visual Memory (PVM), a lightweight learnable module designed to strengthen sustained, on-demand access to visual evidence. Integrated as a parallel branch alongside the Feed-Forward Network (FFN) in LVLMs, PVM establishes a distance-agnostic retrieval pathway that directly provides visual embeddings for enhanced visual perception, thereby structurally mitigating the signal suppression inherent to deep generation. Extensive experiments on Qwen3-VL models demonstrate that PVM brings notable improvements with negligible parameter overhead, delivering consistent average accuracy gains across both 4B and 8B scales, particularly in complex reasoning tasks that demand persistent visual perception. Furthermore, in-depth analysis reveals that PVM shows improved robustness in longer generations and accelerates internal prediction convergence.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。