看见关键之处：一种免训练的自引导多模态细节感知与推理框架
Seeing What Matters: A Training-Free Self-Guided Framework for Multimodal Detail Perception and Reasoning
提出免训练的双分支关键区域定位（梯度语义相关图 + 视觉词元独特性）与渐进式视觉词元剪枝，可直接插入现成 MLLM 的推理流程，用于提升高分辨率细节感知并降低计算量，对做 VLM 细粒度理解与高效推理的人可复用。
原文依据 SLoFo applies a progressive visual token pruning strategy to improve attention focus on key areas while reducing computational overhead
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论多模态大语言模型（MLLM）在多种视觉-语言任务上已取得显著成功。然而，固定分辨率模型在感知细粒度视觉细节方面仍面临挑战，其主要原因在于注意力分散与视觉模糊。为解决这些问题，我们提出 SLoFo，一个免训练、自引导的推理框架，模仿人类的“扫描-定位-聚焦”（Scan-Locate-Focus）过程。SLoFo 首先采用双分支机制识别关键图像区域：语义分支构建基于梯度的语义相关图，结构分支估计视觉词元的独特性，二者提供互补且鲁棒的证据。通过融合两个分支，SLoFo 感知并显式裁剪出关键区域。在推理过程中，借助额外裁剪出的子图像，SLoFo 应用渐进式视觉词元剪枝策略，以提升对关键区域的注意力集中程度，同时降低计算开销。在细节敏感基准与通用基准上的实验表明，SLoFo 无需训练或外部模块即可持续提升准确率（TextVQA 上 +4.79%，GQA 上 +2.58%）与鲁棒性（POPE-MSCOCO 对抗设置下 +4.60%）。
英文原文摘要 1185 字符
Multimodal large language models (MLLMs) have achieved remarkable success on diverse visual-language tasks. However, fixed-resolution models face challenges in perceiving fine-grained visual details, particularly due to *distracted attention* and *blurry vision*. To address these issues, we propose **SLoFo**, a training-free and self-guided inference framework that mimics the human "**S**can-**Lo**cate-**Fo**cus" process. SLoFo first adopts a dual-branch mechanism to identify critical image regions: the Semantic branch constructs a gradient-based semantic relevance map, and the Structure branch estimates visual token uniqueness offering complementary and robust evidence. By combining both branches, SLoFo perceives and explicitly crop critical regions. During inference, with additional cropped sub-image, SLoFo applies a progressive visual token pruning strategy to improve attention focus on key areas while reducing computational overhead. Experiments on detail-sensitive and general-purpose benchmarks show that SLoFo consistently improves accuracy (+4.79% on TextVQA, +2.58% on GQA) and robustness (+4.60% on POPE-MSCOCO adversarial) without training or external modules.