Chart-FR1：面向稠密图表的视觉聚焦驱动细粒度推理
Chart-FR1: Visual Focus-Driven Fine-Grained Reasoning on Dense Charts
Focus-CoT 把推理步骤显式锚定到局部图像区域与 OCR 信号上，配合带信息效率奖励的 Focus-GRPO 压缩冗余视觉 token，这套「视觉聚焦+自适应推理深度」的机制可直接迁移到文档、图表、密集标注场景的多模态推理，也为视觉 token 冗余问题提供了 RL 层面的解法。
原文依据 we propose Focus-CoT, a visual focusing chain-of-thought that enhances fine-grained perception by explicitly linking reasoning steps to key visual cues
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（MLLMs）在图表理解与推理任务中已展现出相当大的潜力。然而，面对以多子图、图例和密集标注为特征的高信息密度（HID）图表，它们仍然表现不佳，原因主要有三方面挑战：（1）细粒度感知能力有限，导致关键视觉线索被遗漏；（2）冗余或噪声视觉信息削弱了多模态推理的性能；（3）缺乏相对于视觉信息量的自适应深度推理。为应对这些挑战，我们提出了一种新颖的聚焦驱动细粒度图表推理模型 Chart-FR1，以提升在 HID 图表上的感知能力、聚焦效率与自适应深度推理能力。具体而言，我们提出 Focus-CoT，一种视觉聚焦思维链，通过将推理步骤与关键视觉线索（如局部图像区域和 OCR 信号）显式关联来增强细粒度感知。在此基础上，我们引入 Focus-GRPO，一种聚焦驱动的强化学习算法，其信息效率奖励可压缩冗余视觉信息以实现高效聚焦，其自适应 KL 惩罚机制可在发现更多视觉线索时灵活控制推理深度。此外，为填补 HID 图表基准的空白，我们构建了 HID-Chart，这是一个具有挑战性的基准，并设计了信息密度指标以评估细粒度图表推理能力。在多个图表基准上的大量实验表明，Chart-FR1 在图表理解与推理方面优于最优的 MLLMs。代码已发布于 https://github.com/phkhub/Chart-FR1。
英文原文摘要 1688 字符
Multimodal large language models (MLLMs) have shown considerable potential in chart understanding and reasoning tasks. However, they still struggle with high information density (HID) charts characterized by multiple subplots, legends, and dense annotations due to three major challenges: (1) limited fine-grained perception results in the omission of critical visual cues; (2) redundant or noisy visual information undermines the performance of multimodal reasoning; (3) lack of adaptive deep reasoning relative to the amount of visual information. To tackle these challenges, we present a novel focus-driven fine-grained chart reasoning model, Chart-FR1, to improve perception, focusing efficiency, and adaptive deep reasoning on HID charts. Specifically, we propose Focus-CoT, a visual focusing chain-of-thought that enhances fine-grained perception by explicitly linking reasoning steps to key visual cues, such as local image regions and OCR signals. Building on this, we introduce Focus-GRPO, a focus-driven reinforcement learning algorithm with an information-efficiency reward that compresses redundant visual information for efficient focusing, and an adaptive KL penalty mechanism that enables flexible control over reasoning depth as more visual cues are discovered. Furthermore, to fill the gap in benchmarks for HID charts, we build HID-Chart, a challenging benchmark with an information-density metric designed to evaluate fine-grained chart reasoning capabilities. Extensive experiments on multiple chart benchmarks demonstrate that Chart-FR1 outperforms state-of-the-art MLLMs in chart understanding and reasoning. Code is available at https://github.com/phkhub/Chart-FR1.