能力路由的视觉检索与证据串联用于长上下文文档问答
Capability-Routed Visual Retrieval and Evidence Threading for Long-Context Document Question Answering
- University of Malaya（马来亚大学）
机构由 AI 辅助整理，请以论文原文为准。
AI总结：
针对长上下文文档问答，提出CAVR、WSPS和VET三个模块，通过能力路由、弱到强选择和证据串联，显著提升多页视觉问答准确率。
AI中文摘要：
年度报告、尽职调查资料包和信息图表仪表板将数字埋藏在页面图像中：坐标轴、单元格网格和脚注，这些内容被OCR流水线扁平化，而页面级视觉检索器仍将它们视为可互换的上下文示例。我们保持冻结的Qwen2.5-VL-7B-Instruct生成器和ColPali / VisRAG-Ret页面索引，并插入三个模块。能力感知的视觉路由器（CAVR）将每个检索到的页面标记为文本、表格、图表、布局或混合类型，并在生成前混合专家。弱到强的页面选择（WSPS）将冻结的7B可回答性教师模型蒸馏到3B选择头，使排序不再是单一的InfoNCE分数。视觉证据串联（VET）构建长度至多为三的布局锚定路径，让生成器读取串联路径而非扁平的top-k列表。在黄金页面DocVQA / ChartQA / InfographicVQA上，7B系统达到96.3 / 90.1 / 85.4。在VisRAG top-3协议下，平均生成准确率为62.74，而相同骨干网络使用拼接时为59.39。在MMLongBench-Doc的检索后读取任务中，F1从19.2提升到22.6，多页准确率从16.4提升到21.2。经过WSPS重排序后，ViDoRe nDCG@5为83.6，其中TAT-DQA财务报告为70.4。
英文摘要：
Annual reports, diligence packs, and infographic dashboards bury numbers in page images: axes, cell grids, and footnotes that OCR pipelines flatten and that page-level visual retrievers still treat as interchangeable in-context examples. We keep a frozen Qwen2.5-VL-7B-Instruct generator and a ColPali / VisRAG-Ret page index, and insert three modules. A capability-aware visual router (CAVR) tags each retrieved page as text, table, chart, layout, or mixed and mixes specialist experts before generation. Weak-to-strong page selection (WSPS) distils a frozen 7B answerability teacher into a 3B selection head so ranking is no longer a single InfoNCE score. Visual evidence threading (VET) builds layout-anchored paths of length at most three and lets the generator read the thread rather than a flat top-$k$ list. On gold-page DocVQA / ChartQA / InfographicVQA the 7B system reaches 96.3 / 90.1 / 85.4. Under the VisRAG top-3 protocol the mean generation accuracy is 62.74 versus 59.39 for the same backbone with concatenation. On MMLongBench-Doc retrieve-then-read, F1 moves from 19.2 to 22.6 and multi-page accuracy from 16.4 to 21.2. ViDoRe nDCG@5 after WSPS reranking is 83.6, with TAT-DQA financial reports at 70.4.
