doc-intelligence Intelligent Document Processing service. Classifies documents, splits multi-document batches, extracts structured fields, validates them against business rules, and flags ...
The best approach is a combination of existing OCR and LLMs. We will trace the experience of our development team in combining the accumulation of OCR and layout analysis with LLMs as a design to ...
We now come to the training and evaluation of LayoutLMv2 and LayoutXLM. We use the same split that we used for training LayoutLMv1 and load the artifact from W&B for this.
A large-scale synthetic dataset for document AI research in underrepresented Turkic languages — Kazakh and Kyrgyz. Built to cover the full document understanding pipeline: text detection, recognition ...
microsoft 's Collections Fara1.5 MagenticLite HARC Mage SWE-FastContext Rho Frog Models GridSFM MSR-ACC Skala ChatBench VibeVoice MediPhi Dayhoff Atlas NatureLM Paza NextCoder Phi-4 Phi-3 Phi-1 ...
你有没有试过复制一份PDF里的表格粘贴到Excel里？多半会得到一堆散乱的文字碎片，行列全乱了，数字和标题对不上号。这不是软件的bug，而是PDF这种格式天生的毛病。它从设计之初就是给人眼看的排版格式，不是给机器读的语义格式。文字在PDF里是按几何坐 ...
D-RAC是一套让AI读懂PDF等企业文档的分块方案，先统一转成PDF渲染图片，再用一次多模态大模型转成检索友好的Markdown，最后靠轻量AI规划分块边界，成本降低超77%，检索效果媲美甚至超越全文 ...
在RAG知识库构建中，PDF解析质量直接影响检索召回率。多栏排版和水印是常见难点，其根源在于PDF不存储阅读顺序，仅记录字符坐标。通过提取文本块的边界框（bbox），结合XY-cut版面分析算法，可递归切割页面并还原正确阅读顺序；同时利用水印的重复性、字体和颜色特征进行过滤。PyMuPDF提供了 ...
想象一下，你正在为公司搭建一个AI知识库。这个知识库就像一座智能图书馆，里面存放着成千上万份文档——有扫描的合同 ...
本内容遵循CC 4.0 BY-SA版权协议 从原理到工程落地，系统拆大模型应用全栈：原理与生态、Agent 与多智能体、MCP/A2A/ANP 协议、记忆管理、Skill、CLI、人工介入、幻觉治理、RAG 全链路（加载/分块 ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results