Table and structured fields not detected — large table region skipped entirely

[ramrohit3545]

🔎 Search before asking | 提交之前请先搜索

• I have searched the MinerU Readme and found no similar bug report.
• I have searched the MinerU Issues and found no similar bug report.
• I have searched the MinerU Discussions and found no similar bug report.

🤖 Consult the online AI assistant for assistance | 在线 AI 助手咨询

• I have consulted the online AI assistant but was unable to obtain a solution to the issue.

Description of the bug | 错误描述

MinerU is failing to detect a large table block and multiple structured fields in the PDF.
The entire highlighted region in the PDF (routing, flight info, handling info, and goods details) is not extracted, and the JSON output only contains text from other sections.

How to reproduce the bug | 如何复现

not extract all.miss some

Operating System Mode | 操作系统类型

Linux

Operating System Version| 操作系统版本

Ubuntu 22.04

Python version | Python 版本

3.12

Software version | 软件版本 (mineru --version)

>=2.5

Backend name | 解析后端

vlm

Device mode | 设备模式

cuda

[dosubot]

This is a known limitation of the MinerU VLM backend: large or complex tables—especially those without clear borders, with high-resolution scans, or irregular layouts—may be skipped or misclassified as plain text, resulting in missing structured fields in the output. The VLM backend relies on the model's output order and does not perform spatial correction like the pipeline backend, so it can miss or fragment large tables and multi-field regions. There is no direct CLI/config fix for this in VLM mode; the merging logic only works if the model output is already well-structured, and can't recover missing or split tables if the model fails to detect them in the first place [issue #4123, issue #3783].

Recommended steps:

• Try the pipeline backend (--backend pipeline) or force OCR mode (--method ocr), which often improves extraction for scanned or complex documents [issue #3943].
• Make sure you are using MinerU >=2.5.3 and have up-to-date dependencies (mineru-vl-utils >=0.1.14, latest vllm image) for recent table merging and recognition fixes [issue #3998].
• Check that table detection is enabled (MINERU_TABLE_ENABLE=true in your environment/config).
• Preprocess the PDF/image: enhance contrast, crop to the table region, and remove marks or watermarks to help the model detect tables [issue #4123].
• Inspect middle.json and layout.pdf outputs to see if the table/fields are present but misclassified or discarded. Sometimes, content is present in the discarded_blocks section and can be manually recovered [docs].
• For extreme cases, manual post-processing of the output JSON or Markdown may be needed to merge split rows or recover missed content [issue #3861].

If you can share a sample of your middle.json or layout.pdf, I can help pinpoint where the extraction is failing.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}