Skip to content
Permalink

Comparing changes

Choose two branches to see what’s changed or to start a new pull request. If you need to, you can also or learn more about diff comparisons.

Open a pull request

Create a new pull request by comparing changes across two branches. If you need to, you can also . Learn more about diff comparisons here.
base repository: opendatalab/MinerU
Failed to load repositories. Confirm that selected base ref is valid, then try again.
Loading
base: master
Choose a base ref
...
head repository: Zhruoshui/Embedded_MinerU
Failed to load repositories. Confirm that selected head ref is valid, then try again.
Loading
compare: master
Choose a head ref
Checking mergeability… Don’t worry, you can still create the pull request.
  • 5 commits
  • 231 files changed
  • 1 contributor

Commits on May 8, 2026

  1. feat: 接入在线多模态视觉模型增强图片语义描述 

    新增 image-description-config 配置，支持通过在线视觉模型（Qwen3-VL-32B-Thinking 等
OpenAI-compatible API）对文档中的图片/chart 块进行语义内容增强，替代本地 VLM
模型输出，显著提升图片描述准确率。

功能要点：
- 配置驱动：mineru.template.json 中新增 image-description-config 段，字段包括
  enable/api_key/base_url/model/enable_thinking/max_tokens/temperature
- 统一泛化提示词：模型自动识别图片类型（照片/图表/流程图/表格/UI截图等），
  选择最佳输出格式（自然语言描述/markdown表格/mermaid/结构化文本）
- 后处理增强：在 VLM 版面分析完成、图片写入磁盘后，对 image/chart 块的
  content 字段调用在线模型重写，发生在 finalize_middle_json 之前
- 图片预处理：最长边缩放至 2048px，JPEG 质量 85，base64 编码传输
- 降级策略：API 调用失败时保留本地 VLM 原始 content，记录 warning 日志
- Thinking 模式支持：自动解析 <think>...</think> 标签，仅保留实际回答
- 并发处理：ThreadPoolExecutor 控制最大并发数为 3
- 覆盖 VLM 和 Hybrid 所有后端（同步 + 异步 共 6 个调用点）

新增文件：
- mineru/backend/vlm/image_enhance.py    核心逻辑模块

修改文件：
- mineru/utils/config_reader.py          新增 get_image_description_config()
- mineru/backend/vlm/vlm_analyze.py      同步/异步流程集成
- mineru/backend/vlm/model_output_to_middle_json.py  独立入口集成
- mineru/backend/hybrid/hybrid_analyze.py            同步/异步流程集成
- mineru/backend/hybrid/hybrid_model_output_to_middle_json.py  独立入口集成
- mineru/utils/llm_client.py             修复 pylint W0718 (broad-exception-caught)
    @Zhruoshui
    Zhruoshui committed May 8, 2026
    Configuration menu
    Copy the full SHA
    c21219f View commit details
    Browse the repository at this point in the history

  2. fix: 修复图片描述增强的两个问题 

    1. _collect_image_spans 未处理两层嵌套结构：MagicModel 生成的 image/chart 块是
   image → blocks → image_body → lines → spans 的嵌套结构，修复后正确遍历
2. enable_thinking 参数不兼容：Qwen3-VL-32B-Thinking 在硅基流动 API 中已内置
   thinking 模式，不再通过 extra_body 传递该参数
3. 增强计数逻辑改为对比原始内容与增强后内容，确保只统计真正被替换的图片
    @Zhruoshui
    Zhruoshui committed May 8, 2026
    Configuration menu
    Copy the full SHA
    290fb52 View commit details
    Browse the repository at this point in the history

  3. chore: remove non-core resources (web UI, docker, CI, docs, demo) and… 

    … add system overview explore doc
    @Zhruoshui
    Zhruoshui committed May 8, 2026
    Configuration menu
    Copy the full SHA
    6089b9c View commit details
    Browse the repository at this point in the history

  4. 删除非必要的文档及测试流程

    @Zhruoshui
    Zhruoshui committed May 8, 2026
    Configuration menu
    Copy the full SHA
    b3c4037 View commit details
    Browse the repository at this point in the history

  5. docs: add project README and CodeStable workflow files 

    - Add README introducing the fork, embedded domain focus, and quick start
- Add .codestable/ with architecture, requirements, features, issues docs
- Document image semantic enhancement feature (design + acceptance)
- Document image enhancement bug fixes (report + fix-note)
- Document scope decision to strip non-core distribution artifacts
- Add compound knowledge: explore, trick, and learning docs
- Track .codestable/ in git (remove from .gitignore)
    @Zhruoshui
    Zhruoshui committed May 8, 2026
    Configuration menu
    Copy the full SHA
    b6247ee View commit details
    Browse the repository at this point in the history
Loading