opendatalab/MinerU mineru-4.0.11-released

MinerU 4.0.11

on GitHub

更新内容 / What's Changed

改进 / Improvements

• API：/v1/models 统一返回随服务版本变化的虚拟模型 ID（如 mineru-4.0.11），补充模型描述并保持注册时间稳定；/v1/tiers 的 current_model 同步使用该 ID，解析质量通过 tier 选择 / API: expose a single versioned virtual model ID through /v1/models, add its description and keep the registration timestamp stable; /v1/tiers uses the same ID for current_model, while parsing quality is selected through tier
• 图像处理：移除 MinerU 生产代码中的 OpenCV 导入和调用，OCR、公式、表格及印章矫正等图像操作通过 DocVortex 公共接口执行 / Image processing: remove OpenCV imports and calls from MinerU production code and use public DocVortex interfaces for OCR, formula, table and seal image operations

修复 / Fixes

• CLI：修复 mineru parse 单文件输入无法覆盖已有输出文件的问题；拒绝输出路径与输入文件相同，避免覆盖源文件；多文件输入继续要求输出为目录 / CLI: allow mineru parse to overwrite an existing output file for a single input, reject output paths that point to the input file, and continue requiring a directory for multiple inputs

依赖 / Dependencies

• docvortex 最低版本提升至 0.5.12（>=0.5.12,<1）/ Require docvortex>=0.5.12,<1
• 移除 OpenCV 直接依赖；Torch/full 等可选依赖配置保持原样，第三方依赖仍可能间接安装 OpenCV / Remove the direct OpenCV dependency; Torch/full extras retain their existing configuration, and third-party dependencies may still install OpenCV transitively

升级提示 / Upgrade Notes

• API 客户端应通过 GET /v1/models 获取当前虚拟模型 ID，并在服务升级后重新发现；使用 tier 选择解析质量，避免硬编码历史模型 ID / API clients should discover the current virtual model ID through GET /v1/models and refresh it after server upgrades; select parsing quality through tier instead of hard-coding historical model IDs

Full Changelog: mineru-4.0.10-released...mineru-4.0.11-released