文档内容提取引擎之MinerU

背景

解决文档内容提取

前期调研

• Tika
  https://github.com/apache/tika
• docling
  https://github.com/docling-project/docling-serve
• mistral OCR
  https://github.com/nicekate/mistral-ocr
• MinerU
  https://github.com/opendatalab/MinerU/blob/master/README_zh-CN.md
  https://github.com/opendatalab/MinerU/tree/master/projects/web_api
• PaddleOCR
  https://github.com/PaddlePaddle/PaddleOCR

整体上看，大部分都是基于PaddleOCR基础上集成，比较有特点的是 MinerU

部署

docker-compose.yml

version: '3.8'

services:
  mineru-api:
    image: mineru-api-full
    container_name: mineru-api
    runtime: nvidia  # 需要NVIDIA Container Runtime支持
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    shm_size: 32g
    ports:
      - "8888:8888"
      - "30000:30000"
    restart: unless-stopped

相关链接

基于MinerU 2.0的PDF解析API