This directory contains optional MinerU OCR helpers. The extraction agent reads Markdown directly, so OCR is only needed when starting from PDFs, Office files, or images.

• official-api: MinerU official token API. The default base URL is https://mineru.net.
• self-hosted: a self-hosted MinerU HTTP service, invoked through the local mineru -b vlm-http-client command.

API credentials can be passed with --api-token, --api-key, or environment variables:

export MINERU_BACKEND=official-api
export MINERU_API_TOKEN=...
export MINERU_API_KEY=...
export MINERU_API_BASE_URL=https://mineru.net
export MINERU_SELF_HOSTED_URL=http://127.0.0.1:30000
export MINERU_SELF_HOSTED_API_TOKEN=...
export MINERU_SELF_HOSTED_MODEL=mineru

• scripts/ocr_batch.py: stage-1 OCR wrapper.
• scripts/run_paper_ocr_merged.py: merged OCR for one paper.
• scripts/batch_paper_ocr_merged.py: merged OCR for a batch of papers.
• pipeline/run_two_stage_ocr_merged.sh: shell wrapper for one paper.
• pipeline/run_batch_papers_ocr_merged.sh: shell wrapper for a batch.

Single paper:

bash ocr/pipeline/run_two_stage_ocr_merged.sh \
  --input /path/to/raw_paper_dir \
  --output /path/to/ocr_output

Batch:

bash ocr/pipeline/run_batch_papers_ocr_merged.sh \
  --input-root /path/to/raw_papers \
  --output-root /path/to/ocr_output \
  --paper-concurrency 4

Single-paper output layout:

<paper>/
├── <paper>.md
└── images/

OCR outputs are generated artifacts. Write them outside the repository, or to ignored directories such as runs/ or ocr/output/.