OCR (PDF → Markdown) with Ollama + deepseek-ocr

Converts a PDF to text (Markdown) by rendering each page to an image and sending it through the deepseek-ocr:latest model served by Ollama via its OpenAI-compatible API (Responses API).

Requirements

• Ollama running locally and the model pulled:
  • ollama pull deepseek-ocr:latest
• On Linux, pdf2image requires Poppler:
  • Debian/Ubuntu: sudo apt-get install poppler-utils

Installation (pipx)

pipx install git+https://github.com/arrase/OCR.git

Usage

ocr 2512.15741v1.pdf

Page selection

You can include/exclude pages (1-based) and use both at the same time; --include is applied first and then --exclude.

Examples:

# Only page 1
ocr --include 1 2512.15741v1.pdf

# Pages 1 to 5 except 3
ocr --include 1-5 --exclude 3 2512.15741v1.pdf

# Combinations
ocr --include 1,3,5-8 --exclude 6-7 2512.15741v1.pdf

Output: creates 2512.15741v1.md in the same directory.

Environment variables (optional)

• OLLAMA_BASE_URL (default http://localhost:11434/v1)
• OLLAMA_MODEL (default deepseek-ocr:latest)