Add page-level text and image extraction

#2
BigLAM: BigScience Libraries, Archives and Museums org
edited 12 days ago

Changes in v2.0:

  • Changed from single extracted_text to page_texts/page_images lists
  • Added page_numbers field to track which pages were extracted- Configurable page selection (first N + last N pages)
  • Both text and images available for VLM vs LLM benchmarking
    Structure:
  • page_texts: List of text strings, one per page- page_images: List of PIL Images, one per page- page_numbers: List of page numbers (1-indexed)- num_pages: Total pages in source PDF
    This enables:
  • Page-by-page analysis
  • Fair comparison between vision and text-only models
  • Clear mapping of which content came from which page
davanstrien changed pull request title from Add page-level text and image extraction Changes in v2.0: - Changed from single extracted_text to page_texts/page_images lists - Added page_numbers field to track which pages were extracted - Configurable page selection (first N + last N pages) - Both text and images available for VLM vs LLM benchmarking Structure: - page_texts: List of text strings, one per page - page_images: List of PIL Images, one per page - page_numbers: List of page numbers (1-indexed) - num_pages: Total pages in source PDF This enables: - Page-by-page analysis - Fair comparison between vision and text-only models - Clear mapping of which content came from which page to Add page-level text and image extraction
davanstrien changed pull request status to merged

Sign up or log in to comment