Extract Page Content from Chandra (Datalab) OCR Results — chandra_extract

Splits Chandra's paginated markdown into per-page entries in the Tensorlake-compatible format used by the other providers. Datalab separates pages with a delimiter of the form \n\n{PAGE_NUMBER} followed by a run of dashes; this function splits on that marker.

Usage

chandra_extract_pages(result, pages = NULL)

Arguments

result: List. The parsed response from chandra_ocr().
pages: Integer vector. Page numbers to extract. If NULL (default), extracts all pages.

Value

List with one element per page, each containing page_number, page_header, section_header, text, tables, and other.

Details

Tables and equations are emitted inline in the markdown text, so the tables and other fields are returned empty for compatibility. Extracted figures/images are available on the raw result (result$structured_output$images) as a named list of base64 strings.

Author

Nathan C. Layman

Examples

if (FALSE) { # \dontrun{
result <- chandra_ocr("paper.pdf")
pages <- chandra_extract_pages(result)
pages[[1]]$text
} # }