PdfToDocument

Import

import { PdfToDocument } from '@stabrise/scaledp/pdf'

Group

Read

Reads

bytes

Writes

document

Needs

pdfjs-dist

Note

Expands one row into several.

Most PDFs already contain their text. Lifting it is instant, exact, and needs no model at all — so the first question for any PDF pipeline is whether OCR is needed.

import { Pipeline } from '@stabrise/scaledp'
import { PdfToDocument, hasUsableTextLayer } from '@stabrise/scaledp/pdf'

const rows = await new Pipeline([new PdfToDocument({ resolution: 300 })]).transform(file)

const needsOcr = rows.filter((row) => !hasUsableTextLayer(row.document))

See Skip OCR when the PDF already has text for the full two-pass pipeline.

Coordinates are pixels, not points

That is what resolution controls: it is not a render DPI here, it is the pixel space the boxes are expressed in. Match it to your PdfToImage.

Words versus runs

pdf.js reports text as runs — a line, a styled span, whatever the producer emitted. splitWords: true (the default) splits those into word boxes by measuring the run's own font, which is what makes entity boxes usable. Turning it off gives run-level boxes, which are coarser but exactly what the PDF said.

Document.text is the run texts joined by newlines either way, and Document.type is 'pdf'.

Parameters