NEW Native Support for Conversational Data: 4 Use Cases to Ship Better Chatbots

PDF - Beta 🧪

note

We’re testing a new PDF tag. To enable it for your account please contact your Customer Success Manager or apply here: https://humansignal.com/pdf-interest-signup

The Pdf tag displays a PDF document in the labeling interface. You can use this tag to:

  • Perform document-level annotations such as classification, transcription, and summarization. See PDF for an example labeling config.
  • Create regions within pages for OCR, NER, and other types of annotations (beta).

Supports:

  • Zoom
  • Rotation
  • PDFs up to 100 pages

Use with the following data types: PDF.

Enterprise

You can also use the PDF tag with Prompts to perform auto-labeling work such as PDF summarization, classification, information extraction, and document intelligence.

Note that since this tag is still in beta, Label Studio’s AI tools will default to the basic implementation of the tag that only supports document-level classification.

Parameters

Param Type Description
value string Data field value containing the URL to the PDF

Example: OCR 🧪

Beta and Label Studio Enterprise only.

Labeling configuration for PDFs:

<View>
  <OcrLabels name="ocr" toName="pdf">
    <Label value="Typo"/>
    <Label value="Incorrect amount"/>
    <Label value="Incorrect name"/>
  </OcrLabels>

  <Pdf name="pdf" value="$pdf"/>
</View>

Example Input data:

{
  "pdf": "https://app.humansignal.com/static/samples/ocr-receipts.pdf"
}

OcrLabels

The above example uses OcrLabels.

This is a new tag to add bounding boxes to the PDF and assign labels to them. This tag must have one or more Label tag children, and supports standard parameters such as maxUsages (see RectangleLabels as an example).

Results:

Result Type Description
x, y, width, height Number Numbers from 0 to 1 that are relative to the page dimensions.
rotation Number Number in degrees clockwise from 0–360. Rotation is calculated with the origin at (x, y) (the top-left corner of the region).
pageIndex Number Page number, 1-based.
ocrtext String Captured text. This can be edited by selecting the region and then editing the text from the Info panel.

note

When you are rotating within the UI, it appears to originate from the center of the region. However, we store the origin as (x, y) - meaning in the top left corner of the region.



Designed for teams of all sizes Contact Sales