Guide-linked quiz · Medium Level

Table Structure Recognition Quiz

Tests table detection and structure recognition, Table Transformer inputs, grid metrics, and validation of extracted cell relationships.

Related guide pathsTable Structure Recognition
Question 1 of 8

What does table-structure recognition recover beyond a table’s outer boundary?

Preview every question8 questions

Answers and explanations appear as you take the quiz.

  1. What does table-structure recognition recover beyond a table’s outer boundary?
    • A list of all words without cell positions
    • Rows, columns, cells, spans, and their relationships
    • Only the document’s page size
    • The legal meaning of each paragraph
  2. In the documented Table Transformer inference pipeline, where can cell text come from?
    • A face-landmark model
    • A density map predicted from the page
    • A separate OCR result or text extracted directly from the PDF
    • Only the table detector’s bounding boxes
  3. What does GriTS evaluate in table structure recognition?
    • Similarity between predicted and reference table grids across topology, content, and location
    • Only OCR character accuracy in isolation
    • The number of pages in a PDF
    • Whether a document contains a table boundary
  4. Why does PubTables-1M canonicalize some table annotations?
    • To remove all empty cells from tables
    • To convert every PDF into an image-only format
    • To guarantee models perform equally on every domain
    • To reduce oversegmentation and improve consistency in the structure ground truth
  5. A parser exports the right words but shifts one column. What failed?
    • The PDF file has no text encoding
    • The OCR language model changed its vocabulary
    • Cell-to-text alignment or structure recovery
    • The page-level table detector necessarily failed
  6. Which tables should be included in a structure-recognition evaluation?
    • Examples with merged headers, borderless cells, and varied document styles
    • Only tables with clear grid lines and one header row
    • Only tables selected after inspecting model output
    • Only blank tables without text content
  7. What should a team do with an extracted total used for a payment?
    • Assume a valid CSV means the value is correct
    • Discard source coordinates once export succeeds
    • Accept a structure score as proof of the total
    • Check it against the source document and preserve a review path
  8. How should a benchmark score for table extraction be interpreted?
    • As a reason to skip inspection of merged cells
    • As performance on its evaluated datasets, not a guarantee for every organization’s PDFs
    • As proof a model cannot extract any tables
    • As direct validation of every downstream financial formula

Keep testing yourself

More quizzes picked for your level.

View all quizzes