引導式測驗 · 中等 等級

文件版面分析測驗

測試文件區域偵測、閱讀順序、資料集範圍、評估設計以及從擷取內容到原始頁面的可追溯性。

相關引導路徑文件版面分析
問題 1 的 8

Which task does document layout analysis perform?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. Which task does document layout analysis perform?
    • Locate page regions and infer their categories or relationships
    • Verify the factual accuracy of every sentence
    • Identify the author of an unsigned page
    • Replace OCR by recovering all text exactly
  2. Which reading-order failure can a coordinate-only top-to-bottom sort cause on a two-column page?
    • A caption can be assigned to the wrong figure because its relation is unclear
    • A detector can merge adjacent text lines into one region
    • It can alternate lines from the left and right columns instead of reading each column in order
    • A recognizer can reverse character order within a line
  3. How does PubLayNet differ in annotation source from DocLayNet?
    • PubLayNet annotations are manually drawn on mixed documents, while DocLayNet is derived from XML tags
    • PubLayNet labels were matched from article XML and content, while DocLayNet uses human-annotated page layouts
    • Both datasets derive labels from article XML; DocLayNet differs mainly by adding OCR transcripts
    • PubLayNet uses manually labeled scans of forms, while DocLayNet maps PubMed XML to page regions
  4. What does a high detection score on one dataset establish?
    • Guaranteed accuracy on forms and photographed pages
    • Correct reading order for all page layouts
    • That the model verifies document claims
    • Performance on that dataset’s evaluated distribution, not every document type
  5. Which test set would best reveal domain shift for an archive of forms?
    • Repeated copies of the training pages
    • Synthetic text boxes without rendered pages
    • Held-out forms from the archive’s target sources and formats
    • Only research-paper pages used during training
  6. Which output lets a reviewer locate the exact image region that produced recognized text?
    • A confidence value for each recognized token without any location data
    • The coordinates or bounding box linked to the extracted text region
    • A page-level reading-order index attached to the full OCR transcript
    • An average character-recognition score for the whole page
  7. What might a mean average precision score fail to evaluate?
    • Whether any region boxes exist
    • Whether a model produced class labels
    • Whether test pages include annotations
    • Reading order or semantic relationships between regions
  8. Before relying on parsed content from a high-impact record, what should a reviewer do?
    • Use dataset scores as a substitute for checking the record
    • Compare the extracted output with the rendered source page
    • Assume correct layout labels prove the text is true
    • Discard page coordinates after OCR