Guide-linked quiz · Medium Level

Scene Text Detection and Recognition Quiz

Tests the separate detection and recognition stages, CRAFT and CRNN design ideas, evaluation metrics, and limits of scene-text outputs.

Question 1 of 8

What does a scene-text detector return?

Preview every question8 questions

Answers and explanations appear as you take the quiz.

  1. What does a scene-text detector return?
    • A verified transcript with legal meaning
    • A translation into the user’s language
    • An identity for the person who wrote the sign
    • Locations or regions in the image that are likely to contain text
  2. How does CRAFT use character-affinity scores?
    • To estimate links between nearby character regions so they can be grouped into text instances
    • To translate each word into another language
    • To measure the author’s handwriting speed
    • To prove that a text string is factually true
  3. Why is detection evaluated separately from recognition?
    • Recognition determines whether the camera has a GPS fix
    • Detection scores measure word spelling only
    • A recognizer cannot read text that the detector missed or cropped poorly
    • Both stages always produce the same errors
  4. What does the CRNN approach combine according to its paper?
    • A table-cell parser and document language model
    • Convolutional image features, recurrent sequence modeling, and transcription
    • Only a dictionary lookup over detected word boxes
    • A crowd-density map and a face detector
  5. Which test set is most useful for a sign-reading application?
    • Only clean horizontal English text from a lab
    • Images selected after the model’s predictions are known
    • A benchmark with no examples from the intended setting
    • Images with representative orientation, curvature, language, lighting, and resolution
  6. What can character error rate fail to show by itself?
    • Whether two strings have different lengths
    • Whether a reference transcript exists
    • Whether the detector missed a region or produced a false text region
    • Whether the transcript contains characters
  7. Why should recognized coordinates be kept with the transcript?
    • They eliminate the need to store the source image
    • They let a reviewer locate the image crop that generated the text
    • They prove the transcription is correct
    • They guarantee an accurate translation
  8. Before using a recognized safety label to trigger an action, what should a system do?
    • Show the image and uncertain read for verification in context
    • Assume the most fluent transcript is authoritative
    • Discard the crop to avoid human review
    • Treat recognition confidence as proof of instruction authority

Keep testing yourself

More quizzes picked for your level.

View all quizzes