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.
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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.