Guide-linked quiz · Medium Level

COCO Dataset and Annotation Format Quiz

Check COCO image-category links, x-y-width-height boxes, instance segmentation, coordinate conversion, visual QA, and dataset scope.

Question 1 of 8

A detection annotation has an image_id. What should that value link to?

Preview every question8 questions

Answers and explanations appear as you take the quiz.

  1. A detection annotation has an image_id. What should that value link to?
    • The category definition rather than an image
    • The model training run identifier
    • The image record containing the annotated object
    • The annotation’s own unique ID
  2. How is a standard COCO-style detection box commonly expressed?
    • Two opposing corners [x1,y1,x2,y2]
    • Upper-left x and y, then width and height
    • Center coordinates and half-width/height
    • A category ID followed by pixel count
  3. A source box is [x1,y1,x2,y2]. How should its width be converted?
    • Compute x2 minus x1
    • Use x2 directly as the width
    • Use x1+x2 as the width
    • Divide the image width by x2
  4. Why can one image have several annotation records?
    • One image must have a new ID per object
    • A category can occur only once per image
    • Each record represents an entire image rather than an instance
    • Each object instance can have its own category and geometry
  5. A warehouse dataset is converted to COCO JSON. What does that conversion itself establish?
    • Its records may be compatible with tools expecting the chosen COCO schema
    • Its image distribution now matches original COCO
    • Its category labels have been independently verified
    • Models trained on it will generalize to every warehouse
  6. A box stored in normalized coordinates is read as image pixels. What likely happens?
    • The box becomes larger because values are normalized
    • The category ID changes from integer to decimal
    • The parser automatically infers the missing pixel scale
    • The rendered box becomes tiny or misplaced
  7. Which check can reveal a category mapping error after conversion?
    • Verify only that the JSON parses
    • Draw sampled annotations and compare IDs, labels and image content
    • Inspect only total file size
    • Assume contiguous IDs imply correct category meaning
  8. What extra structure can instance segmentation supply beyond a box?
    • Guaranteed correctness of the category label
    • A verified camera-pose transform
    • A polygon or encoded mask describing an object’s pixels
    • A single image-level label replacing instances

Keep testing yourself

More quizzes picked for your level.

View all quizzes