Panoptic Segmentation Quiz

Answers and explanations appear as you take the quiz.

1. What does panoptic segmentation assign to every pixel in an image?
   • A category label, plus an instance ID for countable objects
   • Only a depth value
   • A single bounding box
   • A blur level
2. In panoptic terminology, what are 'stuff' classes?
   • Countable objects like cars and people
   • Amorphous regions like sky, road, or grass
   • Text labels in the image
   • Camera metadata
3. What is the main limitation of semantic segmentation that panoptic segmentation overcomes?
   • It cannot label the sky
   • It only works on video
   • It cannot distinguish individual instances of the same category
   • It ignores all background
4. Which metric is used to evaluate panoptic segmentation quality?
   • Frames per second
   • Compression ratio
   • Color accuracy
   • Panoptic Quality (PQ)
5. What does segmentation quality (SQ) measure within Panoptic Quality?
   • The average IoU of matched predicted and ground-truth segments
   • The number of images in the dataset
   • The speed of inference
   • The fraction of classes that are things
6. Which modern architecture handles both things and stuff in one unified, query-based model?
   • JPEG
   • Mask2Former
   • ResNet-18 alone
   • A rule-based filter
7. How is Panoptic Quality (PQ) decomposed in the original formulation?
   • As image resolution multiplied by object count
   • As precision plus recall only
   • As segmentation quality multiplied by recognition quality
   • As frame rate divided by latency
8. Why can a single PQ score conceal useful diagnostic information?
   • PQ measures only image brightness
   • PQ excludes stuff classes
   • PQ is not computed from predictions
   • Segmentation and recognition errors can contribute differently to the aggregate