Non-Maximum Suppression Quiz

Answers and explanations appear as you take the quiz.

1. What is the main purpose of Non-Maximum Suppression in object detection?
   • To remove redundant overlapping boxes, keeping one per object
   • To increase the number of detected boxes
   • To resize the input image
   • To train the detector's weights
2. Which metric does standard NMS use to decide whether two boxes overlap too much?
   • Euclidean distance between centers
   • Intersection over Union (IoU)
   • Cosine similarity of features
   • Pixel color difference
3. In greedy NMS, which box is selected first in each round?
   • The largest box by area
   • The box closest to the image center
   • The box with the highest confidence score
   • A randomly chosen box
4. What is a known weakness of greedy NMS in crowded scenes?
   • It cannot handle color images
   • It always produces too few boxes overall
   • It requires GPU training to run
   • It may suppress a genuine nearby object that overlaps a kept box
5. How does Soft-NMS differ from standard NMS?
   • It decays the scores of overlapping boxes instead of removing them outright
   • It deletes all boxes regardless of score
   • It ignores confidence scores entirely
   • It only works during training
6. Why is NMS typically applied separately for each object class?
   • Because each class needs a different image resolution
   • So a box of one class doesn't suppress a box of a different class
   • To make the algorithm slower on purpose
   • Because IoU cannot be computed across classes
7. How is Intersection over Union (IoU) calculated for two detection boxes?
   • Union area divided by the intersection area
   • The sum of both box perimeters
   • Intersection area divided by the union area
   • The distance between the box centers
8. What is a typical effect of lowering the IoU threshold in greedy NMS?
   • Fewer boxes are suppressed in all cases
   • The detector retrains its class labels
   • The confidence scores are converted into image pixels
   • More overlapping boxes are suppressed, which can risk removing nearby real objects