Multi-Object Tracking Quiz

Answers and explanations appear as you take the quiz.

1. What is the core goal of multi-object tracking?
   • To follow multiple objects across frames while keeping a consistent identity for each
   • To classify a single image
   • To compress a video
   • To generate new frames
2. What does the 'tracking-by-detection' paradigm involve?
   • Tracking without any detector
   • Detecting objects each frame, then linking those detections across frames
   • Only using audio cues
   • Tracking by manually drawing boxes
3. What role does the Kalman filter play in trackers like SORT?
   • It compresses images
   • It detects new objects from scratch
   • It predicts where each tracked object will move next
   • It adds color to frames
4. What did DeepSORT add on top of SORT?
   • A larger image size
   • Audio tracking
   • Automatic subtitles
   • A learned appearance embedding for re-identifying objects after occlusion
5. What is an 'identity switch' in multi-object tracking?
   • When the tracker swaps the IDs of two objects, often as they cross paths
   • Changing the video resolution
   • Turning the camera off
   • Renaming the video file
6. How did ByteTrack improve tracking accuracy?
   • By ignoring all detections
   • By also associating low-confidence detections instead of discarding them
   • By removing the detector
   • By using only one frame
7. What does HOTA aim to balance in multi-object tracking evaluation?
   • Image resolution and model size
   • Training speed and GPU memory
   • Detection accuracy and association accuracy
   • Object color and camera frame rate
8. How does SORT assign detections to predicted tracks?
   • It uses a language model to name each object
   • It matches only by color histograms
   • It trains a new detector for each frame
   • It solves a frame-level assignment problem using IoU-based costs and the Hungarian algorithm