Idanwo Ohun elo

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

1. A detector returns a category label and a rectangle around each visible item. What task is it performing?
   • Object detection
   • Image classification only
   • Pixel-level segmentation only
   • Itumọ ọrọ
2. What does image classification often do without object detection?
   • Return a bounding box for each instance
   • Assign a label without locating the object
   • Segment each object pixel
   • Measure box overlap
3. Why define annotation rules for partly hidden or tiny objects?
   • Small objects do not affect evaluation
   • The rules guarantee perfect accuracy
   • Inconsistent boxes or omitted objects can confuse training and evaluation
   • They determine image resolution
4. What can a score threshold change in detector postprocessing?
   • The crop’s lighting
   • The image’s original dimensions
   • The ground-truth labels
   • The balance between missed objects and false detections
5. Two 10-by-10 boxes overlap over 50 pixels; their union is 150 pixels. What is IoU?
   • 0.50
   • About 0.33
   • 0.67
   • 1.00
6. Why do matching rules matter when reporting precision and recall?
   • They decide category names after training
   • They set camera exposure
   • The match threshold and included object sizes affect which predictions count
   • They remove the need for references
7. A detector works on large isolated products but misses small crowded items. What should be tested?
   • Real capture conditions, including crowding and occlusion
   • Only a clean laboratory image
   • Only the average score
   • The model’s name
8. A low-confidence detection could trigger a consequential action. What should be defined beforehand?
   • An automatic action with no review
   • An assumption that every box is correct
   • A rule to ignore all low scores
   • A human review process for uncertain detections