Bildsegmentierungsquiz

Während Sie am Quiz teilnehmen, werden Antworten und Erklärungen angezeigt.

1. A project labels every road pixel as road. Which segmentation type is this?
   • Semantic segmentation
   • Instance segmentation
   • Bildklassifizierung
   • Object detection only
2. A project must distinguish each car from other cars of the same category. Which task fits?
   • Semantic segmentation only
   • Instance segmentation
   • Bildklassifizierung
   • Clustering
3. Why document how reviewers label uncertain boundaries?
   • Annotation rules replace testing
   • All boundaries are always objectively obvious
   • Different reviewers may draw different masks, affecting evaluation
   • Only image colors matter
4. An image has 950 background pixels and 50 target pixels; a prediction labels everything background. What pixel accuracy results?
   • 50%, with balanced classes
   • 5%, with perfect target recall
   • 100%, with perfect segmentation
   • 95%, despite detecting none of the target
5. Why can overlap metrics such as IoU be more informative than pixel accuracy?
   • They guarantee accurate boundaries
   • They assess overlap of predicted and reference regions
   • They ignore class balance
   • They prove the mask is safe to use
6. A mask looks correct before export but shifts when placed back on the original image. What should the team inspect?
   • The class name alphabetically
   • The model’s training company
   • Cropping, resizing, and coordinate conversion
   • Only the overall pixel score
7. Why inspect segmentation overlays at the scale where they will be used?
   • Boundary errors can matter differently depending on the application
   • A thumbnail always shows every boundary accurately
   • Overlay scale changes the ground truth
   • Pixel accuracy proves usability
8. What should be preserved when transforming an image for segmentation?
   • An average color value
   • Only the compressed thumbnail
   • The user interface theme
   • Source dimensions and coordinate alignment