AI-900Domain 3: Describe Features of Computer Vision Workloads on AzureComputer VisionhardSingle choice
A self-driving system must label every pixel in a road scene (road, pedestrian, vehicle, sign) for precise scene understanding. Which computer vision technique is this?
A.Image classification
Whole-image label — far too coarse.
B.Object detection
Bounding boxes locate objects but don't label every pixel.
C.Semantic segmentation
Per-pixel labeling = semantic segmentation. Correct.
D.OCR
OCR extracts text only.
Why this answer
Semantic segmentation assigns a class label to every pixel — the finest-grained vision task. Classification = one label per image; object detection = bounding boxes; segmentation = pixel-level masks.
Correct answer: C · Last reviewed 2026-02-14
Official AWS references
Practice this interactively
Answer without seeing the solution first in the Computer Vision practice session.
Practice Computer Vision