Microsoft Azure AI Fundamentals (AI-900)Describe features of computer vision workloads on AzureHard
A smart city initiative aims to analyze urban traffic flow by understanding the density and movement of vehicles. The system needs to precisely delineate the boundaries of each vehicle (cars, buses, trucks) in real-time video streams, without necessarily distinguishing between individual instances of the same type (e.g., it doesn't need to know 'this is car A' vs 'this is car B', just 'this is a car'). Which computer vision capability is most suitable for this pixel-level understanding of vehicle presence?
- AInstance Segmentation
- BSemantic Segmentation
- CImage Classification
- DObject Detection
Show answer & explanationAnswer & explanation
Correct answer: B. Semantic Segmentation
Semantic Segmentation is the most suitable capability. It classifies every pixel in an image into a predefined class (e.g., 'car', 'road', 'sky'), providing precise boundaries for all vehicles without needing to differentiate between individual instances of the same class.
Why the other options are wrong
- A. Instance Segmentation would differentiate between individual cars (e.g., 'car_1', 'car_2'), which is more detailed than required and computationally more intensive for just density and general flow.
- C. Image Classification would label the entire scene (e.g., 'traffic jam'), not individual vehicles.
- D. Object Detection would draw bounding boxes, which are less precise than pixel-level delineation for density analysis.
Semantic Segmentation
A computer vision task that assigns a class label to every pixel in an image, effectively outlining objects and regions.
- Provides pixel-level understanding of a scene.
- Treats multiple objects of the same class as a single entity.
- Used in autonomous driving for scene understanding (road, sidewalk, car).
Memory trick: Segment pixels semantically to understand the scene.