AI-900 Practice Question: Describe features of computer vision workloads on Azure
A quality inspection system uses cameras to examine metal parts for surface defects. The system must identify the exact location and shape of each scratch, dent, or crack. Which Azure Computer Vision capability is best suited for this?
⚠ Common exam trap
Many candidates confuse Object Detection (bounding boxes) with Semantic Segmentation (pixel-level masks), failing to recognize that only segmentation can capture the exact shape of irregular defects like cracks or dents.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Semantic Segmentation
Semantic segmentation is the correct choice because it classifies every pixel in an image, allowing the system to precisely delineate the exact location, shape, and boundaries of surface defects like scratches, dents, or cracks on metal parts. This pixel-level granularity is essential for quality inspection where the geometry of each defect must be measured and analyzed.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Image Classification
Why it's wrong here
Image classification assigns a single class label to the entire image; e.g., a typical CNN such as ResNet outputs one probability distribution over predefined categories. When a metal surface contains a crack, a scratch, and a stain, the model can only say 'defective' or 'crack present' without locating the defects or describing their spatial variation. This image-level prediction is fundamentally insufficient for quality inspection because the goal is to output precise regions of concern, not just a verdict.
- ✗
Object Detection
Why it's wrong here
Object detection locates objects by drawing axis-aligned bounding boxes (from models like YOLO or Faster R-CNN) and assigning a label to each box. Cracks and scratches are thin, non-rectangular, and may curve or branch; a bounding box would grossly overestimate the defective area and fail to capture the exact contour. Even though detection provides some localization, the rectangular representation loses the pixel-level geometry needed to assess whether a defect crosses a critical structural boundary.
- ✓
Semantic Segmentation
Why this is correct
Semantic segmentation classifies every pixel of the image into a predefined class, producing a dense label map of the same resolution as the input. With a U-Net or DeepLab model, each pixel belonging to a crack, inclusion, or non-defective metal is predicted precisely, so the model can directly output the exact shape, orientation, and boundary of every defect. This pixel-level output is what enables accurate area computation and downstream decision-making in an automated inspection line.
- ✗
Dense Captioning
Why it's wrong here
Dense captioning runs object detection-like region proposals and then generates a natural-language sentence describing each proposed region, such as 'a long scratch on the shiny surface.' It cannot produce a per-pixel mask, because its output is a set of text strings for rectangular or adaptive regions, not a full-resolution segmentation map. For quality inspection, you need quantitative geometric data (e.g., defect length and area), not a linguistic description, so dense captioning is not the right computer vision technique here.
Go deeper
Related to this question
About these practice questions
This AI-900 question is part of Courseiva's 985-question bank — original exam-style content with full explanations and wrong-answer analysis, never real exam questions or exam dumps. Learn why practice questions differ from exam dumps →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This AI-900 practice question is part of Courseiva's free Microsoft certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AI-900 exam.