AI-900 Practice Question: Describe features of computer vision workloads on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Dividing an image file into smaller files for distributed storage

  Why it's wrong here

  Splitting an image file into smaller chunks for distributed storage is a file-system or data-engineering concern, often done for parallelism or to fit block-size constraints in a distributed database. This operation does not analyze or interpret the image's semantic content—it merely partitions the raw bytes of the file. Image segmentation, in contrast, operates on the pixel grid to derive an understanding of scene structure, so the two are fundamentally different activities that share only the word 'segmentation'.

• ✓

  Classifying every pixel in an image to identify precise boundaries — more detailed than bounding-box object detection

  Why this is correct

  Image segmentation is the task of assigning a class label to each pixel in an image so that every element—objects, background, and fine structural details—is precisely delineated. This is more granular than object detection, which outputs a coarse bounding box that often includes surrounding background pixels, because a segmentation mask identifies the exact contour and shape of each object. The result is a dense per-pixel prediction map that can be used for tasks like medical tumor delineation, autonomous-driving scene understanding, or synthetic-image compositing.

• ✗

  Removing the background from an image by detecting edges

  Why it's wrong here

  Background removal is one downstream application of image segmentation, but the two are not synonymous. Segmentation classifies every pixel in the entire image into its corresponding object or region class, producing a complete label map; edge detection, by contrast, finds only boundary contours and does not assign semantic meaning to the regions they enclose. Removing a background typically requires both a segmentation mask and a masking operation, so the description conflates a part of the pipeline (edge detection) with the full pixel-classification capability that defines segmentation.

• ✗

  Dividing the training dataset into segments for cross-validation

  Why it's wrong here

  Cross-validation is a model-evaluation technique that partitions a labeled training dataset into complementary folds, training on some folds and validating on others to tune hyperparameters or estimate generalization. It has no relation to computer vision, because it operates on row-level data splits rather than on the spatial content of any single image. Segmenting an image, in contrast, is a per-pixel inference task applied to the image's visual features, not to the dataset as a whole.

Go deeper

Related to this question