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.

• ✗

  The physical location where training images were captured

  Why it's wrong here

  The physical location where training images were captured is not ground truth; it is metadata about data provenance and context. While location can matter for domain shift (e.g., a model trained on indoor photos may not generalize outdoors), it is not the target the model is trained to predict. Ground truth refers to the semantic annotation describing what the image actually contains, such as the object class, not where the image was taken.

• ✓

  The verified, accurate labels or annotations for training images that the model learns to predict

  Why this is correct

  Ground truth is the verified, accurate set of labels or annotations assigned to each training example, serving as the correct target output for the model to learn. In supervised computer vision, the model's weights are adjusted to minimize the difference between its predictions and these ground-truth labels, such as class names, bounding boxes, or segmentation masks. Without reliable ground truth, training cannot be properly supervised, because the model has no authoritative answer to imitate.

• ✗

  The minimum image resolution required for accurate model training

  Why it's wrong here

  Minimum image resolution is an input preprocessing requirement, not ground truth. Resolution affects whether a model can discern fine features and is often standardized (e.g., resizing to 224×224 pixels for a convolutional network), but it does not define the content to be predicted. Confusing input specifications with the target label misrepresents what ground truth is: the correct answer associated with an image, not a property of the image's pixel grid or capture settings.

• ✗

  The baseline accuracy of a computer vision model before fine-tuning

  Why it's wrong here

  Baseline accuracy before fine-tuning is an evaluation metric used for comparison, not ground truth. A baseline might come from a pre-trained model or a simple heuristic, and it is measured by comparing predictions against ground-truth labels. Ground truth is the external reference standard itself, while accuracy is a derived score that indicates how well predictions match that standard. The two are fundamentally different: baseline accuracy reflects current performance, whereas ground truth provides the correct answers that define what performance means.

Go deeper

Related to this question