Implement Custom Vision Models AI-102 Practice Questions (Page 3)

Questions 11–15

1. 11foundation · easy

   A manufacturing company needs a custom vision model to count the number of defective components on a conveyor belt in each image. The model must identify each defective component and its location. Which type of custom vision model should be used?

   • AImage captioning
   • BImage tagging
   • CObject detection
   • DImage classification
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   A custom vision model is evaluated on a test set. The model correctly identifies 90 out of 100 actual defective items, but it also incorrectly identifies 10 non-defective items as defective. What is the recall of the model?

   • A0.90
   • B0.82
   • C0.10
   • D1.00
   Reveal Solution Discussion Select an answer first
4. 13expert · hard

   A team has trained a custom vision model and wants to deploy it to multiple Azure regions for low-latency access. They plan to publish the model. What should they consider?

   • AExport the model and deploy it to Azure Functions in each region
   • BPublish the model once and it is automatically available globally
   • CPublish the model in each region where the Custom Vision resource is provisioned
   • DUse Azure Front Door to route to a single endpoint
   Reveal Solution Discussion Select an answer first
5. 14foundation · easy

   What is the purpose of the prediction key when consuming a custom vision model?

   • ATo identify the training dataset used for the model
   • BTo authenticate requests to the prediction endpoint
   • CTo encrypt the images sent to the model
   • DTo specify the model iteration to use
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   A team trained an image classification model to distinguish between healthy and diseased plant leaves. In the evaluation results, the model shows high accuracy (95%) but low recall (0.55) for the 'diseased' class. The business requirement is to flag as many diseased leaves as possible for manual inspection, even if it means more false positives. Which metric should the team prioritize, and what action would best align with the business need?

   • APrioritize recall; retrain with more images of diseased leaves and consider lowering the classification threshold.
   • BPrioritize accuracy; the model is already 95% accurate, so no changes are needed.
   • CPrioritize precision; switch to object detection to get bounding boxes for each diseased leaf.
   • DPrioritize precision; add more images of healthy leaves to reduce false positives.
   Reveal Solution Discussion Select an answer first