MLA-C01 Practice Question 54: Custom Image Labels AI Service
AWS Certified Machine Learning Engineer – Associate (MLA-C01) · Test 6 · question 4 of 10
A manufacturer wants to detect three kinds of surface defects (scratches, dents, and discoloration) in photos of its metal panels on the production line. Its quality team has labeled about 250 photos per defect type. The company has no data scientists and wants a managed service that trains a model from these images and serves predictions through an API, without choosing algorithms or managing training infrastructure.
Which approach should the company use?
Show answer and explanation
Answer:
Rekognition Custom Labels builds a custom image analysis model for labels that Rekognition's pre-trained model does not know, such as defects specific to one company's parts. It builds on Rekognition's existing training, so a small set of images, typically a few hundred or fewer, is enough. It requires no machine learning expertise and serves the trained model through an API, which matches a team with labeled photos and no data scientists (option B).
- A is wrong because DetectLabels returns general-purpose labels such as "Metal" or "Machine". It has no concept of this manufacturer's defect types, so it cannot tell a dent from a normal panel. - C is wrong because the built-in algorithm requires the ML skills the company lacks: preparing data channels, choosing hyperparameters, and deploying and managing an endpoint. - D is wrong because it replaces the model with ongoing human labor. That does not keep up with a production line, adds per-task cost indefinitely, and is not the trained model the company asked for.
Reference: What is Amazon Rekognition Custom Labels?