A manufacturer wants to detect five specific defect types unique to its own product line on an assembly line, using photos its quality team has already labeled. A prebuilt Azure AI Vision image analysis model cannot recognize these defects because they are not part of its general object catalog. What should the team do using Azure AI Foundry's vision capabilities?
Select an answer to reveal the explanation.
Short Explanation
A general-purpose vision model has learned to recognize common everyday objects, not the specific scratches or dents that matter to one particular factory floor. No matter how you squint at its output or filter it, it cannot report on categories it was never shown during training, because that knowledge just is not in there. Switching to a tool built for recognizing human faces obviously will not help either, since defects on a product have nothing to do with faces. Turning up the certainty requirement on the generic labels only makes the model pickier about the generic things it already knows, it does not conjure up brand-new defect categories out of thin air. The real fix is to show the system your own labeled examples of exactly the defects you care about and let it learn those specific patterns, which is what a custom-trained model is for.
Full Explanation
The correct answer is A. Training a custom vision model on the team's own labeled images lets the system learn the specific visual patterns of their five defect types, something no general-purpose prebuilt catalog can offer since it was never trained on those categories. Option B is incorrect because no amount of post-processing can extract defect labels that were never part of the generic model's vocabulary in the first place; the underlying model simply does not know those categories exist. Option C is incorrect because the Face API is built to detect and analyze human faces, which has no relevance to identifying manufacturing defects on a product. Option D is incorrect because raising a confidence threshold only filters how certain the model is about its existing generic tags; it cannot invent new categories that were never in its training data, no matter how the threshold is tuned. Custom training on labeled, domain-specific images is the only path that produces a model capable of recognizing the manufacturer's particular defects.